nmWTAI-Platform/Src/nmNum/nmCalculation/nmCalculationAutoFitPSO.cpp

#include "nmCalculationAutoFitPSO.h"
#include "nmCalculationDllPebiSolverTask.h"
#include "nmDataAnalyzeManager.h"
#include "nmDataWellBase.h"
#include "nmDataVerticalWell.h"
#include "nmDataVerticalFracturedWell.h"
#include "nmDataHorizontalFracturedWell.h"
#include "nmDataReservoir.h"
#include "nmDataAutomaticFitting.h"

#include "nmCalculationPebiGrid.h"

#include <QApplication>
#include <QDebug>
#include <QTime>
#include <QDir>
#include <QTextStream>
#include <QFileInfo>
#include <QProcess>
#include <QProcessEnvironment>
#include <QPair>
#include <QtCore/qmath.h>
#include <cmath>
#include <algorithm>
#include <limits>
#include <QTimer>

#ifdef Q_OS_WIN
#include <windows.h>
#include <float.h>
#define DEBUG_OUT(msg) OutputDebugStringA(QString("[AutoFit] %1\n").arg(msg).toLocal8Bit().data())
#endif

// ===== PSO基础常量 =====
// MIN_FITNESS_IMPROVEMENT：判断适应度是否有有效改善的最小阈值。
// VELOCITY_LIMIT_FACTOR：单次速度最大值占参数搜索区间的比例，防止粒子一步跨太远。
// CONVERGENCE_CHECK_INTERVAL：收敛检查的基础间隔，部分历史逻辑和日志会使用它。
const double nmCalculationAutoFitPSO::MIN_FITNESS_IMPROVEMENT = 1e-8;
const double nmCalculationAutoFitPSO::VELOCITY_LIMIT_FACTOR = 0.2;
const int nmCalculationAutoFitPSO::CONVERGENCE_CHECK_INTERVAL = 5;

// ===== 代理模型筛选参数 =====
//
// PSO acceleration 的原则是“少跑一部分真实求解器，但不能完全信任代理模型”：
// - keep_fraction：代理评分排名靠前的粒子进入真实求解器；
// - audit_fraction：从被筛掉的粒子中随机抽查一部分，避免代理模型系统性漏掉好解；
// - min_solver_fraction：每代至少有一定比例粒子跑真实求解器，保证 pbest/gbest 仍由真实误差支撑；
// - warmup_iterations：前几代全量真实评价，让每个粒子先拿到可靠的个体最优；
// - full_solver_interval：周期性全量审计，用于纠偏代理模型排序误差。
static const double kSurrogateKeepFraction = 0.50;              // 常规工况：代理排序前 50% 粒子进入真实求解器。
static const double kSurrogateHighRiskKeepFraction = 0.70;      // 高风险流量制度：保留更多 top 粒子。
static const double kSurrogateAuditFraction = 0.05;             // 从被筛掉候选中随机审计 5%。
static const double kSurrogateMinSolverFraction = 0.55;         // 常规工况：每代至少 55% 粒子跑真实求解器。
static const double kSurrogateHighRiskMinSolverFraction = 0.80; // 高风险流量制度：每代至少 80% 粒子跑真实求解器。
static const int kSurrogateWarmupIterations = 2;                // 前 2 代全量真实评价，给每个粒子建立 pbest。
static const int kSurrogateFullSolverInterval = 10;             // 每 10 代做一次全量真实求解审计。
static const int kSurrogateScoringMaxAttempts = 2;              // Python 评分失败时最多重试 2 次。
static const bool kUseFixedPsoSeed = true;                      // 固定随机种子便于复现 trace 和交接演示。
static const unsigned int kFixedPsoSeed = 1792008679u;          // 当前固定种子值。

// ===== 代理模型训练域约束 =====
//
// 这些范围必须与 ML/nmWTAI-ML/configs/data_gen_family_random_v2_q.yaml 中的
// 数据生成范围保持一致。超出范围时，C++ 侧会强制退回真实求解器，避免代理模型
// 在训练域外给出不可靠排序。
static const double kSurrogateKMin = 1.0e-3;               // 代理模型训练域：渗透率下界。
static const double kSurrogateKMax = 10.0;                 // 代理模型训练域：渗透率上界。
static const double kSurrogateSkinMin = -10.0;             // 代理模型训练域：skin 下界。
static const double kSurrogateSkinMax = 10.0;              // 代理模型训练域：skin 上界。
static const double kSurrogateWellboreCMin = 1.0e-4;       // 代理模型训练域：井筒储集下界。
static const double kSurrogateWellboreCMax = 2.0;          // 代理模型训练域：井筒储集上界。
static const double kSurrogatePhiMin = 1.0e-2;             // 代理模型训练域：孔隙度下界。
static const double kSurrogatePhiMax = 0.50;               // 代理模型训练域：孔隙度上界。
static const double kSurrogateHMin = 2.0;                  // 代理模型训练域：储层厚度下界。
static const double kSurrogateHMax = 50.0;                 // 代理模型训练域：储层厚度上界。
static const double kSurrogateCfFixed = 4.315e-4;          // 当前代理模型按近似固定 Cf 训练。
static const double kSurrogateCfRelTol = 0.25;             // Cf 允许相对偏差，超过则不使用代理筛选。
static const int kSurrogateMinProdSections = 3;            // 代理训练流量制度：生产段数量下界。
static const int kSurrogateMaxProdSections = 5;            // 代理训练流量制度：生产段数量上界。
static const double kSurrogateProdTimeMin = 24.0;          // 代理训练流量制度：生产总时长下界。
static const double kSurrogateProdTimeMax = 220.0;         // 代理训练流量制度：生产总时长上界。
static const double kSurrogateShutinTimeMin = 12.0;        // 代理训练流量制度：关井段时长下界。
static const double kSurrogateShutinTimeMax = 140.0;       // 代理训练流量制度：关井段时长上界。
static const double kSurrogateQMin = 50.0;                 // 代理训练流量制度：产量下界。
static const double kSurrogateQMax = 500.0;                // 代理训练流量制度：产量上界。
static const double kSurrogateQZeroThreshold = 1.0e-6;     // 判断最后一段是否关井的零产量阈值。
static const double kSurrogateMaxRateStepRatio = 2.2;      // 相邻生产段产量最大跳变比。
static const double kSurrogateStrongDeclineEndStartRatio = 0.70; // 末段/首段产量小于该值视为强递减。
static const double kSurrogateRateTrendRelTol = 1.0e-9;    // 判断产量趋势时的相对容差。

// 判断 double 是否为有限数。PSO、误差函数、trace 写出都依赖它过滤 NaN/Inf。
static inline bool isFiniteNumber(double value)
{
#ifdef Q_OS_WIN
    return _finite(value) != 0;
#else
    return std::isfinite(value);
#endif
}

static inline bool isInClosedRange(double value, double lower, double upper)
{
    return isFiniteNumber(value) && value >= lower && value <= upper;
}

// 判断 value 是否接近 reference。用于 Cf 这类“近似固定”的训练域检查。
static inline bool isNearRelative(double value, double reference, double relTol)
{
    if(!isFiniteNumber(value) || !isFiniteNumber(reference)) {
        return false;
    }

    return qAbs(value - reference) <= qMax(1e-12, qAbs(reference) * relTol);
}

// Windows 下的毫秒级睡眠封装。用于重试间隔、UI 事件循环间隔和模拟模式节奏控制。
static inline void msleep(int ms)
{
#ifdef Q_OS_WIN
    Sleep(ms);
#endif
}

static QString csvEscape(const QString& text)
{
    // CSV 字段转义。trace 中的 run_id、phase、screeningDecision 等字符串都走这里，
    // 防止逗号或引号破坏表格结构。
    QString escaped = text;
    escaped.replace("\"", "\"\"");
    return QString("\"%1\"").arg(escaped);
}

static QString traceNumber(double value)
{
    // trace CSV 中的数值格式化。不可用数值写空字段，便于后续 pandas/Excel 读取。
    if(!isFiniteNumber(value)) {
        return QString();
    }

    return QString::number(value, 'g', 17);
}

static QString traceParamAt(const QVector<double>& params, int index)
{
    // 从完整参数向量中安全取值并格式化。越界时写空字段，避免 trace 写出崩溃。
    if(index < 0 || index >= params.size()) {
        return QString();
    }

    return traceNumber(params[index]);
}

static QString jsonEscape(const QString& text)
{
    // 手写 JSON 字符串转义。trace meta 不依赖第三方 JSON writer，
    // 所以所有字符串字段写入前必须统一经过这里。
    QString escaped = text;
    escaped.replace("\\", "\\\\");
    escaped.replace("\"", "\\\"");
    escaped.replace("\b", "\\b");
    escaped.replace("\f", "\\f");
    escaped.replace("\n", "\\n");
    escaped.replace("\r", "\\r");
    escaped.replace("\t", "\\t");
    return QString("\"%1\"").arg(escaped);
}

static QString jsonNumber(double value)
{
    // JSON 数值格式化。NaN/Inf 在 JSON 中没有合法表示，因此写 null。
    if(!isFiniteNumber(value)) {
        return "null";
    }

    return QString::number(value, 'g', 17);
}

static QString jsonDoubleArray(const QVector<double>& values)
{
    // 将 double 数组写成 JSON 数组，用于 trace meta 的目标曲线、参数上下界等。
    QStringList items;

    for(int i = 0; i < values.size(); ++i) {
        items << jsonNumber(values[i]);
    }

    return QString("[%1]").arg(items.join(","));
}

static QString jsonIntArray(const QVector<int>& values)
{
    // 将 int 数组写成 JSON 数组，用于 enabled_param_indices 等字段。
    QStringList items;

    for(int i = 0; i < values.size(); ++i) {
        items << QString::number(values[i]);
    }

    return QString("[%1]").arg(items.join(","));
}

static QString jsonBoolArray(const QVector<bool>& values)
{
    // 将 bool 数组写成 JSON 数组，用于参数是否启用等字段。
    QStringList items;

    for(int i = 0; i < values.size(); ++i) {
        items << (values[i] ? "true" : "false");
    }

    return QString("[%1]").arg(items.join(","));
}

static double deterministicAuditRandom01(unsigned int seed, int generation, int particleIndex)
{
    // 代理筛选的随机审计需要“随机但可复现”：
    // 同一个 PSO 种子、代号、粒子号会得到同一个 0-1 值，方便复盘 trace。
    unsigned int x = seed;
    x ^= 0x9e3779b9u + static_cast<unsigned int>(generation + 1) + (x << 6) + (x >> 2);
    x ^= 0x85ebca6bu + static_cast<unsigned int>(particleIndex + 1) + (x << 6) + (x >> 2);
    x ^= x >> 16;
    x *= 0x7feb352du;
    x ^= x >> 15;
    x *= 0x846ca68bu;
    x ^= x >> 16;
    return static_cast<double>(x) / 4294967295.0;
}

static QString jsonStringArray(const QStringList& values)
{
    // 将字符串列表写成 JSON 数组，用于参数名、启用参数名等字段。
    QStringList items;

    for(int i = 0; i < values.size(); ++i) {
        items << jsonEscape(values[i]);
    }

    return QString("[%1]").arg(items.join(","));
}

static QString jsonPointArray(const QVector<QPointF>& points)
{
    // 将 QPointF 数组写成 [{"x":...,"y":...}]，用于 trace meta 保存流量制度。
    QStringList items;

    for(int i = 0; i < points.size(); ++i) {
        items << QString("{\"x\":%1,\"y\":%2}")
              .arg(jsonNumber(points[i].x()))
              .arg(jsonNumber(points[i].y()));
    }

    return QString("[%1]").arg(items.join(","));
}

static bool isValidMlRootDirectory(const QString& mlRootPath)
{
    // 判断一个目录是不是可用的 ML/nmWTAI-ML 根目录。
    // 目前以评分脚本是否存在作为最小可用条件。
    if(mlRootPath.isEmpty()) {
        return false;
    }

    QFileInfo scriptInfo(QDir(mlRootPath).absoluteFilePath("scripts/score_pso_candidates.py"));
    return scriptInfo.exists() && scriptInfo.isFile();
}

static QString findExecutableInPath(const QString& executableName)
{
    // 在系统 PATH 中查找可执行文件。找不到 .venv 里的 python.exe 时会用它兜底。
    if(executableName.isEmpty()) {
        return QString();
    }

    QString pathValue = QProcessEnvironment::systemEnvironment().value("PATH");

    if(pathValue.isEmpty()) {
        pathValue = QString::fromLocal8Bit(qgetenv("PATH"));
    }

    if(pathValue.isEmpty()) {
        return QString();
    }

    QStringList executableCandidates;
    executableCandidates << executableName;
#ifdef Q_OS_WIN

    if(!executableName.contains('.')) {
        executableCandidates << (executableName + ".exe")
                             << (executableName + ".bat")
                             << (executableName + ".cmd");
    }

#endif

#ifdef Q_OS_WIN
    QChar pathSeparator(';');
#else
    QChar pathSeparator(':');
#endif
    QStringList pathEntries = pathValue.split(pathSeparator, QString::SkipEmptyParts);

    for(int i = 0; i < pathEntries.size(); ++i) {
        QDir dir(pathEntries[i].trimmed());

        if(!dir.exists()) {
            continue;
        }

        for(int j = 0; j < executableCandidates.size(); ++j) {
            QFileInfo executableInfo(dir.absoluteFilePath(executableCandidates[j]));

            if(executableInfo.exists() && executableInfo.isFile()) {
                return QDir::cleanPath(executableInfo.absoluteFilePath());
            }
        }
    }

    return QString();
}

static QStringList traceParameterNames()
{
    // trace 和 trace meta 使用的完整参数名顺序。
    // 这个顺序必须与 buildTraceParameterVector() 和 m_parameterSelected 的 0-10 索引一致。
    QStringList names;
    names << "k"
          << "skin"
          << "wellboreC"
          << "phi"
          << "initialPressure"
          << "h"
          << "Ct"
          << "Cf"
          << "Soi"
          << "Swi"
          << "Sgi";
    return names;
}

// 构造函数：初始化 PSO 参数、收敛判断阈值、trace/代理统计和模拟模式状态。
// 这里设置的是默认值，真正运行前会由 loadOptimizationConfig() 和
// loadParameterBounds() 从 DataManager 读取界面配置覆盖一部分字段。
nmCalculationAutoFitPSO::nmCalculationAutoFitPSO(QObject* parent)
    : QObject(parent)
    , m_isRunning(false)
    , m_shouldStop(false)
    , m_isPaused(false)
    , m_currentIteration(0)
    , m_globalBestFitness(1e10)
    , m_previousBestFitness(1e10)
    , m_swarmSize(20)
    , m_maxIterations(100)
    , m_targetError(0.001)
    , m_inertiaWeight(0.4)
    , m_cognitiveParam(2.0)
    , m_socialParam(1.2)
    , m_totalEvaluations(0)
    , m_successfulEvaluations(0)
    , m_evaluationInProgress(0)
    , m_consecutiveFailures(0)
    , m_improvementThreshold(0.05)
    , m_userInitialFitness(1e10)
    , m_consecutiveFailedIterations(0)
    , m_maxConsecutiveFailures(3)
    , m_hasValidUserSolution(false)
    , m_diversityThreshold(0.05)
    , m_convergenceVarianceThreshold(1e-8)
    , m_velocityConvergenceThreshold(0.01)
    , m_trueConvergenceWindow(15)
    , m_localOptimumWindow(8)
    , m_nearTargetFactor(2.0)
    , m_farTargetFactor(10.0)
    , m_targetWellName("")
    , m_traceEnabled(true)
    , m_traceRunId("")
    , m_traceFilePath("")
    , m_traceMetaFilePath("")
    , m_surrogateScreeningEnabled(false)
    , m_psoRandomSeed(0)
    , m_surrogateRunContextSummary("")
    , m_surrogateWarmupIterationCount(0)
    , m_surrogatePeriodicAuditIterationCount(0)
    , m_surrogateContextBlockedIterationCount(0)
    , m_surrogateActiveIterationCount(0)
    , m_surrogateSelectedParticleCount(0)
    , m_surrogateScreenedParticleCount(0)
    , m_surrogateTopKParticleCount(0)
    , m_surrogateAuditParticleCount(0)
    , m_surrogateFallbackParticleCount(0)
    , m_surrogateDomainParticleCount(0)
    , m_surrogateMinFloorParticleCount(0)
    , m_surrogateScoringProcess(nullptr)
    , m_surrogateScoringServerKey("")

    , m_simulationMode(false)
    , m_simulationTimer(nullptr)
    , m_simulationIteration(0)
    , m_simulationMaxIterations(100)
    , m_simulationTargetError(0.001)
    , m_simulationStartError(0.5)
    , m_simulationCurrentError(0.5)
    , m_simulationLogInterval(2)
    , m_simulationCurrentParticle(0)
    , m_simulationSuccessCount(0)
    , m_simulationFailCount(0)
    , m_simulationIterationStarted(false)
    , m_simulationPreviousIterError(0.5)
{
    DEBUG_OUT(QString("PSO Constructor: this=0x%1").arg((quintptr)this, 0, 16));

    initializeTemporaryDirectory();

    DEBUG_OUT("AutoFit calculator initialized (data-driven mode)");
    DEBUG_OUT("PSO Constructor completed");
}

// 析构函数：停止仍在进行的拟合、关闭代理评分服务、关闭 trace 文件并清理临时目录。
// 自动拟合可能在 UI 线程中被窗口关闭打断，因此析构时要尽量温和地等待当前评价结束；
// 如果等待超时，再强制清除运行标志，避免对象销毁后还有信号回调或后台进程访问成员变量。
nmCalculationAutoFitPSO::~nmCalculationAutoFitPSO()
{
    DEBUG_OUT(QString("PSO Destructor: this=0x%1").arg((quintptr)this, 0, 16));

    if(m_simulationTimer) {
        m_simulationTimer->stop();
        delete m_simulationTimer;
        m_simulationTimer = nullptr;
    }

    // Stop algorithm.
    if(m_isRunning) {
        m_shouldStop = true;

        int waitCount = 0;

        while(m_isRunning && waitCount < 100) {
            QApplication::processEvents(QEventLoop::ExcludeUserInputEvents, 50);
            msleep(50);
            waitCount++;
        }

        if(m_isRunning) {
            DEBUG_OUT("Force stopping PSO - timeout reached");
            m_isRunning = false;
        }
    }

    // Clean temp directory.
    stopSurrogateScoringServer();
    closeTraceFile();
    cleanupTemporaryDirectory();

    // Disconnect all signals to avoid callbacks after destruction.
    disconnect(this, nullptr, nullptr, nullptr);

    DEBUG_OUT("PSO Destructor completed");
}

// ===== 临时目录工具 =====
//
// 真实求解器 DLL 和自动拟合过程会产生中间文件，因此每次创建独立的
// autofit_temp_<pid>_<timestamp> 目录。退出时只删除本类创建的目录，启动时顺便清理
// 旧进程遗留的 autofit_temp_*，避免长期调试后应用目录被临时文件堆满。

void nmCalculationAutoFitPSO::initializeTemporaryDirectory()
{
    // 先清理历史遗留目录，再为本次对象创建唯一目录。
    // 目录名包含进程 ID 和毫秒时间戳，通常足够唯一；counter 是极端重名时的兜底。
    cleanupOldTemporaryDirectories();

    QString timestamp = QDateTime::currentDateTime().toString("yyyyMMdd_hhmmss_zzz");
    QString processId = QString::number(QCoreApplication::applicationPid());

    m_tempDirectory = QApplication::applicationDirPath() +
                      "/autofit_temp_" + processId + "_" + timestamp;

    int counter = 0;
    QString originalPath = m_tempDirectory;

    while(QDir(m_tempDirectory).exists() && counter < 100) {
        m_tempDirectory = originalPath + "_" + QString::number(counter);
        counter++;
    }

    if(QDir().mkpath(m_tempDirectory)) {
        DEBUG_OUT(QString("Initialized temp directory: %1").arg(m_tempDirectory));
    } else {
        DEBUG_OUT(QString("Warning: Failed to create temp directory: %1").arg(m_tempDirectory));
        m_tempDirectory = QApplication::applicationDirPath();
    }
}

void nmCalculationAutoFitPSO::cleanupTemporaryDirectory()
{
    // 析构或用户停止时调用。删除失败一般是文件仍被 DLL/系统占用，
    // 这里只记录 debug 信息，不让清理失败影响 UI 退出。
    if(QDir(m_tempDirectory).exists()) {
        if(removeDirectoryRecursively(m_tempDirectory)) {
            DEBUG_OUT("Temp directory cleaned up successfully");
        } else {
            DEBUG_OUT("Warning: Failed to clean up temp directory completely");
        }
    }
}

bool nmCalculationAutoFitPSO::removeDirectoryRecursively(const QString& path)
{
    // Qt 旧版本没有统一可用的 removeRecursively 行为时，用本函数递归删除。
    // 调用方传入的是本类创建的临时目录或旧 autofit_temp_* 目录。
    QDir dir(path);

    if(!dir.exists()) {
        return true;
    }

    // 递归删除子目录和文件。包含 Hidden，避免隐藏中间文件阻塞目录删除。
    QFileInfoList entries = dir.entryInfoList(QDir::NoDotAndDotDot | QDir::AllEntries | QDir::Hidden);
    bool allRemoved = true;

    for(int i = 0; i < entries.size(); ++i) {
        const QFileInfo& entry = entries[i];

        if(entry.isDir()) {
            if(!removeDirectoryRecursively(entry.absoluteFilePath())) {
                allRemoved = false;
            }
        } else {
            QFile file(entry.absoluteFilePath());

            // 处理只读文件。某些求解器输出可能带只读属性，删除前先补写权限。
            if(!file.permissions().testFlag(QFile::WriteUser)) {
                file.setPermissions(file.permissions() | QFile::WriteUser);
            }

            if(!file.remove()) {
                DEBUG_OUT(QString("Failed to remove file: %1").arg(entry.absoluteFilePath()));
                allRemoved = false;
            }
        }
    }

    // 删除目录本身
    if(allRemoved) {
        return dir.rmdir(path);
    }

    return false;
}

void nmCalculationAutoFitPSO::cleanupOldTemporaryDirectories()
{
    // 应用启动或新建自动拟合对象时清理旧目录。
    // 不删除当前进程 ID 对应目录，防止同进程内多个拟合对象或正在运行的求解器被误删。
    QString appDir = QApplication::applicationDirPath();
    QDir dir(appDir);

    // 获取当前进程ID，避免删除当前进程可能使用的目录
    QString currentProcessId = QString::number(QCoreApplication::applicationPid());

    // 查找所有以 "autofit_temp_" 开头的目录
    QStringList filters;
    filters << "autofit_temp_*";
    QFileInfoList tempDirs = dir.entryInfoList(filters, QDir::Dirs | QDir::NoDotAndDotDot);

    if(tempDirs.isEmpty()) {
        DEBUG_OUT("No old temporary directories found");
        return;
    }

    DEBUG_OUT(QString("Found %1 potential old temporary directories to clean up").arg(tempDirs.size()));

    int cleanedCount = 0;
    int failedCount = 0;
    int skippedCount = 0;

    for(int i = 0; i < tempDirs.size(); ++i) {
        const QFileInfo& dirInfo = tempDirs[i];
        QString dirPath = dirInfo.absoluteFilePath();
        QString dirName = dirInfo.fileName();

        // 检查是否是当前进程的目录（虽然理论上不应该存在，但为了安全起见）
        if(dirName.contains("_" + currentProcessId + "_")) {
            DEBUG_OUT(QString("Skipping current process directory: %1").arg(dirName));
            skippedCount++;
            continue;
        }

        // 尝试删除目录
        DEBUG_OUT(QString("Attempting to remove old temp directory: %1").arg(dirName));

        if(removeDirectoryRecursively(dirPath)) {
            DEBUG_OUT(QString("Successfully cleaned up: %1").arg(dirName));
            cleanedCount++;
        } else {
            DEBUG_OUT(QString("Failed to clean up: %1 (may be in use by another process)").arg(dirName));
            failedCount++;
        }
    }

    // 输出清理统计信息
    DEBUG_OUT(QString("Old temp directories cleanup summary: %1 removed, %2 failed, %3 skipped")
              .arg(cleanedCount).arg(failedCount).arg(skippedCount));
}


// ==================== 公共接口方法 ====================

void nmCalculationAutoFitPSO::setTargetLogLogData(const QVector<QVector<double> >& targetData)
{
    // 目标曲线由界面层从目标井 history log-log 传入。
    // 约定 targetData[0]=time，targetData[1]=pressure，targetData[2]=pressure derivative。
    m_targetLogLogData = targetData;
    DEBUG_OUT(QString("Target LogLog data set: %1 arrays").arg(targetData.size()));

    if(targetData.size() >= 3) {
        DEBUG_OUT(QString("LogLog data points: X=%1, Y1=%2, Y2=%3")
                  .arg(targetData[0].size())
                  .arg(targetData[1].size())
                  .arg(targetData[2].size()));
    }
}


void nmCalculationAutoFitPSO::stopFitting()
{
    // 用户点击停止时走这里。停止策略是“请求式停止”：
    // 先置 m_shouldStop，让主循环/求解器等待逻辑自然退出；短时间内还在评价时再重置计数。
    // 这样可以减少 DLL 任务被硬中断导致的数据状态残留。
    if(m_simulationMode && m_simulationTimer) {
        m_simulationTimer->stop();
    }

    if(m_isRunning) {
        DEBUG_OUT("Stop request received, setting stop flag...");

        // 添加停止日志
        emit logMessageGenerated(tr("=== User Stop Request Received ==="));
        emit logMessageGenerated(tr("Gracefully stopping PSO optimization..."));

        m_shouldStop = true;

        // 等待当前评估完成，缩短超时时间
        int waitCount = 0;

        while(m_evaluationInProgress > 0 && waitCount < 30) {  // 减少等待时间
            QApplication::processEvents(QEventLoop::ExcludeUserInputEvents, 50);
            msleep(50);
            waitCount++;
        }

        // 超时时强制重置
        if(m_evaluationInProgress > 0) {
            DEBUG_OUT("Force resetting evaluation counter");
            emit logMessageGenerated(tr("Force stopping current evaluation..."));
            m_evaluationInProgress = 0;
        }

        // 确保运行标志被清除
        m_isRunning = false;

        emit logMessageGenerated(tr("PSO optimization stop request processed"));
        DEBUG_OUT("Stop request processed");
    } else {
        DEBUG_OUT("Stop request received but PSO is not running");
        emit logMessageGenerated(tr("Stop request received but optimization is not running"));
    }

    closeTraceFile();
    cleanupTemporaryDirectory();
}

bool nmCalculationAutoFitPSO::isRunning() const
{
    // UI 查询当前是否处于自动拟合运行状态。
    return m_isRunning;
}

int nmCalculationAutoFitPSO::getCurrentIteration() const
{
    // UI 进度条和日志展示用的当前迭代序号。
    return m_currentIteration;
}

QVector<double> nmCalculationAutoFitPSO::getBestSolution() const
{
    // 返回紧凑的“启用参数向量”，顺序与 m_enabledParamIndices 一致。
    return m_globalBestPosition;
}

double nmCalculationAutoFitPSO::getBestFitness() const
{
    // 当前全局最优真实误差。越小越好，1e10 附近通常表示尚无有效解。
    return m_globalBestFitness;
}

QString nmCalculationAutoFitPSO::getLastError() const
{
    // 上一次失败的人类可读错误信息，主要给 UI 层弹窗或日志使用。
    return m_lastError;
}

void nmCalculationAutoFitPSO::resetOptimizer()
{
    // 清空一次运行产生的状态，但不销毁对象。
    // 配置字段会在 startAutoFitting() 中重新从 DataManager 读取；
    // trace 文件先关闭，避免新一轮 run 继续写到旧 CSV。
    closeTraceFile();
    m_swarm.clear();
    m_globalBestPosition.clear();
    m_globalBestFitness = 1e10;
    m_previousBestFitness = 1e10;
    m_lastEvaluatedLogLogData.clear();
    m_globalBestLogLogData.clear();
    m_userInitialLogLogData.clear();
    m_currentIteration = 0;
    m_totalEvaluations = 0;
    m_successfulEvaluations = 0;
    m_convergenceHistory.clear();
    m_lastError.clear();
    m_initialValues.clear();
    m_userInitialSolution.clear();
    m_userInitialFitness = 1e10;
    m_hasValidUserSolution = false;
    m_diversityHistory.clear();
    m_velocityHistory.clear();
    m_particleStagnationHistory.clear();
    m_traceMetaFilePath.clear();
    resetRunSummary();

    DEBUG_OUT("Optimizer reset");
}

void nmCalculationAutoFitPSO::setPSOTargetWellName(const QString& wellName)
{
    // 目标井名是贯穿拟合流程的关键索引：
    // 读目标曲线、写 skin/wellboreC、求解后取 resultLogLog 都依赖这个名字。
    m_targetWellName = wellName;
}

void nmCalculationAutoFitPSO::initializeTraceFile()
{
    // 创建本次 PSO 的可复盘文件：
    // - pso_baseline_trace_<run_id>.csv：逐代逐粒子的参数、真实误差、代理误差和筛选决策；
    // - pso_baseline_trace_<run_id>.meta.json：目标曲线、流量制度、参数上下界、PSO/代理配置。
    //
    // 代理模型评分脚本也会读取 meta.json，因此 trace meta 不是单纯日志，而是C++ 与 Python 代理模型之间的运行上下文契约。
    if(!m_traceEnabled) {
        return;
    }

    closeTraceFile();
    captureSurrogateRunContextSummary();

    m_traceRunId = QDateTime::currentDateTime().toString("yyyyMMdd_hhmmss_zzz");
    QDir traceDir(QDir(getMlRootPath()).absoluteFilePath("data/temp"));

    if(!traceDir.exists() && !QDir().mkpath(traceDir.absolutePath())) {
        DEBUG_OUT(QString("Failed to create PSO trace directory: %1").arg(traceDir.absolutePath()));
        m_traceFilePath.clear();
        return;
    }

    m_traceFilePath = traceDir.absoluteFilePath(QString("pso_baseline_trace_%1.csv").arg(m_traceRunId));
    m_traceMetaFilePath = traceDir.absoluteFilePath(QString("pso_baseline_trace_%1.meta.json").arg(m_traceRunId));
    m_traceFile.setFileName(m_traceFilePath);

    if(!m_traceFile.open(QIODevice::WriteOnly | QIODevice::Text)) {
        DEBUG_OUT(QString("Failed to open PSO trace file: %1").arg(m_traceFilePath));
        m_traceFilePath.clear();
        m_traceMetaFilePath.clear();
        return;
    }

    writeTraceHeader();
    writeTraceMetaFile();
    DEBUG_OUT(QString("PSO baseline trace initialized: %1").arg(m_traceFilePath));
    emit logMessageGenerated(tr("PSO baseline trace: %1").arg(m_traceFilePath));

    if(!m_traceMetaFilePath.isEmpty()) {
        emit logMessageGenerated(tr("PSO baseline trace meta: %1").arg(m_traceMetaFilePath));
    }

    emit logMessageGenerated(tr("PSO surrogate screening: %1, model=%2, keep=%3, audit=%4, warmup=%5, min_solver=%6")
                             .arg(isSurrogateScreeningEnabled() ? tr("enabled") : tr("disabled"))
                             .arg(getSurrogateTag())
                             .arg(getSurrogateKeepFraction(), 0, 'f', 2)
                             .arg(getSurrogateAuditFraction(), 0, 'f', 2)
                             .arg(getSurrogateWarmupIterations())
                             .arg(getSurrogateMinSolverFraction(), 0, 'f', 2));

    if(!isSurrogateScreeningEnabled()) {
        emit logMessageGenerated(tr("Surrogate screening is turned off in the automatic fitting configuration"));
    } else {
        if(m_surrogateRunContextSummary == "passed") {
            emit logMessageGenerated(tr("Surrogate run context check: passed"));
        } else {
            emit logMessageGenerated(tr("Surrogate run context check: %1").arg(m_surrogateRunContextSummary));
        }
    }
}

void nmCalculationAutoFitPSO::resetRunSummary()
{
    // 重置本次运行的代理筛选统计。这里只记录统计口径，不参与 PSO 更新。
    m_surrogateRunContextSummary.clear();
    m_surrogateWarmupIterationCount = 0;
    m_surrogatePeriodicAuditIterationCount = 0;
    m_surrogateContextBlockedIterationCount = 0;
    m_surrogateActiveIterationCount = 0;
    m_surrogateSelectedParticleCount = 0;
    m_surrogateScreenedParticleCount = 0;
    m_surrogateTopKParticleCount = 0;
    m_surrogateAuditParticleCount = 0;
    m_surrogateFallbackParticleCount = 0;
    m_surrogateDomainParticleCount = 0;
    m_surrogateMinFloorParticleCount = 0;
}

void nmCalculationAutoFitPSO::captureSurrogateRunContextSummary()
{
    // 在 trace 初始化阶段记录一次“本项目是否支持代理模型”的结论。
    // 后续每代仍会再次检查，防止运行中 DataManager 状态发生变化。
    if(!isSurrogateScreeningEnabled()) {
        m_surrogateRunContextSummary = "disabled_by_config";
        return;
    }

    QString contextReason;

    if(isSurrogateRunContextSupported(&contextReason)) {
        m_surrogateRunContextSummary = "passed";
    } else {
        m_surrogateRunContextSummary = QString("failed - %1").arg(contextReason);
    }
}

void nmCalculationAutoFitPSO::emitRunSummary(bool success, StopReasonPSO finalReason)
{
    // 运行结束时输出一组交接/排障最有用的摘要：
    // 真实求解次数、成功失败数、代理模型启用情况、各类筛选决策计数，以及 trace 文件路径。
    emit logMessageGenerated(tr("=== PSO Run Summary ==="));
    emit logMessageGenerated(tr("Stop reason: %1").arg(getStopReasonDescription(finalReason)));
    emit logMessageGenerated(tr("Result: %1, final error=%2, iterations=%3, evaluations=%4 (successful=%5, failed=%6)")
                             .arg(success ? "SUCCESS" : "FAILED")
                             .arg(m_globalBestFitness, 0, 'e', 4)
                             .arg(m_currentIteration + 1)
                             .arg(m_totalEvaluations)
                             .arg(m_successfulEvaluations)
                             .arg(m_totalEvaluations - m_successfulEvaluations));
    emit logMessageGenerated(tr("Surrogate config: %1, context=%2")
                             .arg(isSurrogateScreeningEnabled() ? "enabled" : "disabled")
                             .arg(m_surrogateRunContextSummary.isEmpty() ? QString("unknown") : m_surrogateRunContextSummary));
    emit logMessageGenerated(tr("Surrogate iterations: active=%1, warmup=%2, periodic_audit=%3, context_blocked=%4")
                             .arg(m_surrogateActiveIterationCount)
                             .arg(m_surrogateWarmupIterationCount)
                             .arg(m_surrogatePeriodicAuditIterationCount)
                             .arg(m_surrogateContextBlockedIterationCount));
    emit logMessageGenerated(tr("Surrogate particles: selected=%1, screened_out=%2, topk=%3, random_audit=%4, fallback=%5, domain_gate=%6, min_solver_floor=%7")
                             .arg(m_surrogateSelectedParticleCount)
                             .arg(m_surrogateScreenedParticleCount)
                             .arg(m_surrogateTopKParticleCount)
                             .arg(m_surrogateAuditParticleCount)
                             .arg(m_surrogateFallbackParticleCount)
                             .arg(m_surrogateDomainParticleCount)
                             .arg(m_surrogateMinFloorParticleCount));

    if(!m_traceFilePath.isEmpty()) {
        emit logMessageGenerated(tr("Artifacts: trace=%1").arg(m_traceFilePath));
    }

    if(!m_traceMetaFilePath.isEmpty()) {
        emit logMessageGenerated(tr("Artifacts: trace_meta=%1").arg(m_traceMetaFilePath));
    }
}

void nmCalculationAutoFitPSO::closeTraceFile()
{
    // trace 与代理评分 server 生命周期绑定在一次 PSO run 内。
    // 关闭 trace 时同时停止 server，避免 Python 进程继续占用模型或 stdout 管道。
    stopSurrogateScoringServer();

    if(m_traceFile.isOpen()) {
        m_traceFile.flush();
        m_traceFile.close();
    }
}

void nmCalculationAutoFitPSO::writeTraceHeader()
{
    // trace CSV 字段说明：
    // - 当前粒子参数只记录代理模型关心的 k/skin/wellboreC/phi/h/Cf；
    // - solver_objective 是真实求解器误差；
    // - surrogate_objective 是 Python 代理评分；
    // - screening_decision 说明该粒子为什么跑/不跑真实求解器；
    // - pbest/gbest 字段用于离线复盘 PSO 更新是否只依赖真实误差。
    if(!m_traceFile.isOpen()) {
        return;
    }

    QStringList cols;
    cols << "run_id"
         << "generation"
         << "particle_id"
         << "phase"
         << "k"
         << "skin"
         << "wellboreC"
         << "phi"
         << "h"
         << "Cf"
         << "solver_objective"
         << "solver_success"
         << "elapsed_ms"
         << "surrogate_objective"
         << "screening_decision"
         << "pbest_objective"
         << "pbest_k"
         << "pbest_skin"
         << "pbest_wellboreC"
         << "pbest_phi"
         << "pbest_h"
         << "pbest_Cf"
         << "gbest_objective"
         << "gbest_k"
         << "gbest_skin"
         << "gbest_wellboreC"
         << "gbest_phi"
         << "gbest_h"
         << "gbest_Cf"
         << "enabled_param_indices";

    QTextStream out(&m_traceFile);
    out << cols.join(",") << "\n";
}

void nmCalculationAutoFitPSO::writeTraceMetaFile()
{
    // 写 trace 的配套 JSON。它把“如何解释 CSV”的上下文固化下来：
    // 目标井、目标 log-log 曲线、流量制度、参数名和上下界、初始参数、PSO 配置、
    // 代理模型 tag/stage/筛选比例等。Python 评分脚本会读取这个文件来构造模型输入。
    if(m_traceMetaFilePath.isEmpty()) {
        return;
    }

    QFile metaFile(m_traceMetaFilePath);

    if(!metaFile.open(QIODevice::WriteOnly | QIODevice::Text)) {
        DEBUG_OUT(QString("Failed to open PSO trace meta file: %1").arg(m_traceMetaFilePath));
        m_traceMetaFilePath.clear();
        return;
    }

    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
    nmDataWellBase* pTargetWell = dataManager ? dataManager->findWellByName(m_targetWellName) : nullptr;

    QStringList parameterNames = traceParameterNames();
    QStringList enabledNames;

    for(int i = 0; i < m_enabledParamIndices.size(); ++i) {
        int paramIndex = m_enabledParamIndices[i];
        enabledNames << ((paramIndex >= 0 && paramIndex < parameterNames.size()) ? parameterNames[paramIndex] : QString::number(paramIndex));
    }

    QVector<double> initialFullParams = buildTraceParameterVector(m_initialValues);
    QVector<double> targetTime = m_targetLogLogData.size() > 0 ? m_targetLogLogData[0] : QVector<double>();
    QVector<double> targetPressure = m_targetLogLogData.size() > 1 ? m_targetLogLogData[1] : QVector<double>();
    QVector<double> targetDerivative = m_targetLogLogData.size() > 2 ? m_targetLogLogData[2] : QVector<double>();
    QVector<QPointF> flowPoints = pTargetWell ? pTargetWell->getFlowPoints() : QVector<QPointF>();

    QTextStream out(&metaFile);
    out << "{\n";
    out << "  \"schema_version\": 1,\n";
    out << "  \"trace_type\": \"pso_baseline_replay_meta\",\n";
    out << "  \"run_id\": " << jsonEscape(m_traceRunId) << ",\n";
    out << "  \"created_at\": " << jsonEscape(QDateTime::currentDateTime().toString(Qt::ISODate)) << ",\n";
    out << "  \"trace_csv\": " << jsonEscape(QFileInfo(m_traceFilePath).fileName()) << ",\n";
    out << "  \"target\": {\n";
    out << "    \"well_name\": " << jsonEscape(m_targetWellName) << ",\n";
    out << "    \"well_type\": " << (pTargetWell ? QString::number(static_cast<int>(pTargetWell->getWellType())) : "null") << ",\n";
    out << "    \"section_policy\": \"current_target_well_indexF\",\n";
    out << "    \"section_index\": " << (pTargetWell ? QString::number(pTargetWell->getIndexF()) : "null") << ",\n";
    out << "    \"target_loglog\": {\n";
    out << "      \"time\": " << jsonDoubleArray(targetTime) << ",\n";
    out << "      \"pressure\": " << jsonDoubleArray(targetPressure) << ",\n";
    out << "      \"derivative\": " << jsonDoubleArray(targetDerivative) << "\n";
    out << "    },\n";
    out << "    \"flow_points\": " << jsonPointArray(flowPoints) << "\n";
    out << "  },\n";
    out << "  \"pso\": {\n";
    out << "    \"swarm_size\": " << m_swarmSize << ",\n";
    out << "    \"max_iterations\": " << m_maxIterations << ",\n";
    out << "    \"target_error\": " << jsonNumber(m_targetError) << ",\n";
    out << "    \"inertia_weight\": " << jsonNumber(m_inertiaWeight) << ",\n";
    out << "    \"cognitive_param\": " << jsonNumber(m_cognitiveParam) << ",\n";
    out << "    \"social_param\": " << jsonNumber(m_socialParam) << ",\n";
    out << "    \"random_seed\": " << m_psoRandomSeed << "\n";
    out << "  },\n";
    out << "  \"surrogate_screening\": {\n";
    out << "    \"enabled\": " << (isSurrogateScreeningEnabled() ? "true" : "false") << ",\n";
    out << "    \"model_tag\": " << jsonEscape(getSurrogateTag()) << ",\n";
    out << "    \"keep_fraction\": " << jsonNumber(getSurrogateKeepFraction()) << ",\n";
    out << "    \"audit_fraction\": " << jsonNumber(getSurrogateAuditFraction()) << ",\n";
    out << "    \"min_solver_fraction\": " << jsonNumber(getSurrogateMinSolverFraction()) << ",\n";
    out << "    \"warmup_iterations\": " << getSurrogateWarmupIterations() << ",\n";
    out << "    \"full_solver_interval\": " << getSurrogateFullSolverInterval() << ",\n";
    out << "    \"audit_random_source\": \"deterministic_hash\"\n";
    out << "  },\n";
    out << "  \"parameters\": {\n";
    out << "    \"names\": " << jsonStringArray(parameterNames) << ",\n";
    out << "    \"enabled_indices\": " << jsonIntArray(m_enabledParamIndices) << ",\n";
    out << "    \"enabled_names\": " << jsonStringArray(enabledNames) << ",\n";
    out << "    \"selected_flags\": " << jsonBoolArray(m_parameterSelected) << ",\n";
    out << "    \"lower\": " << jsonDoubleArray(m_parameterLower) << ",\n";
    out << "    \"upper\": " << jsonDoubleArray(m_parameterUpper) << ",\n";
    out << "    \"initial_selected\": " << jsonDoubleArray(m_initialValues) << ",\n";
    out << "    \"initial_full\": " << jsonDoubleArray(initialFullParams) << "\n";
    out << "  },\n";
    out << "  \"replay_notes\": [\n";
    out << "    \"CSV rows contain true solver objectives for evaluated particles.\",\n";
    out << "    \"This meta file contains the target curve and run context needed for surrogate replay.\",\n";
    out << "    \"pbest and gbest must be updated only from true solver objectives.\"\n";
    out << "  ]\n";
    out << "}\n";
    metaFile.flush();
    metaFile.close();
}

QVector<double> nmCalculationAutoFitPSO::buildTraceParameterVector(const QVector<double>& selectedParameters) const
{
    // 将粒子内部使用的“启用参数向量”还原成完整 11 维参数向量。
    // 未启用的参数从当前 DataManager 读取，启用的参数用 selectedParameters 覆盖。
    // trace CSV、候选 CSV、代理训练域检查都需要这个完整向量。
    QVector<double> fullParams(11, 0.0);

    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();

    if(dataManager) {
        nmDataReservoir reservoirData = dataManager->getReservoirDataCopy();
        fullParams[0] = reservoirData.getPermeability().getValue().toDouble();
        fullParams[3] = reservoirData.getPorosity().getValue().toDouble();
        fullParams[4] = reservoirData.getInitialPressure().getValue().toDouble();
        fullParams[5] = reservoirData.getThickness().getValue().toDouble();
        fullParams[6] = reservoirData.getCt().getValue().toDouble();
        fullParams[7] = reservoirData.getCf().getValue().toDouble();
        fullParams[8] = reservoirData.getSoi().getValue().toDouble();
        fullParams[9] = reservoirData.getSwi().getValue().toDouble();
        fullParams[10] = reservoirData.getSgi().getValue().toDouble();

        nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);

        if(pTargetWell) {
            fullParams[1] = pTargetWell->getPerforation(0)->getSkin().getValue().toDouble();
            fullParams[2] = pTargetWell->getWellboreStorage().getValue().toDouble();
        }
    }

    for(int i = 0; i < selectedParameters.size() && i < m_enabledParamIndices.size(); ++i) {
        int paramIndex = m_enabledParamIndices[i];

        if(paramIndex >= 0 && paramIndex < fullParams.size()) {
            fullParams[paramIndex] = selectedParameters[i];
        }
    }

    return fullParams;
}

void nmCalculationAutoFitPSO::writeTraceRow(int generation,
        int particleIndex,
        const QString& phase,
        const QVector<double>& parameters,
        double solverObjective,
        bool solverSuccess,
        int elapsedMs,
        double surrogateObjective,
        const QString& screeningDecision,
        const QVector<double>& pbestPosition,
        double pbestObjective)
{
    // 写一行 trace。generation=-1/particleIndex=-1 表示用户初始解；
    // 普通粒子行的 phase 为 particle_solver 或 particle_not_evaluated。
    // 被代理筛掉的粒子 solver_objective 写空值，但仍保留 surrogate_objective 和 decision。
    if(!m_traceFile.isOpen()) {
        return;
    }

    QVector<double> currentParams = buildTraceParameterVector(parameters);
    QVector<double> pbestParams = pbestPosition.isEmpty() ? QVector<double>() : buildTraceParameterVector(pbestPosition);
    QVector<double> gbestParams = m_globalBestPosition.isEmpty() ? QVector<double>() : buildTraceParameterVector(m_globalBestPosition);

    QStringList enabledIndices;

    for(int i = 0; i < m_enabledParamIndices.size(); ++i) {
        enabledIndices << QString::number(m_enabledParamIndices[i]);
    }

    QStringList cols;
    cols << csvEscape(m_traceRunId)
         << QString::number(generation)
         << QString::number(particleIndex)
         << csvEscape(phase)
         << traceParamAt(currentParams, 0)
         << traceParamAt(currentParams, 1)
         << traceParamAt(currentParams, 2)
         << traceParamAt(currentParams, 3)
         << traceParamAt(currentParams, 5)
         << traceParamAt(currentParams, 7)
         << traceNumber(solverObjective)
         << QString::number(solverSuccess ? 1 : 0)
         << QString::number(elapsedMs)
         << traceNumber(surrogateObjective)
         << csvEscape(screeningDecision)
         << traceNumber(pbestObjective)
         << traceParamAt(pbestParams, 0)
         << traceParamAt(pbestParams, 1)
         << traceParamAt(pbestParams, 2)
         << traceParamAt(pbestParams, 3)
         << traceParamAt(pbestParams, 5)
         << traceParamAt(pbestParams, 7)
         << traceNumber(m_globalBestFitness)
         << traceParamAt(gbestParams, 0)
         << traceParamAt(gbestParams, 1)
         << traceParamAt(gbestParams, 2)
         << traceParamAt(gbestParams, 3)
         << traceParamAt(gbestParams, 5)
         << traceParamAt(gbestParams, 7)
         << csvEscape(enabledIndices.join(";"));

    QTextStream out(&m_traceFile);
    out << cols.join(",") << "\n";
    m_traceFile.flush();
}

void nmCalculationAutoFitPSO::writeIterationTraceRows()
{
    // 每代真实评价和全局最优更新后写出所有粒子。
    // 这样 trace 中能同时看到“这一代谁被筛掉”和“当前 pbest/gbest 是什么”。
    if(!m_traceFile.isOpen()) {
        return;
    }

    for(int i = 0; i < m_swarm.size(); ++i) {
        const AutoFitParticle& particle = m_swarm[i];
        double solverObjective = particle.evaluatedThisIteration
                                 ? particle.fitness
                                 : std::numeric_limits<double>::quiet_NaN();
        writeTraceRow(m_currentIteration,
                      i,
                      particle.evaluatedThisIteration ? "particle_solver" : "particle_not_evaluated",
                      particle.position,
                      solverObjective,
                      particle.lastEvaluationSuccess,
                      particle.lastEvaluationElapsedMs,
                      particle.surrogateObjective,
                      particle.screeningDecision,
                      particle.bestPosition,
                      particle.bestFitness);
    }
}

bool nmCalculationAutoFitPSO::isSurrogateScreeningEnabled() const
{
    // 对应界面上的 PSO acceleration / surrogate screening 开关。
    return m_surrogateScreeningEnabled;
}

QString nmCalculationAutoFitPSO::getMlRootPath() const
{
    // 查找 ML/nmWTAI-ML 根目录。程序可能从 build/bin/release 等不同目录启动，
    // 因此按多个相对路径尝试。isValidMlRootDirectory() 用评分脚本是否存在来确认。
    QString appDir = QApplication::applicationDirPath();
    QString cwd = QDir::currentPath();

    QStringList candidateRoots;
    candidateRoots << QDir::cleanPath(appDir + "/../../ML/nmWTAI-ML")
                   << QDir::cleanPath(appDir + "/../ML/nmWTAI-ML")
                   << QDir::cleanPath(cwd + "/ML/nmWTAI-ML")
                   << QDir::cleanPath(cwd + "/../ML/nmWTAI-ML")
                   << QDir::cleanPath(cwd + "/../../ML/nmWTAI-ML");

    for(int i = 0; i < candidateRoots.size(); ++i) {
        if(isValidMlRootDirectory(candidateRoots[i])) {
            return candidateRoots[i];
        }
    }

    return candidateRoots.isEmpty()
           ? QString()
           : candidateRoots.first();
}

QString nmCalculationAutoFitPSO::getPythonExecutablePath() const
{
    // 优先使用 ML 项目内的虚拟环境 python，其次使用 PATH 中的 python。
    // 这样开发机上只要 ML/nmWTAI-ML/.venv 已配置，就不依赖全局 Python 环境。
    QString mlRoot = getMlRootPath();
    QStringList candidateExecutables;

    if(!mlRoot.isEmpty()) {
        candidateExecutables << QDir(mlRoot).absoluteFilePath(".venv/Scripts/python.exe");
        candidateExecutables << QDir(mlRoot).absoluteFilePath(".venv/bin/python");
    }

    for(int i = 0; i < candidateExecutables.size(); ++i) {
        QFileInfo pythonInfo(candidateExecutables[i]);

        if(pythonInfo.exists() && pythonInfo.isFile()) {
            return QDir::cleanPath(pythonInfo.absoluteFilePath());
        }
    }

    QString resolvedPython = findExecutableInPath("python");

    if(!resolvedPython.isEmpty()) {
        return QDir::cleanPath(resolvedPython);
    }

    return "python";
}

QString nmCalculationAutoFitPSO::getSurrogateTag() const
{
    // 当前 C++ 绑定的代理模型实验标识，对应 ML 侧训练输出目录/checkpoint 命名。
    // 如果以后换模型，需要同步更新这里和 ML 脚本可识别的 tag。
    return "family_random_v2_q_50k_noslope";
}

QString nmCalculationAutoFitPSO::getSurrogateStage() const
{
    // 当前代理模型的数据生成阶段。v2_q 表示训练时包含流量制度 q 的变化。
    return "family_random_v2_q";
}

double nmCalculationAutoFitPSO::getSurrogateKeepFraction() const
{
    // 强递减生产制度下代理排序风险更高，因此保留更多 top 粒子进入真实求解器。
    if(getSurrogateStage().contains("v2_q") && isStrongDecliningProductionSchedule()) {
        return kSurrogateHighRiskKeepFraction;
    }

    return kSurrogateKeepFraction;
}

double nmCalculationAutoFitPSO::getSurrogateAuditFraction() const
{
    // 随机审计比例，审计的是代理模型筛掉的候选。
    return kSurrogateAuditFraction;
}

double nmCalculationAutoFitPSO::getSurrogateMinSolverFraction() const
{
    // 每代真实求解器最低覆盖比例。强递减生产制度下提高到高风险配置。
    if(getSurrogateStage().contains("v2_q") && isStrongDecliningProductionSchedule()) {
        return kSurrogateHighRiskMinSolverFraction;
    }

    return kSurrogateMinSolverFraction;
}

int nmCalculationAutoFitPSO::getSurrogateWarmupIterations() const
{
    // 前若干代不筛选，确保粒子先建立真实 pbest。
    return kSurrogateWarmupIterations;
}

int nmCalculationAutoFitPSO::getSurrogateFullSolverInterval() const
{
    // 周期性全量真实评价间隔，用于审计代理模型排序。
    return kSurrogateFullSolverInterval;
}

bool nmCalculationAutoFitPSO::writeSurrogateCandidateCsv(const QString& candidatePath) const
{
    // 写给 Python 代理模型的候选粒子 CSV。
    //
    // 文件位置通常是 ML/nmWTAI-ML/data/temp/pso_screen_candidates_<run_id>_gen<N>.csv。
    // 字段顺序必须与 ML 脚本 score_pso_candidates*.py 保持一致：
    // particle_id,k,skin,wellboreC,phi,h,Cf
    //
    // 这里写的是“完整参数体系”中的关键字段，不是粒子内部紧凑向量。
    // 例如用户没有勾选 Cf 时，Cf 会从当前 DataManager 取值写入 CSV。
    QFile file(candidatePath);

    if(!file.open(QIODevice::WriteOnly | QIODevice::Text)) {
        DEBUG_OUT(QString("Failed to open surrogate candidate file: %1").arg(candidatePath));
        return false;
    }

    QTextStream out(&file);
    // 代理模型只需要这些输入字段；其它参数当前不在训练输入集中。
    out << "particle_id,k,skin,wellboreC,phi,h,Cf\n";

    for(int i = 0; i < m_swarm.size(); ++i) {
        QVector<double> params = buildTraceParameterVector(m_swarm[i].position);
        QStringList cols;
        cols << QString::number(i)
             << traceParamAt(params, 0)
             << traceParamAt(params, 1)
             << traceParamAt(params, 2)
             << traceParamAt(params, 3)
             << traceParamAt(params, 5)
             << traceParamAt(params, 7);
        out << cols.join(",") << "\n";
    }

    file.flush();
    file.close();
    return true;
}

bool nmCalculationAutoFitPSO::runSurrogateScoringProcess(const QString& candidatePath, const QString& scorePath, QString* failureReason)
{
    // 代理评分统一入口。优先走常驻 Python server，失败后降级为一次性脚本。
    // 常驻 server 可以避免每一代重复加载模型，速度更快；一次性脚本更稳，适合作为兜底。
    QString serverFailure;

    if(requestSurrogateScoresFromServer(candidatePath, scorePath, &serverFailure)) {
        return true;
    }

    emit logMessageGenerated(tr("Surrogate scoring server unavailable; falling back to one-shot Python scoring"));

    if(!serverFailure.isEmpty()) {
        emit logMessageGenerated(tr("Surrogate scoring server detail: %1").arg(serverFailure));
    }

    stopSurrogateScoringServer();
    return runSurrogateScoringScriptOnce(candidatePath, scorePath, failureReason);
}

bool nmCalculationAutoFitPSO::ensureSurrogateScoringServer(QString* failureReason)
{
    // 确保常驻代理评分 server 已启动且与当前 run 配置匹配。
    //
    // serverKey 由 python 路径、脚本路径、trace meta、tag、stage 组成。
    // 只要其中任何一个变化，就停止旧进程并启动新进程，避免拿旧模型或旧目标曲线评分。
    if(m_traceMetaFilePath.isEmpty() || !QFileInfo(m_traceMetaFilePath).exists()) {
        QString reason = "trace meta file is missing";
        DEBUG_OUT(QString("Surrogate scoring server skipped: %1").arg(reason));

        if(failureReason) {
            *failureReason = reason;
        }

        return false;
    }

    QString mlRoot = getMlRootPath();
    QString scriptPath = QDir(mlRoot).absoluteFilePath("scripts/score_pso_candidates_server.py");

    if(!QFileInfo(scriptPath).exists()) {
        QString reason = QString("surrogate scoring server script not found: %1").arg(scriptPath);
        DEBUG_OUT(reason);

        if(failureReason) {
            *failureReason = reason;
        }

        return false;
    }

    QString pythonPath = getPythonExecutablePath();
    QString serverKey = QString("%1|%2|%3|%4|%5")
                        .arg(pythonPath)
                        .arg(scriptPath)
                        .arg(m_traceMetaFilePath)
                        .arg(getSurrogateTag())
                        .arg(getSurrogateStage());

    if(m_surrogateScoringProcess &&
            m_surrogateScoringProcess->state() == QProcess::Running &&
            m_surrogateScoringServerKey == serverKey) {
        return true;
    }

    stopSurrogateScoringServer();

    QStringList args;
    // -u 让 Python stdout/stderr 不缓冲，C++ 可以及时读到 ready/ok 行。
    args << "-u"
         << scriptPath
         << "--trace-meta" << m_traceMetaFilePath
         << "--tag" << getSurrogateTag()
         << "--stage" << getSurrogateStage();

    m_surrogateScoringProcess = new QProcess(this);
    m_surrogateScoringProcess->setWorkingDirectory(mlRoot);

    emit logMessageGenerated(tr("Starting surrogate scoring server: python=%1, mlRoot=%2")
                             .arg(pythonPath)
                             .arg(mlRoot));

    m_surrogateScoringProcess->start(pythonPath, args);

    if(!m_surrogateScoringProcess->waitForStarted(15000)) {
        QString reason = QString("failed to start scoring server: python=%1, script=%2, error=%3")
                         .arg(pythonPath)
                         .arg(scriptPath)
                         .arg(m_surrogateScoringProcess->errorString());
        DEBUG_OUT(reason);
        stopSurrogateScoringServer();

        if(failureReason) {
            *failureReason = reason;
        }

        return false;
    }

    QTime timer;
    timer.start();
    QByteArray readyLine;

    while(timer.elapsed() < 120000) {
        // server 启动协议：Python 首先输出一行 JSON，包含 ready=true 或 ok=false。
        // C++ 不做完整 JSON 解析，只检查关键字段，降低 Qt 版本依赖。
        if(m_surrogateScoringProcess->state() != QProcess::Running) {
            QString stdOut = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardOutput()).trimmed();
            QString stdErr = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardError()).trimmed();
            QString reason = QString("scoring server exited before ready: stdout=%1, stderr=%2")
                             .arg(stdOut.isEmpty() ? QString("<empty>") : stdOut)
                             .arg(stdErr.isEmpty() ? QString("<empty>") : stdErr);
            DEBUG_OUT(reason);
            stopSurrogateScoringServer();

            if(failureReason) {
                *failureReason = reason;
            }

            return false;
        }

        if(!m_surrogateScoringProcess->canReadLine()) {
            m_surrogateScoringProcess->waitForReadyRead(500);
            QApplication::processEvents();
            continue;
        }

        readyLine = m_surrogateScoringProcess->readLine().trimmed();

        if(readyLine.contains("\"ready\"") && readyLine.contains("true")) {
            m_surrogateScoringServerKey = serverKey;
            QString stdErr = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardError()).trimmed();

            if(!stdErr.isEmpty()) {
                emit logMessageGenerated(tr("Surrogate scoring server stderr: %1").arg(stdErr));
            }

            return true;
        }

        if(readyLine.contains("\"ok\"") && readyLine.contains("false")) {
            QString reason = QString("scoring server failed during initialization: %1").arg(QString::fromLocal8Bit(readyLine));
            DEBUG_OUT(reason);
            stopSurrogateScoringServer();

            if(failureReason) {
                *failureReason = reason;
            }

            return false;
        }
    }

    QString stdErr = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardError()).trimmed();
    QString reason = QString("scoring server ready timeout: stderr=%1")
                     .arg(stdErr.isEmpty() ? QString("<empty>") : stdErr);
    DEBUG_OUT(reason);
    stopSurrogateScoringServer();

    if(failureReason) {
        *failureReason = reason;
    }

    return false;
}

bool nmCalculationAutoFitPSO::requestSurrogateScoresFromServer(const QString& candidatePath, const QString& scorePath, QString* failureReason)
{
    // 向常驻 Python server 发送一次评分请求。
    // 请求通过 stdin 传 JSON 行，server 读取候选 CSV 后把评分写入 scorePath，
    // 再通过 stdout 返回 ok=true/false。
    if(!ensureSurrogateScoringServer(failureReason)) {
        return false;
    }

    QFile::remove(scorePath);

    QString command = QString("{\"cmd\":\"score\",\"candidates\":%1,\"output\":%2}\n")
                      .arg(jsonEscape(candidatePath))
                      .arg(jsonEscape(scorePath));

    QByteArray bytes = command.toUtf8();
    qint64 written = m_surrogateScoringProcess->write(bytes);

    if(written != bytes.size() || !m_surrogateScoringProcess->waitForBytesWritten(10000)) {
        QString reason = QString("failed to send scoring request to server: written=%1 expected=%2 error=%3")
                         .arg(written)
                         .arg(bytes.size())
                         .arg(m_surrogateScoringProcess->errorString());
        DEBUG_OUT(reason);

        if(failureReason) {
            *failureReason = reason;
        }

        return false;
    }

    QTime timer;
    timer.start();

    while(timer.elapsed() < 120000) {
        // 每次请求最多等待 120 秒。超时或 server 退出都会让本代代理筛选降级为全量真实求解器。
        if(m_surrogateScoringProcess->state() != QProcess::Running) {
            QString stdOut = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardOutput()).trimmed();
            QString stdErr = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardError()).trimmed();
            QString reason = QString("scoring server exited during request: stdout=%1, stderr=%2")
                             .arg(stdOut.isEmpty() ? QString("<empty>") : stdOut)
                             .arg(stdErr.isEmpty() ? QString("<empty>") : stdErr);
            DEBUG_OUT(reason);

            if(failureReason) {
                *failureReason = reason;
            }

            return false;
        }

        if(!m_surrogateScoringProcess->canReadLine()) {
            m_surrogateScoringProcess->waitForReadyRead(500);
            QApplication::processEvents();
            continue;
        }

        QByteArray lineBytes = m_surrogateScoringProcess->readLine().trimmed();
        QString line = QString::fromLocal8Bit(lineBytes);

        if(line.contains("\"ok\"") && line.contains("true")) {
            QFileInfo scoreInfo(scorePath);

            if(scoreInfo.exists() && scoreInfo.size() > 0) {
                QString stdErr = QString::fromLocal8Bit(m_surrogateScoringProcess->readAllStandardError()).trimmed();

                if(!stdErr.isEmpty()) {
                    emit logMessageGenerated(tr("Surrogate scoring server stderr: %1").arg(stdErr));
                }

                return true;
            }

            QString reason = QString("scoring server reported success but score file is missing: %1").arg(scorePath);
            DEBUG_OUT(reason);

            if(failureReason) {
                *failureReason = reason;
            }

            return false;
        }

        if(line.contains("\"ok\"") && line.contains("false")) {
            QString reason = QString("scoring server request failed: %1").arg(line);
            DEBUG_OUT(reason);

            if(failureReason) {
                *failureReason = reason;
            }

            return false;
        }
    }

    QString reason = "scoring server request timed out";
    DEBUG_OUT(reason);

    if(failureReason) {
        *failureReason = reason;
    }

    return false;
}

void nmCalculationAutoFitPSO::stopSurrogateScoringServer()
{
    // 停止常驻 Python 评分进程。
    // 先发送 quit 指令，短时间内不退出再 kill，避免应用关闭时遗留 python.exe。
    if(!m_surrogateScoringProcess) {
        return;
    }

    if(m_surrogateScoringProcess->state() == QProcess::Running) {
        m_surrogateScoringProcess->write("{\"cmd\":\"quit\"}\n");
        m_surrogateScoringProcess->waitForBytesWritten(1000);
        m_surrogateScoringProcess->waitForFinished(3000);

        if(m_surrogateScoringProcess->state() == QProcess::Running) {
            m_surrogateScoringProcess->kill();
            m_surrogateScoringProcess->waitForFinished(3000);
        }
    }

    delete m_surrogateScoringProcess;
    m_surrogateScoringProcess = nullptr;
    m_surrogateScoringServerKey.clear();
}

bool nmCalculationAutoFitPSO::runSurrogateScoringScriptOnce(const QString& candidatePath, const QString& scorePath, QString* failureReason)
{
    // 一次性代理评分脚本兜底路径。
    //
    // 输入：
    // - candidates：writeSurrogateCandidateCsv() 写出的粒子参数；
    // - trace-meta：目标曲线、流量制度、参数上下界等上下文；
    // - tag/stage：选择 ML 侧模型。
    //
    // 输出：
    // - scorePath：Python 写出的评分 CSV，随后由 readSurrogateScores() 读取。
    if(m_traceMetaFilePath.isEmpty() || !QFileInfo(m_traceMetaFilePath).exists()) {
        QString reason = "trace meta file is missing";
        DEBUG_OUT(QString("Surrogate scoring skipped: %1").arg(reason));

        if(failureReason) {
            *failureReason = reason;
        }

        return false;
    }

    QString mlRoot = getMlRootPath();
    QString scriptPath = QDir(mlRoot).absoluteFilePath("scripts/score_pso_candidates.py");

    if(!QFileInfo(scriptPath).exists()) {
        QString reason = QString("surrogate scoring script not found: %1").arg(scriptPath);
        DEBUG_OUT(reason);

        if(failureReason) {
            *failureReason = reason;
        }

        return false;
    }

    QString pythonPath = getPythonExecutablePath();
    QStringList args;
    args << scriptPath
         << "--candidates" << candidatePath
         << "--trace-meta" << m_traceMetaFilePath
         << "--output" << scorePath
         << "--tag" << getSurrogateTag()
         << "--stage" << getSurrogateStage();

    QFile::remove(scorePath);

    QString lastFailure;

    for(int attempt = 1; attempt <= kSurrogateScoringMaxAttempts; ++attempt) {
        // 评分脚本可能因首次加载模型、文件占用或环境抖动失败，因此允许有限重试。
        emit logMessageGenerated(tr("Surrogate scoring attempt %1/%2: python=%3, mlRoot=%4")
                                 .arg(attempt)
                                 .arg(kSurrogateScoringMaxAttempts)
                                 .arg(pythonPath)
                                 .arg(mlRoot));

        QProcess process;
        process.setWorkingDirectory(mlRoot);
        process.start(pythonPath, args);

        if(!process.waitForStarted(10000)) {
            lastFailure = QString("failed to start scoring process (attempt %1/%2): python=%3, script=%4, error=%5")
                          .arg(attempt)
                          .arg(kSurrogateScoringMaxAttempts)
                          .arg(pythonPath)
                          .arg(scriptPath)
                          .arg(process.errorString());
            DEBUG_OUT(lastFailure);

            if(attempt < kSurrogateScoringMaxAttempts) {
                emit logMessageGenerated(tr("Surrogate scoring start failed, retrying once: %1").arg(process.errorString()));
                msleep(500);
                continue;
            }

            if(failureReason) {
                *failureReason = lastFailure;
            }

            return false;
        }

        if(!process.waitForFinished(120000)) {
            process.kill();
            process.waitForFinished(3000);
            QString stdOut = QString::fromLocal8Bit(process.readAllStandardOutput()).trimmed();
            QString stdErr = QString::fromLocal8Bit(process.readAllStandardError()).trimmed();
            lastFailure = QString("scoring process timed out (attempt %1/%2): python=%3, script=%4, stdout=%5, stderr=%6")
                          .arg(attempt)
                          .arg(kSurrogateScoringMaxAttempts)
                          .arg(pythonPath)
                          .arg(scriptPath)
                          .arg(stdOut.isEmpty() ? QString("<empty>") : stdOut)
                          .arg(stdErr.isEmpty() ? QString("<empty>") : stdErr);
            DEBUG_OUT(lastFailure);

            if(attempt < kSurrogateScoringMaxAttempts) {
                emit logMessageGenerated(tr("Surrogate scoring timed out, retrying once"));
                msleep(500);
                continue;
            }

            if(failureReason) {
                *failureReason = lastFailure;
            }

            return false;
        }

        QString stdOut = QString::fromLocal8Bit(process.readAllStandardOutput()).trimmed();
        QString stdErr = QString::fromLocal8Bit(process.readAllStandardError()).trimmed();

        if(process.exitStatus() != QProcess::NormalExit || process.exitCode() != 0) {
            lastFailure = QString("scoring process failed (attempt %1/%2): exitStatus=%3, exitCode=%4, python=%5, script=%6, stdout=%7, stderr=%8")
                          .arg(attempt)
                          .arg(kSurrogateScoringMaxAttempts)
                          .arg(process.exitStatus() == QProcess::NormalExit ? "NormalExit" : "CrashExit")
                          .arg(process.exitCode())
                          .arg(pythonPath)
                          .arg(scriptPath)
                          .arg(stdOut.isEmpty() ? QString("<empty>") : stdOut)
                          .arg(stdErr.isEmpty() ? QString("<empty>") : stdErr);
            DEBUG_OUT(lastFailure);

            if(attempt < kSurrogateScoringMaxAttempts) {
                emit logMessageGenerated(tr("Surrogate scoring process failed, retrying once: exitCode=%1").arg(process.exitCode()));
                msleep(500);
                continue;
            }

            if(failureReason) {
                *failureReason = lastFailure;
            }

            return false;
        }

        QFileInfo scoreInfo(scorePath);

        if(!scoreInfo.exists() || scoreInfo.size() <= 0) {
            lastFailure = QString("scoring process finished without a usable score file (attempt %1/%2): output=%3, stdout=%4, stderr=%5")
                          .arg(attempt)
                          .arg(kSurrogateScoringMaxAttempts)
                          .arg(scorePath)
                          .arg(stdOut.isEmpty() ? QString("<empty>") : stdOut)
                          .arg(stdErr.isEmpty() ? QString("<empty>") : stdErr);
            DEBUG_OUT(lastFailure);

            if(attempt < kSurrogateScoringMaxAttempts) {
                emit logMessageGenerated(tr("Surrogate scoring produced no score file, retrying once"));
                msleep(500);
                continue;
            }

            if(failureReason) {
                *failureReason = lastFailure;
            }

            return false;
        }

        if(!stdErr.isEmpty()) {
            emit logMessageGenerated(tr("Surrogate scoring stderr: %1").arg(stdErr));
        }

        if(!stdOut.isEmpty()) {
            emit logMessageGenerated(tr("Surrogate scoring stdout: %1").arg(stdOut));
        }

        return true;
    }

    if(failureReason) {
        *failureReason = lastFailure;
    }

    return false;
}

QVector<double> nmCalculationAutoFitPSO::readSurrogateScores(const QString& scorePath) const
{
    // 读取 Python 输出的评分 CSV。
    //
    // 当前脚本约定至少包含：
    // particle_id,surrogate_objective,...,success
    // 本函数只读取 particle_id、score 和 success 标志。失败或缺失的粒子保持 NaN，
    // 后续 buildSurrogateEvaluationMask() 会通过 fallback_gate 强制跑真实求解器。
    QVector<double> scores(m_swarm.size(), std::numeric_limits<double>::quiet_NaN());
    QFile file(scorePath);

    if(!file.open(QIODevice::ReadOnly | QIODevice::Text)) {
        DEBUG_OUT(QString("Failed to open surrogate score file: %1").arg(scorePath));
        return scores;
    }

    QTextStream in(&file);
    bool firstLine = true;

    while(!in.atEnd()) {
        QString line = in.readLine().trimmed();

        if(line.isEmpty()) {
            continue;
        }

        if(firstLine) {
            firstLine = false;
            continue;
        }

        QStringList parts = line.split(",");

        if(parts.size() < 5) {
            continue;
        }

        bool idOk = false;
        bool scoreOk = false;
        int particleId = parts[0].toInt(&idOk);
        double score = parts[1].toDouble(&scoreOk);
        bool success = parts[4].trimmed() == "1";

        if(idOk && scoreOk && success && particleId >= 0 && particleId < scores.size()) {
            scores[particleId] = score;
        }
    }

    return scores;
}

bool nmCalculationAutoFitPSO::isSurrogateRunContextSupported(QString* reason) const
{
    // 判断“当前项目工况”是否适合使用代理模型。
    // 这是运行级别的 gate：只要项目整体不满足训练域条件，本次 PSO 全程退回真实求解器。
    // 这样做比盲目调用代理模型更稳，因为代理模型只在特定 PEBI/单井/垂直井/无复杂几何场景训练过。
    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();

    if(!dataManager) {
        if(reason) *reason = "data manager is unavailable";

        return false;
    }

    // 网格类型 gate：当前代理模型只按 PEBI 网格样本训练。
    if(dataManager->getGridType() != NM_Grid_PEBI) {
        if(reason) *reason = "surrogate model is trained only for PEBI grid cases";

        return false;
    }

    nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);

    if(!pTargetWell) {
        if(reason) *reason = QString("target well '%1' is missing").arg(m_targetWellName);

        return false;
    }

    // 井型 gate：当前只支持垂直井。水平井/压裂井等几何特征不在训练集中。
    if(pTargetWell->getWellType() != NM_WELL_MODEL::Vertical_Well) {
        if(reason) *reason = "surrogate model is currently limited to vertical-well cases";

        return false;
    }

    int validCalculationWells = 0;
    QVector<QPair<NM_WELL_MODEL, QString> > calculationWells = dataManager->getCalculationWells();

    for(int i = 0; i < calculationWells.size(); ++i) {
        if(calculationWells[i].first != NM_WELL_MODEL::Unknow_Well) {
            validCalculationWells++;
        }
    }

    // 单井 gate：训练数据假设只有一口有效计算井，多井干扰会改变压力响应形态。
    if(validCalculationWells != 1) {
        if(reason) *reason = QString("surrogate model expects one active calculation well, got %1").arg(validCalculationWells);

        return false;
    }

    // 几何 gate：断层、裂缝、区域变化会引入训练集中没有的边界/非均质影响。
    if(!dataManager->getFaultDataList().isEmpty() ||
            !dataManager->getFractureDataList().isEmpty() ||
            !dataManager->getRegionDataList().isEmpty()) {
        if(reason) *reason = "surrogate model is not enabled for changed fault/fracture/region geometry";

        return false;
    }

    // 参数 gate：代理输入目前只覆盖 k/skin/wellboreC/phi/h。
    // 用户勾选其它参数时，代理无法可靠反映这些参数变化，直接禁用代理筛选。
    QVector<int> allowedParamIndices;
    allowedParamIndices << 0 << 1 << 2 << 3 << 5;

    for(int i = 0; i < m_enabledParamIndices.size(); ++i) {
        if(!allowedParamIndices.contains(m_enabledParamIndices[i])) {
            if(reason) {
                QStringList names = traceParameterNames();
                int idx = m_enabledParamIndices[i];
                QString name = (idx >= 0 && idx < names.size()) ? names[idx] : QString::number(idx);
                *reason = QString("parameter '%1' is outside surrogate input set").arg(name);
            }

            return false;
        }
    }

    nmDataReservoir reservoirData = dataManager->getReservoirDataCopy();
    double cf = reservoirData.getCf().getValue().toDouble();

    // Cf gate：当前模型按近似固定 Cf 训练，允许小范围相对偏差。
    if(!isNearRelative(cf, kSurrogateCfFixed, kSurrogateCfRelTol)) {
        if(reason) {
            *reason = QString("Cf=%1 is outside surrogate fixed-Cf domain").arg(cf, 0, 'g', 10);
        }

        return false;
    }

    QVector<QPointF> flowPoints = pTargetWell->getFlowPoints();

    // flowPoints 有时第一段是 (0,0) 占位点，不代表真实生产段，先移除。
    if(!flowPoints.isEmpty() && qAbs(flowPoints[0].x()) <= 1e-12 && qAbs(flowPoints[0].y()) <= 1e-12) {
        flowPoints.remove(0);
    }

    if(flowPoints.size() < 2) {
        if(reason) *reason = "flow schedule is too short for surrogate screening";

        return false;
    }

    // 约定最后一段是关井段，因此生产段数量 = 总段数 - 1。
    int productionSections = flowPoints.size() - 1;

    if(productionSections < kSurrogateMinProdSections || productionSections > kSurrogateMaxProdSections) {
        if(reason) {
            *reason = QString("production section count %1 is outside surrogate range [%2,%3]")
                      .arg(productionSections)
                      .arg(kSurrogateMinProdSections)
                      .arg(kSurrogateMaxProdSections);
        }

        return false;
    }

    double productionTime = 0.0;

    for(int i = 0; i < productionSections; ++i) {
        double dt = flowPoints[i].x();
        double q = flowPoints[i].y();

        if(!isFiniteNumber(dt) || dt <= 0.0 || !isInClosedRange(q, kSurrogateQMin, kSurrogateQMax)) {
            if(reason) {
                *reason = QString("production schedule section %1 is outside surrogate dt/q domain").arg(i + 1);
            }

            return false;
        }

        if(i > 0) {
            // 相邻生产段产量跳变不能太大，否则代理模型可能没见过这种制度。
            double previousQ = flowPoints[i - 1].y();
            double stepRatio = qMax(q, previousQ) / qMax(1e-12, qMin(q, previousQ));

            if(stepRatio > kSurrogateMaxRateStepRatio) {
                if(reason) {
                    *reason = QString("rate step ratio %1 exceeds surrogate domain").arg(stepRatio, 0, 'g', 10);
                }

                return false;
            }
        }

        productionTime += dt;
    }

    // 旧阶段模型对强递减制度不稳；v2_q 阶段允许但会提高真实求解比例。
    if(!getSurrogateStage().contains("v2_q")) {
        double endStartRatio = std::numeric_limits<double>::quiet_NaN();

        if(isStrongDecliningProductionSchedule(&endStartRatio)) {
            if(reason) {
                *reason = QString("strongly decreasing production schedule ratio %1 is outside surrogate domain")
                          .arg(endStartRatio, 0, 'g', 10);
            }

            return false;
        }
    }

    if(!isInClosedRange(productionTime, kSurrogateProdTimeMin, kSurrogateProdTimeMax)) {
        if(reason) {
            *reason = QString("production total time %1 is outside surrogate range").arg(productionTime, 0, 'g', 10);
        }

        return false;
    }

    double shutinTime = flowPoints.last().x();
    double shutinRate = flowPoints.last().y();

    if(!isInClosedRange(shutinTime, kSurrogateShutinTimeMin, kSurrogateShutinTimeMax) ||
            qAbs(shutinRate) > kSurrogateQZeroThreshold) {
        if(reason) *reason = "last flow section is not a trained shut-in section";

        return false;
    }

    // 当前目标段必须是最后的关井段，对应试井解释的双对数分析段。
    if(pTargetWell->getIndexF() != flowPoints.size()) {
        if(reason) {
            *reason = QString("target section index %1 is not the last section %2")
                      .arg(pTargetWell->getIndexF())
                      .arg(flowPoints.size());
        }

        return false;
    }

    return true;
}

bool nmCalculationAutoFitPSO::isSurrogateCandidateInDomain(const QVector<double>& selectedParameters, QString* reason) const
{
    // 判断“单个粒子候选参数”是否在代理模型训练域内。
    // 与 isSurrogateRunContextSupported() 不同，这里检查的是粒子参数值本身。
    // 超出范围的粒子会被 domain_gate 强制跑真实求解器。
    QVector<double> params = buildTraceParameterVector(selectedParameters);
    double k = params.size() > 0 ? params[0] : std::numeric_limits<double>::quiet_NaN();
    double skin = params.size() > 1 ? params[1] : std::numeric_limits<double>::quiet_NaN();
    double wellboreC = params.size() > 2 ? params[2] : std::numeric_limits<double>::quiet_NaN();
    double phi = params.size() > 3 ? params[3] : std::numeric_limits<double>::quiet_NaN();
    double h = params.size() > 5 ? params[5] : std::numeric_limits<double>::quiet_NaN();
    double cf = params.size() > 7 ? params[7] : std::numeric_limits<double>::quiet_NaN();

    if(!isInClosedRange(k, kSurrogateKMin, kSurrogateKMax)) {
        if(reason) *reason = QString("k=%1").arg(k, 0, 'g', 10);

        return false;
    }

    if(!isInClosedRange(skin, kSurrogateSkinMin, kSurrogateSkinMax)) {
        if(reason) *reason = QString("skin=%1").arg(skin, 0, 'g', 10);

        return false;
    }

    if(!isInClosedRange(wellboreC, kSurrogateWellboreCMin, kSurrogateWellboreCMax)) {
        if(reason) *reason = QString("wellboreC=%1").arg(wellboreC, 0, 'g', 10);

        return false;
    }

    if(!isInClosedRange(phi, kSurrogatePhiMin, kSurrogatePhiMax)) {
        if(reason) *reason = QString("phi=%1").arg(phi, 0, 'g', 10);

        return false;
    }

    if(!isInClosedRange(h, kSurrogateHMin, kSurrogateHMax)) {
        if(reason) *reason = QString("h=%1").arg(h, 0, 'g', 10);

        return false;
    }

    if(!isNearRelative(cf, kSurrogateCfFixed, kSurrogateCfRelTol)) {
        if(reason) *reason = QString("Cf=%1").arg(cf, 0, 'g', 10);

        return false;
    }

    return true;
}

bool nmCalculationAutoFitPSO::forceSolverByFallbackGate(const QVector<double>& selectedParameters) const
{
    // 经验 fallback gate。
    // 有些点虽然还在训练域边界内，但组合起来属于代理模型容易排序失真的区域，
    // 例如极低渗透率配大井筒储集、极端 skin 等。遇到这些候选时直接跑真实求解器。
    QVector<double> params = buildTraceParameterVector(selectedParameters);
    double k = params.size() > 0 ? params[0] : std::numeric_limits<double>::quiet_NaN();
    double skin = params.size() > 1 ? params[1] : std::numeric_limits<double>::quiet_NaN();
    double wellboreC = params.size() > 2 ? params[2] : std::numeric_limits<double>::quiet_NaN();
    double phi = params.size() > 3 ? params[3] : std::numeric_limits<double>::quiet_NaN();
    double h = params.size() > 5 ? params[5] : std::numeric_limits<double>::quiet_NaN();

    if(!isFiniteNumber(k) || !isFiniteNumber(skin) || !isFiniteNumber(wellboreC) || !isFiniteNumber(phi) || !isFiniteNumber(h)) {
        return true;
    }

    if(k <= 1e-3 && wellboreC > 0.5) {
        return true;
    }

    if(phi < 0.03 && h < 4.0) {
        return true;
    }

    if(k < 0.1 && h >= 50.0) {
        return true;
    }

    if(qAbs(skin) > 6.0) {
        return true;
    }

    return false;
}

bool nmCalculationAutoFitPSO::isStrongDecliningProductionSchedule(double* endStartRatio) const
{
    // 判断生产段产量是否整体非增且首末下降明显。
    // 这类流量制度代理模型风险更高，所以 keep/min_solver 会自动提高。
    if(endStartRatio) {
        *endStartRatio = std::numeric_limits<double>::quiet_NaN();
    }

    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();

    if(!dataManager) {
        return false;
    }

    nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);

    if(!pTargetWell) {
        return false;
    }

    QVector<QPointF> flowPoints = pTargetWell->getFlowPoints();

    if(!flowPoints.isEmpty() && qAbs(flowPoints[0].x()) <= 1e-12 && qAbs(flowPoints[0].y()) <= 1e-12) {
        flowPoints.remove(0);
    }

    if(flowPoints.size() < 2) {
        return false;
    }

    int productionSections = flowPoints.size() - 1;
    bool productionRatesNonIncreasing = true;
    bool productionRatesHaveMeaningfulDrop = false;

    for(int i = 0; i < productionSections; ++i) {
        double q = flowPoints[i].y();

        if(!isFiniteNumber(q)) {
            return false;
        }

        if(i > 0) {
            double previousQ = flowPoints[i - 1].y();

            if(q > previousQ * (1.0 + kSurrogateRateTrendRelTol)) {
                productionRatesNonIncreasing = false;
            }

            if(q < previousQ * (1.0 - kSurrogateRateTrendRelTol)) {
                productionRatesHaveMeaningfulDrop = true;
            }
        }
    }

    if(!productionRatesNonIncreasing || !productionRatesHaveMeaningfulDrop) {
        return false;
    }

    double firstRate = flowPoints.first().y();
    double lastProductionRate = flowPoints[productionSections - 1].y();
    double ratio = lastProductionRate / qMax(1e-12, firstRate);

    if(endStartRatio) {
        *endStartRatio = ratio;
    }

    return ratio < kSurrogateStrongDeclineEndStartRatio;
}

QVector<bool> nmCalculationAutoFitPSO::buildSurrogateEvaluationMask()
{
    // 返回值 evaluateMask 与 m_swarm 一一对应：
    // true  表示该粒子本代需要继续跑真实数值求解器；
    // false 表示该粒子本代被代理模型筛掉，只记录 screened_out，不更新 pbest/gbest。
    //
    // 这个函数是 PSO acceleration 的唯一入口。为了保证结果可信，它包含多层保护：
    // 1. 工况不支持代理模型时，全量真实求解器；
    // 2. 前几代 warmup，全量真实求解器；
    // 3. 周期性审计，全量真实求解器；
    // 4. 代理评分失败时，全量真实求解器；
    // 5. top-k + random audit + fallback + min solver floor 共同决定最终 mask。
    QVector<bool> evaluateMask(m_swarm.size(), true);

    for(int i = 0; i < m_swarm.size(); ++i) {
        m_swarm[i].surrogateObjective = std::numeric_limits<double>::quiet_NaN();
        m_swarm[i].selectedForSolver = true;
        m_swarm[i].selectedByAudit = false;
        m_swarm[i].screeningDecision = "full_solver";
    }

    if(!isSurrogateScreeningEnabled()) {
        // 用户关闭 PSO acceleration：保持原始 PSO 行为，所有粒子都跑真实求解器。
        return evaluateMask;
    }

    QString contextReason;

    if(!isSurrogateRunContextSupported(&contextReason)) {
        // 当前项目工况不在代理模型训练域内。这里不报错，只降级为全量真实求解器，
        // 这样用户仍能完成自动拟合，只是没有代理加速收益。
        m_surrogateContextBlockedIterationCount++;

        for(int i = 0; i < m_swarm.size(); ++i) {
            m_swarm[i].screeningDecision = "context_gate_full_solver";
        }

        emit logMessageGenerated(tr("Surrogate screening disabled for this run context: %1").arg(contextReason));
        return evaluateMask;
    }

    // 前几代全量真实评价，让每个粒子都先获得可信的个体最优 pbest。
    // 否则代理模型一开始就筛掉粒子，可能导致某些粒子从未有真实误差基准。
    if(m_currentIteration < getSurrogateWarmupIterations()) {
        m_surrogateWarmupIterationCount++;

        for(int i = 0; i < m_swarm.size(); ++i) {
            m_swarm[i].screeningDecision = "warmup_full_solver";
        }

        return evaluateMask;
    }

    if(getSurrogateFullSolverInterval() > 0 &&
            m_currentIteration > 0 &&
            ((m_currentIteration + 1) % getSurrogateFullSolverInterval()) == 0) {
        // 周期性全量审计。它牺牲少量速度，换取对代理排序偏差的长期纠偏能力。
        m_surrogatePeriodicAuditIterationCount++;

        for(int i = 0; i < m_swarm.size(); ++i) {
            m_swarm[i].screeningDecision = "periodic_full_solver";
        }

        emit logMessageGenerated(tr("Surrogate audit iteration %1: running full solver for all particles")
                                 .arg(m_currentIteration + 1));
        return evaluateMask;
    }

    QString mlRoot = getMlRootPath();
    QDir tempDir(QDir(mlRoot).absoluteFilePath("data/temp"));

    if(!tempDir.exists()) {
        QDir().mkpath(tempDir.absolutePath());
    }

    QString candidatePath = tempDir.absoluteFilePath(QString("pso_screen_candidates_%1_gen%2.csv")
                            .arg(m_traceRunId)
                            .arg(m_currentIteration));// C++ 写给 Python 的候选粒子参数
    QString scorePath = tempDir.absoluteFilePath(QString("pso_screen_scores_%1_gen%2.csv")
                        .arg(m_traceRunId)
                        .arg(m_currentIteration));// Python 写回给 C++ 的代理评分结果

    QString scoringFailureReason;

    if(!writeSurrogateCandidateCsv(candidatePath) || !runSurrogateScoringProcess(candidatePath, scorePath, &scoringFailureReason)) {
        // Python 环境、checkpoint、processed 数据或脚本异常时都走这里。
        // 降级策略是全量真实求解器，保证代理模块故障不会中断 PSO 拟合。
        emit logMessageGenerated(tr("Surrogate screening unavailable; falling back to full solver for iteration %1")
                                 .arg(m_currentIteration + 1));

        if(!scoringFailureReason.isEmpty()) {
            emit logMessageGenerated(tr("Surrogate scoring failure detail: %1").arg(scoringFailureReason));
        }

        return evaluateMask;
    }

    QVector<double> scores = readSurrogateScores(scorePath);
    QVector<int> finiteScoreIndices;

    // 只接受成功写出的有限代理目标函数。失败候选后面会通过 fallback 强制真实评价。
    for(int i = 0; i < scores.size(); ++i) {
        m_swarm[i].surrogateObjective = scores[i];

        if(isFiniteNumber(scores[i])) {
            finiteScoreIndices.append(i);
        }
    }

    if(finiteScoreIndices.isEmpty()) {
        emit logMessageGenerated(tr("Surrogate scoring returned no valid scores; falling back to full solver"));
        return evaluateMask;
    }

    std::sort(finiteScoreIndices.begin(), finiteScoreIndices.end(), [&](int a, int b) {
        return scores[a] < scores[b];
    });

    // 代理目标函数越小越好。先将所有粒子设为 screened_out，再逐步放回需要真实评价的粒子。
    evaluateMask.fill(false);

    for(int i = 0; i < m_swarm.size(); ++i) {
        m_swarm[i].selectedForSolver = false;
        m_swarm[i].screeningDecision = "screened_out";
    }

    int keepN = qMax(1, qCeil(m_swarm.size() * getSurrogateKeepFraction()));
    keepN = qMin(keepN, finiteScoreIndices.size());

    // 代理评分 top-k：这些粒子最可能给出好解，优先跑真实求解器。
    for(int rank = 0; rank < keepN; ++rank) {
        int idx = finiteScoreIndices[rank];
        evaluateMask[idx] = true;
        m_swarm[idx].selectedForSolver = true;
        m_swarm[idx].screeningDecision = "surrogate_topk";
    }

    QVector<QPair<double, int> > auditCandidates;

    // 随机审计：从代理模型认为“不好”的候选中抽查一部分。
    // 使用 deterministicAuditRandom01 保证同一随机种子下 trace 可复现。
    for(int i = 0; i < finiteScoreIndices.size(); ++i) {
        int idx = finiteScoreIndices[i];

        if(!evaluateMask[idx]) {
            auditCandidates.append(qMakePair(deterministicAuditRandom01(m_psoRandomSeed, m_currentIteration, idx), idx));
        }
    }

    std::sort(auditCandidates.begin(), auditCandidates.end(), [](const QPair<double, int>& a, const QPair<double, int>& b) {
        return a.first < b.first;
    });

    int auditN = qMin(auditCandidates.size(), qCeil(m_swarm.size() * getSurrogateAuditFraction()));

    for(int i = 0; i < auditN; ++i) {
        int idx = auditCandidates[i].second;
        evaluateMask[idx] = true;
        m_swarm[idx].selectedForSolver = true;
        m_swarm[idx].selectedByAudit = true;
        m_swarm[idx].screeningDecision = "random_audit";
    }

    int fallbackCount = 0;
    int domainFallbackCount = 0;

    // fallback gate：只要候选点超出代理训练域、代理评分无效、粒子还没有可靠 pbest，
    // 或落入经验高风险区域，就强制真实求解。这样做是为了保护 PSO 收敛质量。
    for(int i = 0; i < m_swarm.size(); ++i) {
        QString domainReason;
        bool domainFallback = !isSurrogateCandidateInDomain(m_swarm[i].position, &domainReason);
        bool fallback = !isFiniteNumber(scores[i]) ||
                        m_swarm[i].bestFitness >= 1e9 ||
                        forceSolverByFallbackGate(m_swarm[i].position) ||
                        domainFallback;

        if(fallback) {
            if(!evaluateMask[i]) {
                fallbackCount++;
            }

            if(domainFallback) {
                domainFallbackCount++;
            }

            evaluateMask[i] = true;
            m_swarm[i].selectedForSolver = true;
            m_swarm[i].screeningDecision = domainFallback ? "domain_gate" : "fallback_gate";
        }
    }

    int selectedCount = 0;

    for(int i = 0; i < evaluateMask.size(); ++i) {
        if(evaluateMask[i]) {
            selectedCount++;
        }
    }

    int minSolverN = qMax(1, qCeil(m_swarm.size() * getSurrogateMinSolverFraction()));

    if(selectedCount < minSolverN) {
        // 最小真实求解比例：防止代理模型过度筛选，导致一代中真实误差信息太少。
        QVector<QPair<double, int> > extraCandidates;

        for(int i = 0; i < m_swarm.size(); ++i) {
            if(!evaluateMask[i] && isFiniteNumber(scores[i])) {
                extraCandidates.append(qMakePair(scores[i], i));
            }
        }

        std::sort(extraCandidates.begin(), extraCandidates.end(), [](const QPair<double, int>& a, const QPair<double, int>& b) {
            return a.first < b.first;
        });

        for(int i = 0; i < extraCandidates.size() && selectedCount < minSolverN; ++i) {
            int idx = extraCandidates[i].second;
            evaluateMask[idx] = true;
            m_swarm[idx].selectedForSolver = true;
            m_swarm[idx].screeningDecision = "min_solver_floor";
            selectedCount++;
        }
    }

    emit logMessageGenerated(tr("Surrogate screening iteration %1: selected %2/%3 particles (keep=%4, audit=%5, min_solver=%6, fallback=%7, domain=%8)")
                             .arg(m_currentIteration + 1)
                             .arg(selectedCount)
                             .arg(m_swarm.size())
                             .arg(getSurrogateKeepFraction(), 0, 'f', 2)
                             .arg(getSurrogateAuditFraction(), 0, 'f', 2)
                             .arg(getSurrogateMinSolverFraction(), 0, 'f', 2)
                             .arg(fallbackCount)
                             .arg(domainFallbackCount));

    m_surrogateActiveIterationCount++;

    for(int i = 0; i < m_swarm.size(); ++i) {
        const QString& decision = m_swarm[i].screeningDecision;

        if(m_swarm[i].selectedForSolver) {
            m_surrogateSelectedParticleCount++;
        }

        if(decision == "screened_out") {
            m_surrogateScreenedParticleCount++;
        } else if(decision == "surrogate_topk") {
            m_surrogateTopKParticleCount++;
        } else if(decision == "random_audit") {
            m_surrogateAuditParticleCount++;
        } else if(decision == "fallback_gate") {
            m_surrogateFallbackParticleCount++;
        } else if(decision == "domain_gate") {
            m_surrogateDomainParticleCount++;
        } else if(decision == "min_solver_floor") {
            m_surrogateMinFloorParticleCount++;
        }
    }

    return evaluateMask;
}

//QString nmCalculationAutoFitPSO::getTargetWellName() const
//{
//	return m_targetWellName;
//}

// ==================== 数据加载方法 ====================

bool nmCalculationAutoFitPSO::loadAllConfigFromDataManager()
{
    // 统一从 DataManager 加载本次运行所需配置。
    // UI 层只负责把用户选择保存到 nmDataAutomaticFitting，本类从这里开始完全数据驱动。
    try {
        loadOptimizationConfig();
        loadParameterBounds();
        extractUserInitialValues(); // 直接提取初始值，无需条件判断
        return true;
    } catch(...) {
        m_lastError = "Failed to load configuration from data manager";
        return false;
    }
}

void nmCalculationAutoFitPSO::loadOptimizationConfig()
{
    // 读取“优化控制”配置：迭代次数、目标误差、是否启用代理筛选。
    // PSO 的 swarm/inertia/cognitive/social 目前是类内固定默认值，
    // 不是界面输入项；后续若要开放高级设置，可以从这里扩展。
    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
    nmDataAutomaticFitting fittingData = dataManager->getAutomaticFittingDataCopy();

    // 加载基本配置
    m_maxIterations = fittingData.getIterationCount().getValue().toInt();
    m_targetError = fittingData.getErrorTolerance().getValue().toDouble();
    m_surrogateScreeningEnabled = fittingData.getSurrogateScreeningEnabled();

    // 设置 PSO 默认参数：
    // swarmSize=20 控制每代粒子数；inertia/cognitive/social 控制搜索惯性、
    // 个体经验和群体经验的权重。adaptiveParameterUpdate() 会在运行中微调这些值。
    m_swarmSize = 20;
    m_inertiaWeight = 0.4;
    m_cognitiveParam = 2.0;
    m_socialParam = 1.2;

    DEBUG_OUT(QString("Loaded optimization config: iterations=%1, error=%2, surrogate=%3")
              .arg(m_maxIterations).arg(m_targetError).arg(m_surrogateScreeningEnabled ? "enabled" : "disabled"));
}

void nmCalculationAutoFitPSO::loadParameterBounds()
{
    // 读取用户勾选的拟合参数及上下界。
    //
    // 这里构建三个核心数组：
    // - m_parameterSelected[11]：完整参数体系中每个参数是否参与拟合；
    // - m_parameterLower/Upper[11]：完整参数体系的搜索上下界；
    // - m_enabledParamIndices：把粒子内部紧凑向量映射回完整参数索引。
    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
    nmDataAutomaticFitting fittingData = dataManager->getAutomaticFittingDataCopy();

    // 获取参数选择状态
    m_parameterSelected.resize(11);
    m_parameterSelected[0] = fittingData.getPermeabilitySelected();
    m_parameterSelected[1] = fittingData.getSkinSelected();
    m_parameterSelected[2] = fittingData.getWellboreStorageSelected();
    m_parameterSelected[3] = fittingData.getPorositySelected();
    m_parameterSelected[4] = fittingData.getInitialPressureSelected();
    m_parameterSelected[5] = fittingData.getThicknessSelected();
    m_parameterSelected[6] = fittingData.getCtSelected();
    m_parameterSelected[7] = fittingData.getCfSelected();
    m_parameterSelected[8] = fittingData.getSoiSelected();
    m_parameterSelected[9] = fittingData.getSwiSelected();
    m_parameterSelected[10] = fittingData.getSgiSelected();

    // 获取参数边界
    m_parameterLower.resize(11);
    m_parameterUpper.resize(11);

    m_parameterLower[0] = fittingData.getPermeabilityMin().getValue().toDouble();
    m_parameterUpper[0] = fittingData.getPermeabilityMax().getValue().toDouble();

    m_parameterLower[1] = fittingData.getSkinMin().getValue().toDouble();
    m_parameterUpper[1] = fittingData.getSkinMax().getValue().toDouble();

    m_parameterLower[2] = fittingData.getWellboreStorageMin().getValue().toDouble();
    m_parameterUpper[2] = fittingData.getWellboreStorageMax().getValue().toDouble();

    m_parameterLower[3] = fittingData.getPorosityMin().getValue().toDouble();
    m_parameterUpper[3] = fittingData.getPorosityMax().getValue().toDouble();

    m_parameterLower[4] = fittingData.getInitialPressureMin().getValue().toDouble();
    m_parameterUpper[4] = fittingData.getInitialPressureMax().getValue().toDouble();

    m_parameterLower[5] = fittingData.getThicknessMin().getValue().toDouble();
    m_parameterUpper[5] = fittingData.getThicknessMax().getValue().toDouble();

    m_parameterLower[6] = fittingData.getCtMin().getValue().toDouble();
    m_parameterUpper[6] = fittingData.getCtMax().getValue().toDouble();

    m_parameterLower[7] = fittingData.getCfMin().getValue().toDouble();
    m_parameterUpper[7] = fittingData.getCfMax().getValue().toDouble();

    m_parameterLower[8] = fittingData.getSoiMin().getValue().toDouble();
    m_parameterUpper[8] = fittingData.getSoiMax().getValue().toDouble();

    m_parameterLower[9] = fittingData.getSwiMin().getValue().toDouble();
    m_parameterUpper[9] = fittingData.getSwiMax().getValue().toDouble();

    m_parameterLower[10] = fittingData.getSgiMin().getValue().toDouble();
    m_parameterUpper[10] = fittingData.getSgiMax().getValue().toDouble();

    // 更新启用参数索引
    m_enabledParamIndices.clear();

    for(int i = 0; i < m_parameterSelected.size(); ++i) {
        if(m_parameterSelected[i]) {
            m_enabledParamIndices.append(i);
        }
    }

    DEBUG_OUT(QString("Loaded parameter bounds: %1 enabled parameters")
              .arg(m_enabledParamIndices.size()));
}

// ==================== PSO算法核心方法 ====================
bool nmCalculationAutoFitPSO::startAutoFitting()
{
    // 自动拟合的总入口。可以把这个函数当成 PSO 的“运行剧本”：
    // 读取配置 -> 校验输入 -> 评价用户初始解 -> 初始化粒子群 ->
    // 按代循环评价粒子 -> 更新全局最优 -> 判断停止 -> 保存结果。
    if(m_isRunning) {
        m_lastError = "Auto fitting is already running";
        return false;
    }

    // 发送初始化日志
    //emit logMessageGenerated(tr("=== PSO Automatic Fitting Started ==="));
    emit logMessageGenerated(tr("Algorithm: Particle Swarm Optimization"));

    try {
        // 从 DataManager 读取界面保存的自动拟合配置。
        // 本类不直接依赖 UI 控件，便于后续从脚本或其他入口复用。
        if(!loadAllConfigFromDataManager()) {
            emit logMessageGenerated(tr("ERROR: Failed to load configuration from data manager"));
            return false;
        }

        if(m_simulationMode) {
            // 调试/演示用快速路径，不调用真实求解器。正式工况通常不走这里。
            DEBUG_OUT("=== SIMULATION MODE ACTIVATED ===");
            runSimulatedFitting();
            return true;
        }

        int enabledParams = getEnabledParameterCount();
        emit logMessageGenerated(tr("Enabled parameters count: %1").arg(enabledParams));

        if(enabledParams == 0) {
            m_lastError = "No parameters enabled for optimization";
            emit logMessageGenerated(tr("ERROR: No parameters enabled for optimization"));
            return false;
        }

        if(m_targetLogLogData.isEmpty() || m_targetLogLogData.size() < 3) {
            m_lastError = "Target LogLog data is empty or insufficient";
            emit logMessageGenerated(tr("ERROR: Target LogLog data is empty or insufficient"));
            return false;
        }

        // 目标曲线必须至少包含三列：time、pressure、pressure derivative，
        // 且三列长度一致。后续误差函数会按 time 做插值对齐。
        if(m_targetLogLogData[0].size() != m_targetLogLogData[1].size() ||
                m_targetLogLogData[0].size() != m_targetLogLogData[2].size()) {
            m_lastError = "Target LogLog data arrays have inconsistent sizes";
            emit logMessageGenerated(tr("ERROR: Target LogLog data arrays have inconsistent sizes"));
            return false;
        }

        emit logMessageGenerated(tr("Target data validation passed (%1 data points)").arg(m_targetLogLogData[0].size()));

        // 使用保存的初始值进行精英保护。resetOptimizer() 会清空部分运行状态，
        // 所以先把用户当前模型参数缓存下来，后面再恢复用于初始解评价和粒子初始化。
        QVector<double> savedInitialValues = m_initialValues;

        // 重置状态
        resetOptimizer();
        m_isRunning = true;
        m_shouldStop = false;
        m_isPaused = false;
        m_currentIteration = 0;
        m_consecutiveFailures = 0;

        if(kUseFixedPsoSeed) {
            // 固定随机种子便于复现 trace 和代理筛选效果。若做生产随机搜索，
            // 可以关闭 kUseFixedPsoSeed，改用当前时间生成种子。
            m_psoRandomSeed = kFixedPsoSeed;
        } else {
            QTime seedTime = QTime::currentTime();
            m_psoRandomSeed = static_cast<unsigned int>(QDateTime::currentDateTime().toTime_t());
            m_psoRandomSeed ^= static_cast<unsigned int>((seedTime.hour() << 22) ^ (seedTime.minute() << 16) ^ (seedTime.second() << 10) ^ seedTime.msec());
        }

        if(m_psoRandomSeed == 0u) {
            m_psoRandomSeed = 1u;
        }

        qsrand(m_psoRandomSeed);
        m_consecutiveFailedIterations = 0;  // 重置连续失败迭代计数

        // 精英保护：先评价用户当前模型。
        // 如果后续 PSO 没找到更好解，最终结果不会比用户已有模型更差。
        m_initialValues = savedInitialValues;
        initializeTraceFile();

        if(!savedInitialValues.isEmpty()) {
            m_userInitialSolution = savedInitialValues;
            emit logMessageGenerated(tr("=== Evaluating Initial Solution (Elite Protection) ==="));

            // 输出初始参数值，方便在 trace / 日志中确认初始解对应的参数顺序。
            QString paramStr = tr("Initial parameters: ");

            for(int i = 0; i < m_userInitialSolution.size(); ++i) {
                paramStr += QString("[%1]=%2 ").arg(i).arg(m_userInitialSolution[i], 0, 'f', 6);
            }

            emit logMessageGenerated(paramStr);

            try {
                emit logMessageGenerated(tr("Starting initial solution evaluation..."));
                DEBUG_OUT(QString("Before evaluateError: m_userInitialSolution size = %1").arg(m_userInitialSolution.size()));

                if(m_userInitialSolution.isEmpty()) {
                    emit logMessageGenerated(tr("ERROR: m_userInitialSolution is empty!"));
                    return false;
                }

                QTime initialEvalTimer;
                initialEvalTimer.start();
                m_userInitialFitness = evaluateFitness(m_userInitialSolution);
                int initialEvalElapsedMs = initialEvalTimer.elapsed();

                if(m_userInitialFitness < 1e9) {
                    m_hasValidUserSolution = true;
                    m_globalBestFitness = m_userInitialFitness;
                    m_globalBestPosition = m_userInitialSolution;
                    m_userInitialLogLogData = m_lastEvaluatedLogLogData;
                    m_globalBestLogLogData = m_userInitialLogLogData;

                    emit logMessageGenerated(tr("Initial solution evaluation successful"));
                    emit logMessageGenerated(tr("Initial Error: %1").arg(m_userInitialFitness, 0, 'e', 4));
                    emit logMessageGenerated(tr("Elite protection activated - initial solution will be preserved if no significant improvement found"));
                    emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, 0, m_globalBestFitness);
                } else {
                    m_hasValidUserSolution = false;
                    emit logMessageGenerated(tr("Initial solution evaluation failed - starting with random initialization"));
                }

                writeTraceRow(-1,
                              -1,
                              "initial_solution",
                              m_userInitialSolution,
                              m_userInitialFitness,
                              m_userInitialFitness < 1e9,
                              initialEvalElapsedMs,
                              std::numeric_limits<double>::quiet_NaN(),
                              "initial_solution",
                              m_hasValidUserSolution ? m_userInitialSolution : QVector<double>(),
                              m_userInitialFitness);
            } catch(...) {
                m_hasValidUserSolution = false;
                emit logMessageGenerated(tr("Exception during initial solution evaluation"));
            }

            // 恢复初始值供粒子群初始化使用
            m_initialValues = savedInitialValues;
        }

        // 初始化粒子群。粒子维度等于用户勾选的参数数量，而不是固定 11 维。
        if(kUseFixedPsoSeed) {
            emit logMessageGenerated(tr("PSO random seed: %1 ").arg(m_psoRandomSeed));
        } else {
            emit logMessageGenerated(tr("PSO random seed: %1 ").arg(m_psoRandomSeed));
        }

        initializeSwarm();
        emit logMessageGenerated(tr("Swarm initialized: %1 particles, %2 dimensions").arg(m_swarmSize).arg(getEnabledParameterCount()));

        // PSO主循环：每一代先决定哪些粒子需要真实评价，再更新全局最优和粒子位置。
        emit logMessageGenerated(tr("=== Starting PSO Main Loop ==="));

        for(m_currentIteration = 0; m_currentIteration < m_maxIterations && !m_shouldStop; ++m_currentIteration) {
            // 每次迭代都输出标题，或者只在重要迭代输出详细信息
            bool shouldOutputDetail = (m_currentIteration % qMax(1, m_maxIterations / 10) == 0) ||
                                      (m_currentIteration < 5) ||
                                      (m_currentIteration >= m_maxIterations - 2); // 总是输出前5次和最后2次的详细信息

            // 每次迭代都输出标题
            emit logMessageGenerated(tr("--- Iteration %1/%2 ---").arg(m_currentIteration + 1).arg(m_maxIterations));

            // 只在特定迭代输出详细统计信息
            if(shouldOutputDetail) {
                emit logMessageGenerated(tr("Current best error: %1").arg(m_globalBestFitness, 0, 'e', 4));
                emit logMessageGenerated(tr("Total evaluations: %1 (successful: %2, failures: %3)")
                                         .arg(m_totalEvaluations).arg(m_successfulEvaluations).arg(m_totalEvaluations - m_successfulEvaluations));
            }

            // 检查暂停状态
            while(m_isPaused && !m_shouldStop) {
                QApplication::processEvents();
                msleep(100);
            }

            if(m_shouldStop) {
                emit logMessageGenerated(tr("Optimization stopped by user request"));
                break;
            }

            int currentIterationSuccessful = 0;
            int currentIterationTotal = 0;
            int currentIterationFailed = 0;
            int currentIterationSkipped = 0;
            QVector<bool> evaluateMask = buildSurrogateEvaluationMask();

            for(int i = 0; i < m_swarmSize && !m_shouldStop; ++i) {
                try {
                    if(i < evaluateMask.size() && !evaluateMask[i]) {
                        // 被代理模型筛掉的粒子本代不跑真实求解器。
                        // 这里不会用 surrogateObjective 更新 pbest，避免代理误差污染真实最优。
                        AutoFitParticle& particle = m_swarm[i];
                        particle.evaluatedThisIteration = false;
                        particle.lastEvaluationSuccess = false;
                        particle.lastEvaluationElapsedMs = -1;
                        particle.selectedForSolver = false;

                        if(particle.screeningDecision.isEmpty()) {
                            particle.screeningDecision = "screened_out";
                        }

                        currentIterationSkipped++;
                        continue;
                    }

                    double previousFitness = m_swarm[i].fitness;
                    // 对单个粒子执行真实评价：写参数 -> 跑求解器 -> 算曲线误差 -> 更新 pbest。
                    updateParticle(i);
                    currentIterationTotal++;

                    if(m_swarm[i].fitness < 1e9) {
                        currentIterationSuccessful++;

                        if(shouldOutputDetail && m_swarm[i].fitness < previousFitness) {
                            // 计算改进幅度
                            double improvement = (previousFitness - m_swarm[i].fitness) / qMax(1e-10, previousFitness);

                            if(improvement > 0.01) { // 1%以上改进才输出
                                emit logMessageGenerated(tr("  Particle %1 improved: %2 -> %3 (%4% improvement)")
                                                         .arg(i + 1).arg(previousFitness, 0, 'e', 3).arg(m_swarm[i].fitness, 0, 'e', 3)
                                                         .arg(improvement * 100, 0, 'f', 2));
                            }
                        }
                    } else {
                        currentIterationFailed++;

                        if(shouldOutputDetail) {
                            emit logMessageGenerated(tr("  Particle %1: evaluation failed").arg(i + 1));
                        }
                    }

                    // 每评估2个粒子处理一次事件
                    if(i % 2 == 0) {
                        QApplication::processEvents();
                    }

                } catch(const std::exception& e) {
                    if(shouldOutputDetail) {
                        emit logMessageGenerated(tr("  Particle %1: Exception: %2").arg(i + 1).arg(e.what()));
                    }

                    m_swarm[i].fitness = 1e10;
                    currentIterationTotal++;
                    currentIterationFailed++;
                    m_totalEvaluations++;
                } catch(...) {
                    if(shouldOutputDetail) {
                        emit logMessageGenerated(tr("  Particle %1: Unknown exception").arg(i + 1));
                    }

                    m_swarm[i].fitness = 1e10;
                    currentIterationTotal++;
                    currentIterationFailed++;
                    m_totalEvaluations++;
                }
            }

            if(m_shouldStop) break;

            double currentSuccessRate = currentIterationTotal > 0 ?
                                        (double)currentIterationSuccessful / currentIterationTotal : 0.0;

            // 更新连续失败迭代计数
            if(currentIterationSuccessful == 0) {
                m_consecutiveFailedIterations++;
                emit logMessageGenerated(tr("WARNING: No successful evaluations in iteration %1 (consecutive failures: %2)")
                                         .arg(m_currentIteration + 1).arg(m_consecutiveFailedIterations));
            } else {
                m_consecutiveFailedIterations = 0; // 重置连续失败计数
            }

            // 输出当前迭代统计
            if(shouldOutputDetail) {
                emit logMessageGenerated(tr("Current iteration stats: %1 successful, %2 failed, %3 skipped out of %4 particles (solver success rate: %5%)")
                                         .arg(currentIterationSuccessful).arg(currentIterationFailed)
                                         .arg(currentIterationSkipped)
                                         .arg(m_swarmSize).arg(currentSuccessRate * 100, 0, 'f', 1));
            }

            // 检查是否需要停止优化
            if(m_consecutiveFailedIterations >= m_maxConsecutiveFailures) {
                m_lastError = QString("Too many consecutive failed iterations (%1)").arg(m_consecutiveFailedIterations);
                emit logMessageGenerated(tr("ERROR: Too many consecutive failed iterations (%1/%2) - stopping optimization")
                                         .arg(m_consecutiveFailedIterations).arg(m_maxConsecutiveFailures));
                break;
            }

            // 警告低成功率但不立即停止
            if(currentSuccessRate < 0.5 && m_currentIteration > 3) {
                emit logMessageGenerated(tr("WARNING: Low success rate (%1%) in iteration %2, but continuing optimization")
                                         .arg(currentSuccessRate * 100, 0, 'f', 1).arg(m_currentIteration + 1));
            }

            // 更新全局最优。updateGlobalBest() 只读取粒子的真实 bestFitness，
            // 不使用代理模型的 surrogateObjective。
            //double previousGlobalBest = m_globalBestFitness;
            updateGlobalBest();
            // 记录本代所有粒子的真实/代理误差和筛选决策，用于复盘和排障。
            writeIterationTraceRows();

            // 自适应参数更新
            adaptiveParameterUpdate(m_currentIteration);

            emit progressUpdated(m_currentIteration + 1, m_globalBestFitness);

            // 记录收敛历史
            m_convergenceHistory.append(m_globalBestFitness);

            // 每5次迭代输出收敛状态
            if((m_currentIteration + 1) % 5 == 0) {
                double recentImprovement = 0.0;

                if(m_convergenceHistory.size() >= 5) {
                    double oldBest = m_convergenceHistory[m_convergenceHistory.size() - 5];
                    recentImprovement = (oldBest - m_globalBestFitness) / qMax(1e-10, oldBest);
                }

                emit logMessageGenerated(tr("  Convergence check: recent improvement = %1% over last 5 iterations")
                                         .arg(recentImprovement * 100, 0, 'f', 4));
            }

            // 更新收敛指标
            updateConvergenceMetrics();

            // 智能收敛判断：目标达成、稳定收敛、局部最优、连续失败等都会在这里决策。
            StopReasonPSO stopReason = analyzeOptimizationStatus();

            if(stopReason == PSO_TARGET_ACHIEVED) {
                emit logMessageGenerated(tr("=== TARGET ACHIEVED ==="));
                emit logMessageGenerated(tr("Target error achieved! Current error: %1 < Target: %2")
                                         .arg(m_globalBestFitness, 0, 'e', 4).arg(m_targetError, 0, 'e', 4));
                emit logMessageGenerated(tr("Optimization completed successfully after %1 iterations")
                                         .arg(m_currentIteration + 1));
                break;
            } else if(stopReason == PSO_TRUE_CONVERGENCE) {
                emit logMessageGenerated(tr("=== TRUE CONVERGENCE DETECTED ==="));
                emit logMessageGenerated(tr("Algorithm has converged to a stable solution"));
                emit logMessageGenerated(tr("Final error: %1 after %2 iterations")
                                         .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1));
                emit logMessageGenerated(tr("Solution quality: %1 (1.0 = target achieved)")
                                         .arg(m_targetError / qMax(1e-10, m_globalBestFitness), 0, 'f', 3));
                break;
            } else if(stopReason == PSO_LOCAL_OPTIMUM) {
                emit logMessageGenerated(tr("=== LOCAL OPTIMUM DETECTED ==="));
                emit logMessageGenerated(tr("Algorithm appears to be trapped in local optimum"));
                emit logMessageGenerated(tr("Current error: %1 after %2 iterations")
                                         .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1));
                emit logMessageGenerated(tr("Suggestion: Try restarting with different parameters or larger search space"));
                break;
            } else if(stopReason == PSO_CONSECUTIVE_FAILURES) {
                emit logMessageGenerated(tr("=== CONSECUTIVE FAILURES ==="));
                emit logMessageGenerated(tr("Too many consecutive failed iterations (%1/%2)")
                                         .arg(m_consecutiveFailedIterations).arg(m_maxConsecutiveFailures));
                break;
            } else if(stopReason == PSO_CONTINUE_OPTIMIZATION) {
                // 继续优化，每10次迭代输出一次状态
                if(m_currentIteration > 0 && m_currentIteration % 10 == 0) {
                    double diversity = calculateSwarmDiversity();
                    double avgVel = calculateAverageVelocity();
                    double stagnation = calculateParticleStagnationRate();

                    emit logMessageGenerated(tr("  Optimization status: diversity=%1, avgVel=%2, stagnation=%3%")
                                             .arg(diversity, 0, 'f', 4).arg(avgVel, 0, 'f', 4).arg(stagnation * 100, 0, 'f', 1));
                }
            }

            // 更新粒子位置和速度
            updateVelocityAndPosition();

            // 输出迭代结束标记
            if(shouldOutputDetail) {
                emit logMessageGenerated(tr("  Iteration %1 completed - Current best: %2")
                                         .arg(m_currentIteration + 1).arg(m_globalBestFitness, 0, 'e', 4));
            }

            // 强制处理事件，保持界面响应
            QApplication::processEvents();

            // 在迭代间稍作停顿，减少系统负载
            if(m_currentIteration % 3 == 2) {
                msleep(200);
            }
        }

        // 最终结果验证和保护
        validateAndProtectFinalResult();

    } catch(const std::exception& e) {
        m_lastError = QString(tr("Critical exception in PSO main loop: %1")).arg(e.what());
        emit logMessageGenerated(tr("CRITICAL ERROR: %1").arg(e.what()));
        closeTraceFile();
        cleanupTemporaryDirectory();
        m_isRunning = false;
        emit fittingFinished(false, m_lastError);
        return false;
    } catch(...) {
        m_lastError = QString(tr("Unknown critical exception in PSO main loop"));
        emit logMessageGenerated(tr("CRITICAL ERROR: Unknown exception in PSO main loop"));
        closeTraceFile();
        cleanupTemporaryDirectory();
        m_isRunning = false;
        emit fittingFinished(false, m_lastError);
        return false;
    }

    m_isRunning = false;

    // 应用最终参数
    if(!m_globalBestPosition.isEmpty()) {
        try {
            emit logMessageGenerated(tr("Applying optimized parameters to model..."));
            applyParametersToDataManager(m_globalBestPosition);
            saveOptimizationResult();

            // 输出最终优化结果
            emit logMessageGenerated(tr("=== Optimization Results ==="));
            emit logMessageGenerated(tr("Final error: %1").arg(m_globalBestFitness, 0, 'e', 4));
            emit logMessageGenerated(tr("Total iterations: %1").arg(m_currentIteration + 1));  // 显示实际完成的迭代数
            emit logMessageGenerated(tr("Total evaluations: %1 (successful: %2)")
                                     .arg(m_totalEvaluations).arg(m_successfulEvaluations));

            // 输出最优参数值
            QString finalParams = tr("Optimized parameters: ");

            for(int i = 0; i < m_globalBestPosition.size(); ++i) {
                finalParams += QString("[%1]=%2 ").arg(i).arg(m_globalBestPosition[i], 0, 'f', 6);
            }

            emit logMessageGenerated(finalParams);

            emit logMessageGenerated(tr("Parameters applied successfully to data manager"));
        } catch(const std::exception& e) {
            emit logMessageGenerated(tr("ERROR: Failed to apply final parameters: %1").arg(e.what()));
            m_lastError = QString("Failed to apply final parameters: %1").arg(e.what());
        } catch(...) {
            emit logMessageGenerated(tr("ERROR: Unknown error applying final parameters"));
            m_lastError = "Failed to apply final parameters due to unknown error";
        }
    }

    // 判断系统确定最终结果
    bool success;
    QString message;
    StopReasonPSO finalReason = analyzeOptimizationStatus();

    if(finalReason == PSO_TARGET_ACHIEVED) {
        success = true;
        message = QString(tr("Target achieved. Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION SUCCESSFUL ==="));
    } else if(finalReason == PSO_TRUE_CONVERGENCE) {
        success = true;
        message = QString(tr("PSO optimization converged to stable solution. Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION CONVERGED ==="));
    } else if(finalReason == PSO_LOCAL_OPTIMUM) {
        success = true;
        message = QString(tr("PSO optimization trapped in local optimum. Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION - LOCAL OPTIMUM ==="));
    } else if(finalReason == PSO_MAX_ITERATIONS) {
        success = true;
        message = QString(tr("Max iterations reached. Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION - MAX ITERATIONS ==="));
    } else if(finalReason == PSO_USER_STOPPED) {
        success = true;
        message = QString(tr("Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION STOPPED BY USER ==="));
    } else if(finalReason == PSO_CONSECUTIVE_FAILURES) {
        success = false;
        message = QString(tr("PSO optimization failed due to consecutive failures. Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION FAILED ==="));
    } else {
        success = false;
        message = QString(tr("PSO optimization ended unexpectedly. Best error: %1, Iterations: %2"))
                  .arg(m_globalBestFitness, 0, 'e', 4).arg(m_currentIteration + 1);
        emit logMessageGenerated(tr("=== PSO OPTIMIZATION - UNKNOWN END ==="));
    }

    emitRunSummary(success, finalReason);

    // 先发送最终进度更新，确保进度条达到100%
    emit progressUpdated(m_maxIterations, m_globalBestFitness);

    // 确保界面更新完成
    QApplication::processEvents();

    // 短暂延迟，让进度条动画完成
    msleep(200);

    // 再次确保界面更新
    QApplication::processEvents();

    // 现在才发送完成信号，这会触发消息框
    closeTraceFile();
    emit fittingFinished(success, message);

    cleanupTemporaryDirectory();

    return success;

}

void nmCalculationAutoFitPSO::extractUserInitialValues()
{
    // 从当前项目模型读取用户已有初始参数。
    // 只提取用户勾选的参数，并按 m_enabledParamIndices 的顺序写入 m_initialValues。
    // 这些值用于：
    // 1. 初始解真实评价；
    // 2. 粒子群引导初始化；
    // 3. 最终精英保护。
    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
    nmDataReservoir reservoirData = dataManager->getReservoirDataCopy();
    //QVector<nmDataWellBase*> wells = dataManager->getWellDataList();

    nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);

    m_initialValues.clear();

    // 按照启用参数的顺序提取初始值。井参数来自目标井，储层参数来自 reservoirData。
    for(int i = 0; i < m_enabledParamIndices.size(); ++i) {
        int paramIndex = m_enabledParamIndices[i];
        double initialValue = 0.0;

        switch(paramIndex) {
            case 0: // 渗透率
                initialValue = reservoirData.getPermeability().getValue().toDouble();
                break;

            case 1: // 表皮系数
                if(pTargetWell) {
                    initialValue = pTargetWell->getPerforation(0)->getSkin().getValue().toDouble();
                }

                break;

            case 2: // 井筒储集系数
                if(pTargetWell) {
                    initialValue = pTargetWell->getWellboreStorage().getValue().toDouble();
                }

                break;

            case 3: // 孔隙度
                initialValue = reservoirData.getPorosity().getValue().toDouble();
                break;

            case 4: // 初始压力
                initialValue = reservoirData.getInitialPressure().getValue().toDouble();
                break;

            case 5: // 储层厚度
                initialValue = reservoirData.getThickness().getValue().toDouble();
                break;

            case 6: // 综合压缩系数
                initialValue = reservoirData.getCt().getValue().toDouble();
                break;

            case 7: // 岩石压缩系数
                initialValue = reservoirData.getCf().getValue().toDouble();
                break;

            case 8: // 初始含油饱和度
                initialValue = reservoirData.getSoi().getValue().toDouble();
                break;

            case 9: // 初始含水饱和度
                initialValue = reservoirData.getSwi().getValue().toDouble();
                break;

            case 10: // 初始含气饱和度
                initialValue = reservoirData.getSgi().getValue().toDouble();
                break;
        }

        m_initialValues.append(initialValue);
    }

    DEBUG_OUT(QString("Extracted %1 user initial values").arg(m_initialValues.size()));

    for(int i = 0; i < m_initialValues.size(); ++i) {
        DEBUG_OUT(QString("  Initial[%1] = %2").arg(i).arg(m_initialValues[i], 0, 'e', 3));
    }

    // 验证初始值
    if(!validateInitialValues()) {
        DEBUG_OUT("Warning: Some initial values are outside parameter bounds");
    }
}

void nmCalculationAutoFitPSO::initializeSwarm()
{
    // 初始化粒子群。
    // 若用户已有初始模型有效，则一部分粒子围绕初始值做局部搜索；
    // 另一部分粒子在全局上下界内随机分布，保留跳出局部区域的能力。
    int dimensions = getEnabledParameterCount();

    if(dimensions == 0) return;

    m_swarm.resize(m_swarmSize);
    m_globalBestPosition.resize(dimensions);

    bool hasValidInitials = !m_initialValues.isEmpty() && m_initialValues.size() >= dimensions;

    // 引导搜索比例。guided 粒子围绕用户初始值，非 guided 粒子做全局随机搜索。
    //int guidedCount = hasValidInitials ? qMax(2, (m_swarmSize * 3) / 4) : qMax(1, m_swarmSize / 3);
    int guidedCount = hasValidInitials ? qMax(2, m_swarmSize / 2) : qMax(1, m_swarmSize / 3);

    DEBUG_OUT(QString("Enhanced swarm initialization: %1 particles, %2 guided, %3 random")
              .arg(m_swarmSize).arg(guidedCount).arg(m_swarmSize - guidedCount));

    for(int i = 0; i < m_swarmSize; ++i) {
        AutoFitParticle& particle = m_swarm[i];
        particle.position.resize(dimensions);
        particle.velocity.resize(dimensions);
        particle.bestPosition.resize(dimensions);
        particle.bestFitness = 1e10;
        particle.fitness = 1e10;
        particle.evaluatedThisIteration = false;
        particle.lastEvaluationSuccess = false;
        particle.lastEvaluationElapsedMs = -1;
        particle.surrogateObjective = 1e10;
        particle.selectedForSolver = true;
        particle.selectedByAudit = false;
        particle.screeningDecision = "full_solver";

        for(int j = 0; j < dimensions; ++j) {
            int paramIndex = m_enabledParamIndices[j];
            double range = m_parameterUpper[paramIndex] - m_parameterLower[paramIndex];

            if(i == 0 && hasValidInitials) {
                // 第一个粒子：完全使用用户初始值，确保初始模型本身参与粒子群。
                particle.position[j] = m_initialValues[j];

            } else if(i < guidedCount && hasValidInitials) {
                // 引导粒子采用分层搜索半径：
                // 近邻精细搜索 + 中等扰动 + 较大扰动，覆盖初始模型周围不同尺度。
                double searchRadius;

                if(i <= guidedCount / 3) {
                    searchRadius = range * 0.03;  // 3%范围内精细搜索
                } else if(i <= guidedCount * 2 / 3) {
                    searchRadius = range * 0.08;  // 8%范围内搜索
                } else {
                    searchRadius = range * 0.15;  // 15%范围内搜索
                }

                double offset = (random01() - 0.5) * searchRadius;
                particle.position[j] = m_initialValues[j] + offset;

            } else {
                // 少数粒子进行全局搜索，防止所有粒子都被初始模型附近的局部最优限制。
                particle.position[j] = m_parameterLower[paramIndex] + random01() * range;
            }
        }

        // 边界检查
        clampToLimits(particle.position);

        // 速度初始化。引导粒子速度更小，避免刚开始就离开初始模型邻域；
        // 全局粒子速度稍大，便于探索更宽的搜索空间。
        for(int j = 0; j < dimensions; ++j) {
            int paramIndex = m_enabledParamIndices[j];
            double range = m_parameterUpper[paramIndex] - m_parameterLower[paramIndex];

            double velocityFactor;

            if(i < guidedCount) {
                velocityFactor = VELOCITY_LIMIT_FACTOR * 0.15;  // 更小的速度
            } else {
                velocityFactor = VELOCITY_LIMIT_FACTOR * 0.8;   // 适中速度
            }

            particle.velocity[j] = (random01() - 0.5) * range * velocityFactor;
        }

        particle.bestPosition = particle.position;
    }
}

void nmCalculationAutoFitPSO::updateParticle(int particleIndex)
{
    if(particleIndex < 0 || particleIndex >= m_swarm.size()) return;

    AutoFitParticle& particle = m_swarm[particleIndex];

    // 单粒子真实评价入口。
    // 这里调用 evaluateFitness()，因此会真实写 DataManager、调用求解器、计算误差。
    // 被代理模型筛掉的粒子不会进入这个函数。
    particle.evaluatedThisIteration = false;
    particle.lastEvaluationSuccess = false;
    particle.lastEvaluationElapsedMs = -1;
    particle.selectedForSolver = true;

    if(particle.screeningDecision.isEmpty()) {
        particle.screeningDecision = "full_solver";
    }

    QTime evalTimer;
    evalTimer.start();
    particle.fitness = evaluateFitness(particle.position);
    particle.currentLogLogData = m_lastEvaluatedLogLogData;
    particle.lastEvaluationElapsedMs = evalTimer.elapsed();
    particle.evaluatedThisIteration = true;
    particle.lastEvaluationSuccess = (particle.fitness < 1e9);
    m_totalEvaluations++;

    if(particle.fitness < 1e9) {
        m_successfulEvaluations++;
    }

    // 更新个体最优 pbest。这里使用的是真实求解器误差 particle.fitness，
    // 不是代理模型给出的 surrogateObjective。
    if(particle.fitness < particle.bestFitness) {
        particle.bestFitness = particle.fitness;
        particle.bestPosition = particle.position;
        particle.bestLogLogData = particle.currentLogLogData;
        DEBUG_OUT(QString("Particle %1 improved: error = %2")
                  .arg(particleIndex).arg(particle.fitness, 0, 'e', 4));
    }
}

void nmCalculationAutoFitPSO::updateGlobalBest()
{
    // 保存上一轮的全局最优，用于后续自适应参数调整等
    m_previousBestFitness = m_globalBestFitness;
    bool globalBestUpdated = false;

    // 遍历所有粒子，寻找比当前 global best 更好的个体最优。
    // 注意：particle.bestFitness 只有在真实求解器评价成功后才会更新。
    for(int i = 0; i < m_swarm.size(); ++i) {
        const AutoFitParticle& particle = m_swarm[i];

        // 只要个体最优比当前全局最优小，就认为是更好的解
        if(particle.bestFitness < m_globalBestFitness) {

            double improvement = m_globalBestFitness - particle.bestFitness;
            double relativeImprovement =
                improvement / qMax(1e-10, qAbs(m_globalBestFitness));

            // 区分显著改进和微小改进，但无论如何都会更新全局最优
            if(relativeImprovement > m_improvementThreshold) {
                DEBUG_OUT(QString(tr("Global best updated with %1% improvement: %2"))
                          .arg(relativeImprovement * 100.0, 0, 'f', 3)
                          .arg(particle.bestFitness, 0, 'e', 4));
            } else {
                DEBUG_OUT(QString(tr("Global best updated (minor improvement %1% < %2%) to %3"))
                          .arg(relativeImprovement * 100.0, 0, 'f', 3)
                          .arg(m_improvementThreshold * 100.0, 0, 'f', 2)
                          .arg(particle.bestFitness, 0, 'e', 4));
            }

            // 无论相对改进是否超过阈值，都要更新全局最优
            m_globalBestFitness  = particle.bestFitness;
            m_globalBestPosition = particle.bestPosition;
            m_globalBestLogLogData = particle.bestLogLogData;
            globalBestUpdated    = true;
            emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, m_currentIteration + 1, m_globalBestFitness);
        }
    }

    // 只有在本轮没有找到任何更好的解时才考虑恢复用户初始解
    if(!globalBestUpdated && m_hasValidUserSolution && m_userInitialFitness < m_globalBestFitness) {

        double improvement = m_globalBestFitness - m_userInitialFitness;
        double relativeImprovement =
            improvement / qMax(1e-10, qAbs(m_globalBestFitness));

        DEBUG_OUT("Elite protection: Restoring user initial solution as global best");
        DEBUG_OUT(QString("Elite solution is better than current best by %1%")
                  .arg(relativeImprovement * 100.0, 0, 'f', 3));

        m_globalBestFitness  = m_userInitialFitness;
        m_globalBestPosition = m_userInitialSolution;
        m_globalBestLogLogData = m_userInitialLogLogData;
        emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, m_currentIteration + 1, m_globalBestFitness);
    }
}

//void nmCalculationAutoFitPSO::updateGlobalBest()
//{
//    m_previousBestFitness = m_globalBestFitness;
//    bool globalBestUpdated = false;
//
//    for(int i = 0; i < m_swarm.size(); ++i) {
//        const AutoFitParticle& particle = m_swarm[i];
//
//        if(particle.bestFitness < m_globalBestFitness) {
//            // 只有显著改进时才更新
//            double improvement = (m_globalBestFitness - particle.bestFitness);
//            double relativeImprovement = improvement / qMax(1e-10, qAbs(m_globalBestFitness));
//
//            if(relativeImprovement > m_improvementThreshold) {
//                m_globalBestFitness = particle.bestFitness;
//                m_globalBestPosition = particle.bestPosition;
//                globalBestUpdated = true;
//
//                DEBUG_OUT(QString("Global best updated with %1% improvement: %2")
//                          .arg(relativeImprovement * 100, 0, 'f', 3)
//                          .arg(m_globalBestFitness, 0, 'e', 4));
//            } else {
//                DEBUG_OUT(QString("Rejecting minor improvement: %1% (threshold: %2%)")
//                          .arg(relativeImprovement * 100, 0, 'f', 4)
//                          .arg(m_improvementThreshold * 100, 0, 'f', 2));
//            }
//        }
//    }
//
//    // 在没有找到更好解时才检查
//    if(!globalBestUpdated && m_hasValidUserSolution && m_userInitialFitness < m_globalBestFitness) {
//        DEBUG_OUT("Elite protection: No significant improvement found, checking initial solution");
//
//        // 检查初始解是否仍然是最优的
//        double improvement = m_globalBestFitness - m_userInitialFitness;
//        double relativeImprovement = improvement / qMax(1e-10, qAbs(m_globalBestFitness));
//
//        if(relativeImprovement > m_improvementThreshold * 0.5) { // 使用更宽松的阈值
//            DEBUG_OUT("Elite protection: Restoring user initial solution");
//            m_globalBestFitness = m_userInitialFitness;
//            m_globalBestPosition = m_userInitialSolution;
//        }
//    }
//}

void nmCalculationAutoFitPSO::updateVelocityAndPosition()
{
    for(int i = 0; i < m_swarm.size(); ++i) {
        AutoFitParticle& particle = m_swarm[i];

        for(int j = 0; j < particle.position.size(); ++j) {
            // PSO速度更新公式：
            // inertia 保留上一轮搜索方向；
            // cognitive 项让粒子靠近自己的历史最优；
            // social 项让粒子靠近群体全局最优。
            double r1 = random01();
            double r2 = random01();

            particle.velocity[j] = m_inertiaWeight * particle.velocity[j] +
                                   m_cognitiveParam * r1 * (particle.bestPosition[j] - particle.position[j]) +
                                   m_socialParam * r2 * (m_globalBestPosition[j] - particle.position[j]);

            // 速度限制。不同参数量纲差异很大，所以按该参数搜索区间的一定比例限速。
            int paramIndex = m_enabledParamIndices[j];
            double range = m_parameterUpper[paramIndex] - m_parameterLower[paramIndex];
            double maxVel = range * VELOCITY_LIMIT_FACTOR;
            particle.velocity[j] = qMax(-maxVel, qMin(maxVel, particle.velocity[j]));

            // 更新位置
            particle.position[j] += particle.velocity[j];
        }

        // 边界约束。PSO 位置更新后如果越界，直接夹回用户设置的上下界。
        clampToLimits(particle.position);
    }
}

//double nmCalculationAutoFitPSO::evaluateFitness(const QVector<double>& parameters)
//{
//    const QString funcName = QString("evaluateError[%1]").arg(m_currentIteration);
//    static int callCount = 0;
//    callCount++;
//
//    try {
//        DEBUG_OUT(QString("%1: Call #%2 - Starting evaluation with %3 parameters")
//                  .arg(funcName).arg(callCount).arg(parameters.size()));
//
//        // 打印参数值用于对比
//        QString paramStr = "Parameters: ";
//
//        for(int i = 0; i < parameters.size(); ++i) {
//            paramStr += QString("[%1]=%2 ").arg(i).arg(parameters[i], 0, 'f', 6);
//        }
//
//        DEBUG_OUT(QString("%1: %2").arg(funcName).arg(paramStr));
//
//        // 1. 参数有效性检查
//        if(!validateParameters(parameters)) {
//            DEBUG_OUT(QString("%1: Call #%2 - Invalid parameters").arg(funcName).arg(callCount));
//            return 1e10;
//        }
//
//        // 2. 检查数据管理器状态
//        nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
//
//        if(!dataManager) {
//            DEBUG_OUT(QString("%1: Call #%2 - DataManager is null").arg(funcName).arg(callCount));
//            return 1e10;
//        }
//
//        // 3. 应用参数到数据管理器
//        try {
//            DEBUG_OUT(QString("%1: Call #%2 - Applying parameters to DataManager").arg(funcName).arg(callCount));
//            applyParametersToDataManager(parameters);
//            DEBUG_OUT(QString("%1: Call #%2 - Parameters applied successfully").arg(funcName).arg(callCount));
//        } catch(const std::exception& e) {
//            DEBUG_OUT(QString("%1: Call #%2 - Failed to apply parameters: %3").arg(funcName).arg(callCount).arg(e.what()));
//            return 1e10;
//        } catch(...) {
//            DEBUG_OUT(QString("%1: Call #%2 - Unknown error applying parameters").arg(funcName).arg(callCount));
//            return 1e10;
//        }
//
//        // 4. 运行求解器
//        QVector<QVector<double> > solverResult;
//        const int maxRetries = 2;
//        bool solverSuccess = false;
//
//        for(int retry = 0; retry <= maxRetries; ++retry) {
//            if(m_shouldStop) return 1e10;
//
//            try {
//                DEBUG_OUT(QString("%1: Call #%2 - Solver attempt %3/%4")
//                          .arg(funcName).arg(callCount).arg(retry + 1).arg(maxRetries + 1));
//
//                // 在求解器调用前添加短暂延迟，确保状态稳定
//                if(retry > 0) {
//                    DEBUG_OUT(QString("%1: Call #%2 - Retry delay before solver attempt")
//                              .arg(funcName).arg(callCount));
//                    msleep(1000); // 增加延迟时间
//                }
//
//                solverResult = runSolver();
//
//                if(!solverResult.isEmpty() && validateSolverResult(solverResult)) {
//                    DEBUG_OUT(QString("%1: Call #%2 - Solver successful on attempt %3, result size: %4")
//                              .arg(funcName).arg(callCount).arg(retry + 1).arg(solverResult[0].size()));
//                    solverSuccess = true;
//                    break;
//                } else {
//                    DEBUG_OUT(QString("%1: Call #%2 - Solver failed on attempt %3 - empty or invalid result")
//                              .arg(funcName).arg(callCount).arg(retry + 1));
//
//                    if(retry < maxRetries) {
//                        DEBUG_OUT(QString("%1: Call #%2 - Will retry solver").arg(funcName).arg(callCount));
//                    }
//                }
//
//            } catch(const std::exception& e) {
//                DEBUG_OUT(QString("%1: Call #%2 - Solver exception on attempt %3: %4")
//                          .arg(funcName).arg(callCount).arg(retry + 1).arg(e.what()));
//            } catch(...) {
//                DEBUG_OUT(QString("%1: Call #%2 - Unknown solver exception on attempt %3")
//                          .arg(funcName).arg(callCount).arg(retry + 1));
//            }
//        }
//
//        if(!solverSuccess) {
//            DEBUG_OUT(QString("%1: Call #%2 - All solver attempts failed").arg(funcName).arg(callCount));
//            return 1e10;
//        }
//
//        // 5. 获取双对数结果数据
//        QVector<QVector<double> > resultLogLogData;
//
//        try {
//            //wells = dataManager->getWellDataList(); // 重新获取，确保是最新的
//			nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);
//
//			if(pTargetWell) {
//				resultLogLogData = pTargetWell->getResultLogLog();
//
//                if(!validateLogLogData(resultLogLogData)) {
//                    DEBUG_OUT(QString("%1: Call #%2 - Invalid result LogLog data").arg(funcName).arg(callCount));
//                    return 1e10;
//                }
//
//                DEBUG_OUT(QString("%1: Call #%2 - LogLog result data obtained, size: %3")
//                          .arg(funcName).arg(callCount).arg(resultLogLogData[0].size()));
//            } else {
//                DEBUG_OUT(QString("%1: Call #%2 - No wells found after solver").arg(funcName).arg(callCount));
//                return 1e10;
//            }
//        } catch(const std::exception& e) {
//            DEBUG_OUT(QString("%1: Call #%2 - Error getting LogLog result: %3").arg(funcName).arg(callCount).arg(e.what()));
//            return 1e10;
//        } catch(...) {
//            DEBUG_OUT(QString("%1: Call #%2 - Unknown error getting LogLog result").arg(funcName).arg(callCount));
//            return 1e10;
//        }
//
//        // 6. 双对数曲线对齐和误差计算
//        double error;
//
//        try {
//            error = calculateLogLogCurveError(m_targetLogLogData, resultLogLogData);
//
//            if(!isFiniteNumber(error) || error < 0) {
//                DEBUG_OUT(QString("%1: Call #%2 - Invalid error value: %3").arg(funcName).arg(callCount).arg(error));
//                return 1e10;
//            }
//
//            DEBUG_OUT(QString("%1: Call #%2 - Evaluation successful, error = %3")
//                      .arg(funcName).arg(callCount).arg(error, 0, 'e', 6));
//
//        } catch(const std::exception& e) {
//            DEBUG_OUT(QString("%1: Call #%2 - LogLog error calculation failed: %3").arg(funcName).arg(callCount).arg(e.what()));
//            return 1e10;
//        } catch(...) {
//            DEBUG_OUT(QString("%1: Call #%2 - Unknown error in LogLog error calculation").arg(funcName).arg(callCount));
//            return 1e10;
//        }
//
//        return error;
//
//    } catch(const std::exception& e) {
//        DEBUG_OUT(QString("%1: Call #%2 - Top-level exception: %3").arg(funcName).arg(callCount).arg(e.what()));
//        return 1e10;
//    } catch(...) {
//        DEBUG_OUT(QString("%1: Call #%2 - Unknown top-level exception").arg(funcName).arg(callCount));
//        return 1e10;
//    }
//}

double nmCalculationAutoFitPSO::evaluateFitness(const QVector<double>& parameters)
{
    // 粒子适应度评价函数，也是自动拟合最核心的闭环：
    // 1. 校验粒子参数是否在用户设置的上下界和基本物理范围内；
    // 2. 将参数写入 DataManager 的储层/目标井对象；
    // 3. 调用真实数值求解器，生成模拟结果；
    // 4. 从目标井读取模拟后的 result log-log 曲线；
    // 5. 与目标 history log-log 曲线计算误差，误差越小代表拟合越好。
    //
    // 返回 1e10 表示该粒子评价失败或结果不可用。PSO 会把它当成很差的解。
    const QString funcName = QString("evaluateError[%1]").arg(m_currentIteration);
    static int callCount = 0;
    callCount++;
    m_lastEvaluatedLogLogData.clear();

    try {
        DEBUG_OUT(QString("%1: Call #%2 - Starting evaluation with %3 parameters")
                  .arg(funcName).arg(callCount).arg(parameters.size()));

        // 打印参数值
        QString paramStr = "Parameters: ";

        for(int i = 0; i < parameters.size(); ++i) {
            paramStr += QString("[%1]=%2 ").arg(i).arg(parameters[i], 0, 'f', 6);
        }

        DEBUG_OUT(QString("%1: %2").arg(funcName).arg(paramStr));

        // 1. 参数有效性检查。这里先在 PSO 层拦截明显非法的候选，
        // 避免把非有限数、越界值或极端危险值传给求解器。
        if(!validateParameters(parameters)) {
            DEBUG_OUT(QString("%1: Call #%2 - VALIDATION FAILED").arg(funcName).arg(callCount));

            // 详细检查每个参数
            for(int i = 0; i < parameters.size(); ++i) {
                if(!isFiniteNumber(parameters[i])) {
                    DEBUG_OUT(QString("  -> Param[%1] is NOT finite: %2").arg(i).arg(parameters[i]));
                }

                if(i < m_enabledParamIndices.size()) {
                    int paramIndex = m_enabledParamIndices[i];

                    if(paramIndex >= 0 && paramIndex < m_parameterLower.size()) {
                        double lower = m_parameterLower[paramIndex];
                        double upper = m_parameterUpper[paramIndex];

                        if(parameters[i] < lower) {
                            DEBUG_OUT(QString("  -> Param[%1]=%2 < lower bound %3")
                                      .arg(i).arg(parameters[i]).arg(lower));
                        }

                        if(parameters[i] > upper) {
                            DEBUG_OUT(QString("  -> Param[%1]=%2 > upper bound %3")
                                      .arg(i).arg(parameters[i]).arg(upper));
                        }

                        // 检查危险值。这些条件不是严格物理模型定义，
                        // 而是工程保护：避免求解器在明显异常输入下崩溃或返回无意义曲线。
                        switch(paramIndex) {
                            case 0: // 渗透率
                                if(parameters[i] <= 1e-6) {
                                    DEBUG_OUT(QString("  -> REJECTED: Permeability too small: %1").arg(parameters[i]));
                                }

                                break;

                            case 2: // 井筒储集系数
                                if(parameters[i] <= 1e-8) {
                                    DEBUG_OUT(QString("  -> REJECTED: Wellbore storage too small: %1").arg(parameters[i]));
                                }

                                break;

                            case 3: // 孔隙度
                                if(parameters[i] <= 1e-4 || parameters[i] >= 0.95) {
                                    DEBUG_OUT(QString("  -> REJECTED: Unrealistic porosity: %1").arg(parameters[i]));
                                }

                                break;
                        }
                    }
                }
            }

            return 1e10;
        }

        DEBUG_OUT(QString("%1: Call #%2 - Parameters validated OK").arg(funcName).arg(callCount));

        // 2. 数据管理器检查。后续参数写回和求解器组装都依赖当前 DataManager。
        nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();

        if(!dataManager) {
            DEBUG_OUT(QString("%1: Call #%2 - DataManager is NULL").arg(funcName).arg(callCount));
            return 1e10;
        }

        DEBUG_OUT(QString("%1: Call #%2 - DataManager OK").arg(funcName).arg(callCount));

        // 3. 应用参数。parameters 的顺序与 m_enabledParamIndices 对齐，
        // applyParametersToDataManager() 会把它们拆分写入储层参数和目标井参数。
        try {
            DEBUG_OUT(QString("%1: Call #%2 - Applying parameters...").arg(funcName).arg(callCount));
            applyParametersToDataManager(parameters);
            DEBUG_OUT(QString("%1: Call #%2 - Parameters applied successfully").arg(funcName).arg(callCount));
        } catch(const std::exception& e) {
            DEBUG_OUT(QString("%1: Call #%2 - FAILED to apply parameters: %3")
                      .arg(funcName).arg(callCount).arg(e.what()));
            return 1e10;
        }

        // 4. 运行求解器。真实求解器偶发失败时允许重试，避免一次 DLL 调用异常
        // 直接让整个粒子评价失败。
        QVector<QVector<double>> solverResult;
        const int maxRetries = 2;
        bool solverSuccess = false;

        for(int retry = 0; retry <= maxRetries; ++retry) {
            if(m_shouldStop) {
                DEBUG_OUT(QString("%1: Call #%2 - User stop requested").arg(funcName).arg(callCount));
                return 1e10;
            }

            try {
                DEBUG_OUT(QString("%1: Call #%2 - Solver attempt %3/%4")
                          .arg(funcName).arg(callCount).arg(retry + 1).arg(maxRetries + 1));

                if(retry > 0) {
                    DEBUG_OUT(QString("%1: Call #%2 - Retry delay...").arg(funcName).arg(callCount));
                    msleep(1000);
                }

                solverResult = runSolver();

                // 详细检查求解器结果
                if(solverResult.isEmpty()) {
                    DEBUG_OUT(QString("%1: Call #%2 - Solver returned EMPTY result").arg(funcName).arg(callCount));
                } else {
                    DEBUG_OUT(QString("%1: Call #%2 - Solver returned %3 arrays")
                              .arg(funcName).arg(callCount).arg(solverResult.size()));

                    for(int i = 0; i < solverResult.size(); ++i) {
                        DEBUG_OUT(QString("  -> Array[%1] size: %2").arg(i).arg(solverResult[i].size()));
                    }

                    if(validateSolverResult(solverResult)) {
                        DEBUG_OUT(QString("%1: Call #%2 - Solver result VALIDATED on attempt %3")
                                  .arg(funcName).arg(callCount).arg(retry + 1));
                        solverSuccess = true;
                        break;
                    } else {
                        DEBUG_OUT(QString("%1: Call #%2 - Solver result VALIDATION FAILED on attempt %3")
                                  .arg(funcName).arg(callCount).arg(retry + 1));
                    }
                }

            } catch(const std::exception& e) {
                DEBUG_OUT(QString("%1: Call #%2 - Solver EXCEPTION on attempt %3: %4")
                          .arg(funcName).arg(callCount).arg(retry + 1).arg(e.what()));
            }
        }

        if(!solverSuccess) {
            DEBUG_OUT(QString("%1: Call #%2 - ALL SOLVER ATTEMPTS FAILED").arg(funcName).arg(callCount));
            return 1e10;
        }

        // 5. 获取 LogLog 数据。runSolver() 会更新 DataManager 中目标井的计算结果，
        // 这里再从目标井读取 resultLogLogData 作为模拟曲线。
        QVector<QVector<double>> resultLogLogData;

        try {
            nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);

            if(!pTargetWell) {
                DEBUG_OUT(QString("%1: Call #%2 - Target well '%3' NOT FOUND")
                          .arg(funcName).arg(callCount).arg(m_targetWellName));
                return 1e10;
            }

            DEBUG_OUT(QString("%1: Call #%2 - Target well found: %3")
                      .arg(funcName).arg(callCount).arg(m_targetWellName));

            resultLogLogData = pTargetWell->getResultLogLog();

            if(!validateLogLogData(resultLogLogData)) {
                DEBUG_OUT(QString("%1: Call #%2 - LogLog data VALIDATION FAILED")
                          .arg(funcName).arg(callCount));

                // 详细输出LogLog数据问题
                if(resultLogLogData.size() < 3) {
                    DEBUG_OUT(QString("  -> LogLog arrays count: %1 (need 3)")
                              .arg(resultLogLogData.size()));
                } else {
                    DEBUG_OUT(QString("  -> LogLog array sizes: X=%1, Y1=%2, Y2=%3")
                              .arg(resultLogLogData[0].size())
                              .arg(resultLogLogData[1].size())
                              .arg(resultLogLogData[2].size()));

                    // 检查数据有效性
                    for(int i = 0; i < qMin(5, resultLogLogData[0].size()); ++i) {
                        if(!isFiniteNumber(resultLogLogData[0][i]) ||
                                !isFiniteNumber(resultLogLogData[1][i]) ||
                                !isFiniteNumber(resultLogLogData[2][i])) {
                            DEBUG_OUT(QString("  -> Invalid data at index %1: X=%2, Y1=%3, Y2=%4")
                                      .arg(i)
                                      .arg(resultLogLogData[0][i])
                                      .arg(resultLogLogData[1][i])
                                      .arg(resultLogLogData[2][i]));
                        }
                    }
                }

                return 1e10;
            }

            DEBUG_OUT(QString("%1: Call #%2 - LogLog data validated, size: %3")
                      .arg(funcName).arg(callCount).arg(resultLogLogData[0].size()));

        } catch(const std::exception& e) {
            DEBUG_OUT(QString("%1: Call #%2 - Error getting LogLog data: %3")
                      .arg(funcName).arg(callCount).arg(e.what()));
            return 1e10;
        }

        // 6. 计算误差。这里比较的是目标井 history log-log 与当前模拟 result log-log。
        double error;

        try {
            DEBUG_OUT(QString("%1: Call #%2 - Calculating error...")
                      .arg(funcName).arg(callCount));

            error = calculateLogLogCurveError(m_targetLogLogData, resultLogLogData);

            if(!isFiniteNumber(error) || error < 0) {
                DEBUG_OUT(QString("%1: Call #%2 - INVALID error value: %3")
                          .arg(funcName).arg(callCount).arg(error));
                return 1e10;
            }

            DEBUG_OUT(QString("%1: Call #%2 - SUCCESS! Error = %3")
                      .arg(funcName).arg(callCount).arg(error, 0, 'e', 6));
            // 保存最后一次有效曲线，供 updateParticle() 写入粒子 currentLogLogData。
            m_lastEvaluatedLogLogData = resultLogLogData;

        } catch(const std::exception& e) {
            DEBUG_OUT(QString("%1: Call #%2 - Error calculation FAILED: %3")
                      .arg(funcName).arg(callCount).arg(e.what()));
            return 1e10;
        }

        return error;

    } catch(const std::exception& e) {
        DEBUG_OUT(QString("%1: Call #%2 - TOP-LEVEL EXCEPTION: %3")
                  .arg(funcName).arg(callCount).arg(e.what()));
        return 1e10;
    } catch(...) {
        DEBUG_OUT(QString("%1: Call #%2 - UNKNOWN TOP-LEVEL EXCEPTION")
                  .arg(funcName).arg(callCount));
        return 1e10;
    }
}

// ==================== 参数应用方法 ====================

void nmCalculationAutoFitPSO::applyParametersToDataManager(const QVector<double>& parameters)
{
    // 将粒子的“启用参数向量”写回项目数据。
    // parameters 的维度必须等于用户勾选的参数数量，顺序由 m_enabledParamIndices 决定。
    // 这里不直接跑求解器，只负责把 DataManager 调整到该粒子对应的模型状态。
    if(parameters.size() != getEnabledParameterCount()) {
        return;
    }

    updateReservoirParameters(parameters);
    updateWellParameters(parameters);
}

void nmCalculationAutoFitPSO::updateReservoirParameters(const QVector<double>& parameters)
{
    // 更新储层级参数。井级参数 skin/wellboreC 不在这里改，由 updateWellParameters() 负责。
    // 这里先取 DataManager 中 reservoir 的副本，修改后再整体写回 DataManager。
    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
    nmDataReservoir reservoirData = dataManager->getReservoirDataCopy();

    // paramIndex 是粒子 position 中的索引；i 是完整 11 个参数体系中的索引。
    // 只有 m_parameterSelected[i] 为 true 时，才从 parameters 中消费一个值。
    int paramIndex = 0;

    for(int i = 0; i < m_parameterSelected.size() && i < 11; ++i) {
        if(m_parameterSelected[i] && paramIndex < parameters.size()) {
            double value = parameters[paramIndex];

            switch(i) {
                case 0: // 渗透率
                    reservoirData.getPermeability().setValue(value);
                    break;

                case 3: // 孔隙度
                    reservoirData.getPorosity().setValue(value);
                    break;

                case 4: // 初始压力
                    reservoirData.getInitialPressure().setValue(value);
                    break;

                case 5: // 储层厚度
                    reservoirData.getThickness().setValue(value);
                    break;

                case 6: // 综合压缩系数
                    reservoirData.getCt().setValue(value);
                    break;

                case 7: // 岩石压缩系数
                    reservoirData.getCf().setValue(value);
                    break;

                case 8: // 初始含油饱和度
                    reservoirData.getSoi().setValue(value);
                    break;

                case 9: // 初始含水饱和度
                    reservoirData.getSwi().setValue(value);
                    break;

                case 10: // 初始含气饱和度
                    reservoirData.getSgi().setValue(value);
                    break;
            }

            paramIndex++;
        }
    }

    // 更新数据管理器
    dataManager->updateReservoirData(reservoirData);
}

void nmCalculationAutoFitPSO::updateWellParameters(const QVector<double>& parameters)
{
    // 更新目标井上的拟合参数。目前井级可拟合参数主要是：
    // - skin：写入第一个 perforation；
    // - wellboreC：写入井筒储集系数。
    // 如果目标井不存在或没有射孔数据，这里只记录 debug，不抛异常。
    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();

    // 只更新当前目标井的井参数，避免多井项目中误改其他井。
    //QVector<nmDataWellBase*> wells = dataManager->getWellDataList();
    nmDataWellBase* pWell = dataManager->findWellByName(m_targetWellName);

    if(!pWell) return;

    //nmDataWellBase* pWell = wells[0]; // 使用第一口井

    int paramIndex = 0;

    for(int i = 0; i < m_parameterSelected.size() && i < 9; ++i) {
        if(m_parameterSelected[i] && paramIndex < parameters.size()) {
            double value = parameters[paramIndex];

            switch(i) {
                case 1: { // 表皮系数
                        nmDataPerforation* perf = pWell->getPerforation(0);

                        if(perf) {
                            nmDataAttribute skinAttr = perf->getSkin();
                            skinAttr.setValue(value);
                            perf->setSkin(skinAttr);
                        }
                    }
                    break;

                case 2: { // 井筒储集系数
                        nmDataAttribute wellboreAttr = pWell->getWellboreStorage();
                        wellboreAttr.setValue(value);
                        pWell->setWellboreStorage(wellboreAttr);
                    }
                    break;
            }

            paramIndex++;
        }
    }

    // 根据井类型更新到数据管理器
    updateWellToDataManager(pWell);
}

void nmCalculationAutoFitPSO::updateWellToDataManager(nmDataWellBase* pWell)
{
    if(!pWell) return;

    nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
    NM_WELL_MODEL wellType = pWell->getWellType();

    // DataManager 内部按井型维护不同容器。修改基类指针后，需要根据实际井型
    // 调用对应 update 接口，才能让后续求解器组装读到最新 skin / wellboreC。
    switch(wellType) {
        case NM_WELL_MODEL::Vertical_Well: {
                nmDataVerticalWell* pVerticalWell = dynamic_cast<nmDataVerticalWell*>(pWell);

                if(pVerticalWell) {
                    QVector<nmDataVerticalWell> wells;
                    wells.append(*pVerticalWell);
                    dataManager->updateVerticalWells(wells);
                }

                break;
            }

        case NM_WELL_MODEL::Vertical_Fractured_Well: {
                nmDataVerticalFracturedWell* pVFracturedWell = dynamic_cast<nmDataVerticalFracturedWell*>(pWell);

                if(pVFracturedWell) {
                    QVector<nmDataVerticalFracturedWell> wells;
                    wells.append(*pVFracturedWell);
                    dataManager->updateVerticalFracturedWells(wells);
                }

                break;
            }

        case NM_WELL_MODEL::Horizontal_Fractured_Well: {
                nmDataHorizontalFracturedWell* pHFracturedWell = dynamic_cast<nmDataHorizontalFracturedWell*>(pWell);

                if(pHFracturedWell) {
                    QVector<nmDataHorizontalFracturedWell> wells;
                    wells.append(*pHFracturedWell);
                    dataManager->updateHorizontalFracturedWells(wells);
                }

                break;
            }

        default:
            break;
    }
}

// ==================== 求解器相关方法 ====================

QVector<QVector<double>> nmCalculationAutoFitPSO::runSolver()
{
    // 真实求解器入口。目前默认走 DLL 方式，保留 runSolverExe() 是历史/备用路径。
    // evaluateFitness() 不关心具体求解器形态，只要求返回可验证的双对数结果。
    //return runSolverExe();
    return runSolverDll();
}

// ==================== 数据处理方法 ====================
QVector<QPointF> nmCalculationAutoFitPSO::interpolateData(
    const QVector<QPointF>& source, const QVector<double>& targetX) const
{
    // 将一条曲线插值到指定 targetX 网格。
    // 双对数误差计算前会分别把目标曲线和模拟曲线插到同一个 commonX 上，
    // 这样每个时间点才能逐点比较。
    QVector<QPointF> result;

    if(source.isEmpty() || targetX.isEmpty()) {
        DEBUG_OUT("Warning: Empty data in interpolation");
        return result;
    }

    // 数据清理和验证合并：跳过 NaN/Inf 和极端大值，尽量保留可用点继续计算。
    // 这里不立即把整个粒子判失败，是为了提高求解器偶发异常输出下的鲁棒性。
    QVector<QPointF> validSource;
    const double MAX_REASONABLE_VALUE = 1e12;

    for(int i = 0; i < source.size(); ++i) {
        const QPointF& point = source[i];

        // 检查数值有效性
        if(!isFiniteNumber(point.x()) || !isFiniteNumber(point.y())) {
            DEBUG_OUT(QString("Skipping invalid data point at index %1: X=%2, Y=%3")
                      .arg(i).arg(point.x()).arg(point.y()));
            continue;
        }

        // 检查极大值 - 跳过而不是失败
        if(qAbs(point.y()) > MAX_REASONABLE_VALUE) {
            DEBUG_OUT(QString("Skipping extremely large Y value at index %1: %2")
                      .arg(i).arg(point.y()));
            continue;
        }

        // 只保留有效的数据点
        validSource.append(point);
    }

    // 检查清理后的数据是否足够
    if(validSource.size() < 3) {
        DEBUG_OUT(QString("Insufficient valid data points after cleaning: %1")
                  .arg(validSource.size()));
        return result; // 返回空结果，但不算失败
    }

    DEBUG_OUT(QString("Data cleaning: %1 -> %2 valid points")
              .arg(source.size()).arg(validSource.size()));

    // 对清理后的数据按 X 排序。后续线性插值要求横坐标单调。
    QVector<QPointF> sortedSource = validSource;

    for(int i = 0; i < sortedSource.size() - 1; ++i) {
        for(int j = 0; j < sortedSource.size() - 1 - i; ++j) {
            if(sortedSource[j].x() > sortedSource[j + 1].x()) {
                QPointF temp = sortedSource[j];
                sortedSource[j] = sortedSource[j + 1];
                sortedSource[j + 1] = temp;
            }
        }
    }

    double sourceMinX = sortedSource.first().x();
    double sourceMaxX = sortedSource.last().x();

    double targetMinX = targetX[0];
    double targetMaxX = targetX[0];

    for(int i = 1; i < targetX.size(); ++i) {
        if(targetX[i] < targetMinX) targetMinX = targetX[i];

        if(targetX[i] > targetMaxX) targetMaxX = targetX[i];
    }

    DEBUG_OUT(QString("Source X range: [%1, %2], Target X range: [%3, %4]")
              .arg(sourceMinX).arg(sourceMaxX).arg(targetMinX).arg(targetMaxX));

    // 插值算法：
    // - 在源数据范围内用线性内插；
    // - 目标 X 落在源范围外时做有限斜率外推；
    // - 找不到区间时使用最近点兜底。
    for(int i = 0; i < targetX.size(); ++i) {
        double x = targetX[i];

        if(!isFiniteNumber(x)) continue;

        double y = 0.0;

        // 插值逻辑
        if(x <= sourceMinX) {
            if(sortedSource.size() >= 2) {
                double dx = sortedSource[1].x() - sortedSource[0].x();

                if(qAbs(dx) > 1e-10) {
                    double slope = (sortedSource[1].y() - sortedSource[0].y()) / dx;
                    slope = qMax(-1e6, qMin(1e6, slope));
                    y = sortedSource[0].y() + slope * (x - sortedSource[0].x());
                } else {
                    y = sortedSource[0].y();
                }
            } else {
                y = sortedSource[0].y();
            }
        } else if(x >= sourceMaxX) {
            if(sortedSource.size() >= 2) {
                int lastIdx = sortedSource.size() - 1;
                double dx = sortedSource[lastIdx].x() - sortedSource[lastIdx - 1].x();

                if(qAbs(dx) > 1e-10) {
                    double slope = (sortedSource[lastIdx].y() - sortedSource[lastIdx - 1].y()) / dx;
                    slope = qMax(-1e6, qMin(1e6, slope));
                    y = sortedSource[lastIdx].y() + slope * (x - sortedSource[lastIdx].x());
                } else {
                    y = sortedSource[lastIdx].y();
                }
            } else {
                y = sortedSource.last().y();
            }
        } else {
            // 内插
            bool found = false;

            for(int j = 0; j < sortedSource.size() - 1; ++j) {
                if(x >= sortedSource[j].x() && x <= sortedSource[j + 1].x()) {
                    double dx = sortedSource[j + 1].x() - sortedSource[j].x();

                    if(qAbs(dx) > 1e-10) {
                        double ratio = (x - sortedSource[j].x()) / dx;
                        y = sortedSource[j].y() + ratio * (sortedSource[j + 1].y() - sortedSource[j].y());
                    } else {
                        y = sortedSource[j].y();
                    }

                    found = true;
                    break;
                }
            }

            if(!found) {
                // 使用最近点
                double minDist = 1e10;

                for(int k = 0; k < sortedSource.size(); ++k) {
                    double dist = qAbs(sortedSource[k].x() - x);

                    if(dist < minDist) {
                        minDist = dist;
                        y = sortedSource[k].y();
                    }
                }
            }
        }

        // 最终数值检查
        if(!isFiniteNumber(y)) {
            y = sortedSource.size() > 0 ? sortedSource[0].y() : 1.0;
        }

        y = qMax(-1e12, qMin(1e12, y));

        result.append(QPointF(x, y));
    }

    if(result.isEmpty()) {
        DEBUG_OUT("LogLog interpolation failed");
        return result;
    }

    DEBUG_OUT(QString("Interpolation completed: %1 -> %2 points")
              .arg(validSource.size()).arg(result.size()));

    return result;
}

// ==================== 算法辅助方法 ====================

void nmCalculationAutoFitPSO::adaptiveParameterUpdate(int iteration)
{
    // 自适应调整 PSO 系数。
    // 随迭代推进降低惯性权重，让搜索从全局探索逐步转向局部收敛；
    // 当近期改进很小时，提高 cognitive、降低 social，让粒子更多参考自身历史最优。
    double progress = static_cast<double>(iteration) / m_maxIterations;
    // 更快降低惯性权重，增强局部搜索
    m_inertiaWeight = 0.4 - 0.25 * progress;

    if(iteration > 0) {
        double improvement = m_previousBestFitness - m_globalBestFitness;
        double relativeImprovement = improvement / qMax(1e-10, qAbs(m_previousBestFitness));

        if(relativeImprovement < 1e-5) {
            // 改进很小时，增强局部搜索
            m_cognitiveParam = qMin(2.5, m_cognitiveParam * 1.05);
            m_socialParam = qMax(0.8, m_socialParam * 0.95);
        } else {
            // 有明显改进时，保持平衡
            m_cognitiveParam = qMax(1.5, qMin(2.2, m_cognitiveParam));
            m_socialParam = qMax(1.0, qMin(1.5, m_socialParam));
        }
    }
}

void nmCalculationAutoFitPSO::saveOptimizationResult()
{
    // 当前函数只做日志记录。真正把最优参数写回项目数据的是
    // startAutoFitting() 结束阶段的 applyParametersToDataManager(m_globalBestPosition)。
    DEBUG_OUT(QString("Optimization result: error=%1, evaluations=%2/%3")
              .arg(m_globalBestFitness, 0, 'e', 4)
              .arg(m_successfulEvaluations)
              .arg(m_totalEvaluations));
}

void nmCalculationAutoFitPSO::validateAndProtectFinalResult()
{
    // 最终精英保护。
    // PSO 是随机启发式算法，某些工况下可能没有找到比用户初始模型更好的结果。
    // 这里用初始解误差与最终全局最优误差做比较，如果改进不足，就恢复初始解，
    // 避免“自动拟合”把已有模型调坏。
    if(!m_hasValidUserSolution) {
        emit logMessageGenerated(tr("No initial solution for elite protection"));
        return;
    }

    emit logMessageGenerated(tr("=== Final Result Validation (Elite Protection) ==="));

    // 使用已有的评估结果
    double finalFitness = m_globalBestFitness;
    double initialFitness = m_userInitialFitness;

    emit logMessageGenerated(tr("Comparing results: Initial=%1, Final=%2")
                             .arg(initialFitness, 0, 'e', 4).arg(finalFitness, 0, 'e', 4));

    // 计算改进程度
    double improvement = initialFitness - finalFitness;
    double relativeImprovement = improvement / qMax(1e-10, qAbs(initialFitness));

    emit logMessageGenerated(tr("Improvement: %1 (%2%)")
                             .arg(improvement, 0, 'e', 4).arg(relativeImprovement * 100, 0, 'f', 2));

    if(relativeImprovement < m_improvementThreshold) {
        emit logMessageGenerated(tr("Elite protection triggered: insufficient improvement"));
        emit logMessageGenerated(tr("Threshold: %1%, Actual: %2%")
                                 .arg(m_improvementThreshold * 100, 0, 'f', 2)
                                 .arg(relativeImprovement * 100, 0, 'f', 4));
        emit logMessageGenerated(tr("Restoring initial solution as final result"));

        m_globalBestFitness = initialFitness;
        m_globalBestPosition = m_userInitialSolution;
        m_globalBestLogLogData = m_userInitialLogLogData;
        emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, m_currentIteration + 1, m_globalBestFitness);

        emit logMessageGenerated(tr("Initial solution restored successfully"));
    } else {
        emit logMessageGenerated(tr("Final result validated - significant improvement achieved"));
    }
}

// ==================== 工具方法 ====================

double nmCalculationAutoFitPSO::random01() const
{
    // PSO 使用 Qt 的 qrand/qsrand。随机种子在 startAutoFitting() 中设置，
    // 当前默认固定种子，便于复现实验和 trace。
    return static_cast<double>(qrand()) / RAND_MAX;
}

void nmCalculationAutoFitPSO::clampToLimits(QVector<double>& parameters) const
{
    // 将粒子位置夹回用户给定上下界。parameters 是紧凑的启用参数向量，
    // 需要通过 m_enabledParamIndices 找到它在完整参数体系中的上下界。
    for(int i = 0; i < parameters.size() && i < m_enabledParamIndices.size(); ++i) {
        int paramIndex = m_enabledParamIndices[i];

        if(paramIndex >= 0 && paramIndex < m_parameterLower.size()) {
            parameters[i] = qMax(m_parameterLower[paramIndex],
                                 qMin(m_parameterUpper[paramIndex], parameters[i]));
        }
    }
}

int nmCalculationAutoFitPSO::getEnabledParameterCount() const
{
    // 返回粒子维度，即用户勾选参与拟合的参数数量。
    int count = 0;

    for(int i = 0; i < m_parameterSelected.size(); ++i) {
        if(m_parameterSelected[i]) count++;
    }

    return count;
}

// ==================== 验证和处理方法 ====================

bool nmCalculationAutoFitPSO::validateParameters(const QVector<double>& parameters) const
{
    // 参数合法性检查分两层：
    // 1. 与用户界面设置一致：维度、有限数、上下界；
    // 2. 求解器保护：拦截会导致数值崩溃或明显无物理意义的极端值。
    if(parameters.size() != getEnabledParameterCount()) {
        return false;
    }

    for(int i = 0; i < parameters.size(); ++i) {
        if(!isFiniteNumber(parameters[i])) {
            return false;
        }

        // 检查参数范围
        if(i < m_enabledParamIndices.size()) {
            int paramIndex = m_enabledParamIndices[i];

            if(paramIndex >= 0 && paramIndex < m_parameterLower.size()) {
                if(parameters[i] < m_parameterLower[paramIndex] ||
                        parameters[i] > m_parameterUpper[paramIndex]) {
                    return false;
                }
            }
        }
    }

    // 直接拦截会导致求解器数值崩溃的参数值
    for(int i = 0; i < parameters.size() && i < m_enabledParamIndices.size(); ++i) {
        int paramIndex = m_enabledParamIndices[i];
        double value = parameters[i];

        switch(paramIndex) {
            case 0: // 渗透率：必须大于零
                if(value <= 1e-8) {
                    DEBUG_OUT(QString("Rejecting near-zero permeability: %1").arg(value));
                    return false;
                }

                break;

            case 2: // 井筒储集系数：必须大于零
                if(value <= 1e-10) {
                    DEBUG_OUT(QString("Rejecting near-zero wellbore storage: %1").arg(value));
                    return false;
                }

                break;

            case 3: // 孔隙度：必须在合理范围
                if(value <= 1e-6 || value >= 0.99) {
                    DEBUG_OUT(QString("Rejecting unrealistic porosity: %1").arg(value));
                    return false;
                }

                break;

            case 6: // 综合压缩系数：必须大于零
                if(value <= 1e-8) {
                    DEBUG_OUT(QString("Rejecting near-zero total compressibility: %1").arg(value));
                    return false;
                }

                break;
        }
    }

    return true;
}

bool nmCalculationAutoFitPSO::validateLogLogData(const QVector<QVector<double>>& logLogData) const
{
    // 校验双对数曲线结构。约定：
    // logLogData[0]=time，logLogData[1]=pressure，logLogData[2]=pressure derivative。
    // 三列必须长度一致，且至少有足够点数用于插值和误差计算。
    if(logLogData.size() < 3) {
        DEBUG_OUT("LogLog data has less than 3 arrays");
        return false;
    }

    // 检查数组大小一致性
    int size = logLogData[0].size();

    if(size == 0) {
        DEBUG_OUT("Empty LogLog data");
        return false;
    }

    if(logLogData[1].size() != size || logLogData[2].size() != size) {
        DEBUG_OUT(QString("LogLog data size mismatch: X=%1, Y1=%2, Y2=%3")
                  .arg(logLogData[0].size())
                  .arg(logLogData[1].size())
                  .arg(logLogData[2].size()));
        return false;
    }

    // 检查最小数据点数
    if(size < 5) {
        DEBUG_OUT(QString("Too few LogLog data points: %1").arg(size));
        return false;
    }

    // 数据有效性检查
    for(int i = 0; i < size; ++i) {
        if(!isFiniteNumber(logLogData[0][i]) ||
                !isFiniteNumber(logLogData[1][i]) ||
                !isFiniteNumber(logLogData[2][i])) {
            DEBUG_OUT(QString("Invalid LogLog data at index %1").arg(i));
            return false;
        }
    }

    return true;
}

bool nmCalculationAutoFitPSO::validateInitialValues() const
{
    // 检查当前模型读取出的初始参数是否和用户勾选维度一致，并且在上下界内。
    // 如果初始值越界，算法仍可继续，但日志会提示，因为精英保护可能不可用或效果变差。
    if(m_initialValues.size() != m_enabledParamIndices.size()) {
        DEBUG_OUT("Initial values count mismatch with enabled parameters");
        return false;
    }

    bool allValid = true;

    for(int i = 0; i < m_initialValues.size(); ++i) {
        int paramIndex = m_enabledParamIndices[i];
        double value = m_initialValues[i];

        if(!isFiniteNumber(value)) {
            DEBUG_OUT(QString("Initial value[%1] is not finite: %2").arg(i).arg(value));
            allValid = false;
            continue;
        }

        if(paramIndex < m_parameterLower.size() && paramIndex < m_parameterUpper.size()) {
            double minVal = m_parameterLower[paramIndex];
            double maxVal = m_parameterUpper[paramIndex];

            if(value < minVal || value > maxVal) {
                DEBUG_OUT(QString("Initial value[%1] = %2 is outside bounds [%3, %4]")
                          .arg(i).arg(value).arg(minVal).arg(maxVal));
                allValid = false;
            }
        }
    }

    return allValid;
}

bool nmCalculationAutoFitPSO::validateSolverResult(const QVector<QVector<double>>& result) const
{
    // 校验求解器压力结果。这里检查的是 pressure result，至少需要 time 和 pressure 两列。
    // result log-log 的结构会在 validateLogLogData() 中另行检查。
    if(result.size() < 2) {
        DEBUG_OUT("Solver result has less than 2 arrays");
        return false;
    }

    if(result[0].size() != result[1].size()) {
        DEBUG_OUT(QString("Size mismatch: X=%1, Y=%2").arg(result[0].size()).arg(result[1].size()));
        return false;
    }

    if(result[0].size() == 0) {
        DEBUG_OUT("Empty solver result");
        return false;
    }

    // 检查最小数据点数
    if(result[0].size() < 10) {
        DEBUG_OUT(QString("Too few data points: %1").arg(result[0].size()));
        return false;
    }

    // 数据有效性检查
    for(int i = 0; i < result[0].size(); ++i) {
        if(!isFiniteNumber(result[0][i]) || !isFiniteNumber(result[1][i])) {
            DEBUG_OUT(QString("Invalid data at index %1: X=%2, Y=%3")
                      .arg(i).arg(result[0][i]).arg(result[1][i]));
            return false;
        }
    }

    // 检查X值单调性
    bool isMonotonic = true;

    for(int i = 1; i < result[0].size(); ++i) {
        if(result[0][i] <= result[0][i - 1]) {
            isMonotonic = false;
            break;
        }
    }

    if(!isMonotonic) {
        DEBUG_OUT("X values are not monotonically increasing");
    }

    return true;
}

double nmCalculationAutoFitPSO::calculateCurveError(
    const QVector<QPointF>& curve1, const QVector<QPointF>& curve2) const
{
    // 两条已经对齐到同一 X 网格的曲线误差。
    // 使用“相对误差为主、绝对误差为辅”的点误差，并对极端值做保护。
    if(curve1.size() != curve2.size() || curve1.isEmpty()) {
        return 1e10;
    }

    // 预检查：确保没有无穷大值
    for(int i = 0; i < curve1.size(); ++i) {
        if(!isFiniteNumber(curve1[i].y()) || !isFiniteNumber(curve2[i].y())) {
            DEBUG_OUT(QString("Infinite value detected at index %1: Y1=%2, Y2=%3")
                      .arg(i).arg(curve1[i].y()).arg(curve2[i].y()));
            return 1e10;
        }

        if(qAbs(curve1[i].y()) > 1e12 || qAbs(curve2[i].y()) > 1e12) {
            DEBUG_OUT(QString("Extremely large value detected at index %1")
                      .arg(i));
            return 1e10;
        }
    }

    double totalError = 0.0;
    double totalWeight = 0.0;
    int validPoints = 0;

    for(int i = 0; i < curve1.size(); ++i) {
        double y1 = curve1[i].y();
        double y2 = curve2[i].y();

        // 跳过异常值
        if(!isFiniteNumber(y1) || !isFiniteNumber(y2)) {
            continue;
        }

        // 自适应权重：Y 值越大权重越小，避免大幅值段完全主导误差。
        double weightFactor = qMin(100.0, qAbs(y1) * 0.01);
        double weight = 1.0 / (1.0 + weightFactor);

        // 相对误差和绝对误差的组合 - 添加数值保护
        double yMax = qMax(qAbs(y1), qAbs(y2));
        yMax = qMax(1e-12, yMax); // 防止除零

        double relativeError = qAbs(y1 - y2) / yMax;
        double absoluteError = qAbs(y1 - y2);

        // 限制误差值避免爆炸
        relativeError = qMin(1e6, relativeError);
        absoluteError = qMin(1e6, absoluteError);

        // 误差组合：相对误差为主，绝对误差为辅
        double pointError = 0.7 * relativeError + 0.3 * absoluteError;

        if(isFiniteNumber(pointError) && pointError < 1e10) {
            totalError += weight * pointError * pointError;
            totalWeight += weight;
            validPoints++;
        }
    }

    if(totalWeight > 0 && validPoints > 0) {
        double result = sqrt(totalError / totalWeight);

        // 最终检查
        if(!isFiniteNumber(result)) {
            DEBUG_OUT("Final error calculation produced infinite result");
            return 1e10;
        }

        return qMin(1e9, result); // 限制最大误差值
    } else {
        DEBUG_OUT(QString("No valid points for error calculation: validPoints=%1")
                  .arg(validPoints));
        return 1e10;
    }
}

double nmCalculationAutoFitPSO::calculateLogLogCurveError(
    const QVector<QVector<double> >& target,
    const QVector<QVector<double> >& result) const
{
    // 双对数曲线误差计算。
    //
    // target 通常来自目标井历史曲线，result 来自当前粒子参数下的模拟曲线。
    // 两条曲线的时间点往往不完全一致，所以这里先取两者时间范围的重叠区间，
    // 再在公共时间网格上插值对齐，最后分别计算压力曲线和压力导数曲线误差。
    //
    // 返回值越小表示拟合越好；返回 1e10 表示曲线无效或无法比较。
    // 验证数据
    if(!validateLogLogData(target) || !validateLogLogData(result)) {
        return 1e10;
    }

    try {
        // 数据对齐：找到目标曲线与模拟曲线 time 轴的重叠区域。
        // 不在重叠区域内的点不参与误差，避免外推导致误差失真。
        double targetMinX = target[0][0];
        double targetMaxX = target[0][0];

        for(int i = 1; i < target[0].size(); ++i) {
            if(target[0][i] < targetMinX) targetMinX = target[0][i];

            if(target[0][i] > targetMaxX) targetMaxX = target[0][i];
        }

        double resultMinX = result[0][0];
        double resultMaxX = result[0][0];

        for(int i = 1; i < result[0].size(); ++i) {
            if(result[0][i] < resultMinX) resultMinX = result[0][i];

            if(result[0][i] > resultMaxX) resultMaxX = result[0][i];
        }

        double overlapMinX = qMax(targetMinX, resultMinX);
        double overlapMaxX = qMin(targetMaxX, resultMaxX);

        if(overlapMinX >= overlapMaxX) {
            DEBUG_OUT("No overlap between target and result LogLog curves");
            return 1e10;
        }

        // 生成公共 X 网格进行插值。使用对数均匀网格，是为了给早期时间段
        // 更多分辨率；试井双对数曲线的早期形态通常对参数识别很敏感。
        QVector<double> commonX;
        int numPoints = 50;

        if(overlapMinX > 0 && overlapMaxX > 0) {
            // 对数空间均匀分布
            double logMin = qLn(overlapMinX);
            double logMax = qLn(overlapMaxX);

            for(int i = 0; i < numPoints; ++i) {
                double logX = logMin + i * (logMax - logMin) / (numPoints - 1);
                double x = qExp(logX);

                // 数值保护
                if(!isFiniteNumber(x) || x <= 0) {
                    continue;
                }

                commonX.append(x);
            }

            DEBUG_OUT("Using log-uniform grid for better early-time coverage");
        }

        if(commonX.isEmpty()) {
            DEBUG_OUT("Failed to generate common X grid");
            return 1e10;
        }

        // 插值目标曲线。target[1] 是压力，target[2] 是压力导数。
        QVector<QPointF> targetCurve1, targetCurve2;

        for(int i = 0; i < target[0].size(); ++i) {
            // 检查数据有效性
            if(isFiniteNumber(target[0][i]) && isFiniteNumber(target[1][i]) &&
                    isFiniteNumber(target[2][i])) {
                targetCurve1.append(QPointF(target[0][i], target[1][i]));
                targetCurve2.append(QPointF(target[0][i], target[2][i]));
            }
        }

        if(targetCurve1.isEmpty() || targetCurve2.isEmpty()) {
            DEBUG_OUT("Target curves are empty after filtering");
            return 1e10;
        }

        QVector<QPointF> alignedTarget1 = interpolateData(targetCurve1, commonX);
        QVector<QPointF> alignedTarget2 = interpolateData(targetCurve2, commonX);

        // 插值结果曲线。result 与 target 使用同一 commonX，保证逐点可比。
        QVector<QPointF> resultCurve1, resultCurve2;

        for(int i = 0; i < result[0].size(); ++i) {
            // 检查数据有效性
            if(isFiniteNumber(result[0][i]) && isFiniteNumber(result[1][i]) &&
                    isFiniteNumber(result[2][i])) {
                resultCurve1.append(QPointF(result[0][i], result[1][i]));
                resultCurve2.append(QPointF(result[0][i], result[2][i]));
            }
        }

        if(resultCurve1.isEmpty() || resultCurve2.isEmpty()) {
            DEBUG_OUT("Result curves are empty after filtering");
            return 1e10;
        }

        QVector<QPointF> alignedResult1 = interpolateData(resultCurve1, commonX);
        QVector<QPointF> alignedResult2 = interpolateData(resultCurve2, commonX);

        // 检查插值结果
        if(alignedTarget1.isEmpty() || alignedTarget2.isEmpty() ||
                alignedResult1.isEmpty() || alignedResult2.isEmpty()) {
            DEBUG_OUT("LogLog interpolation failed");
            return 1e10;
        }

        if(alignedTarget1.size() != alignedResult1.size() ||
                alignedTarget2.size() != alignedResult2.size()) {
            DEBUG_OUT("LogLog interpolation size mismatch");
            return 1e10;
        }

        // 计算两条曲线的误差。当前压力和导数各占 50%。
        // 如果后续要让导数形态更重要，可以从这里调整权重。
        double error1 = calculateCurveError(alignedTarget1, alignedResult1);
        double error2 = calculateCurveError(alignedTarget2, alignedResult2);

        // 检查个别误差是否有效
        if(!isFiniteNumber(error1) || error1 > 1e9) {
            DEBUG_OUT(QString("Curve1 error is invalid: %1").arg(error1));
            error1 = 1e10;
        }

        if(!isFiniteNumber(error2) || error2 > 1e9) {
            DEBUG_OUT(QString("Curve2 error is invalid: %1").arg(error2));
            error2 = 1e10;
        }

        // 组合误差 - 添加保护
        double combinedError;

        if(error1 > 1e9 && error2 > 1e9) {
            combinedError = 1e10;
        } else if(error1 > 1e9) {
            combinedError = error2;
        } else if(error2 > 1e9) {
            combinedError = error1;
        } else {
            combinedError = 0.5 * error1 + 0.5 * error2;
        }

        DEBUG_OUT(QString("LogLog errors: Curve1=%1, Curve2=%2, Combined=%3")
                  .arg(error1, 0, 'e', 4).arg(error2, 0, 'e', 4).arg(combinedError, 0, 'e', 4));

        return qMin(1e9, combinedError);

    } catch(const std::exception& e) {
        DEBUG_OUT(QString("Exception in LogLog error calculation: %1").arg(e.what()));
        return 1e10;
    } catch(...) {
        DEBUG_OUT("Unknown exception in LogLog error calculation");
        return 1e10;
    }
}

double nmCalculationAutoFitPSO::calculateWeightedPointError(
    double target, double result, double timeWeight) const
{
    if(!isFiniteNumber(target) || !isFiniteNumber(result)) {
        return 1e10;
    }

    // 对数空间误差：关注数量级差异，适合双对数曲线。
    double logTarget = qLn(qMax(1e-10, qAbs(target)));
    double logResult = qLn(qMax(1e-10, qAbs(result)));
    double logError = qAbs(logTarget - logResult);

    // 线性空间相对误差：补充约束实际幅值差异。
    double yMax = qMax(qAbs(target), qAbs(result));
    double relativeError = qAbs(target - result) / qMax(1e-10, yMax);

    // 组合误差：对数误差占 70%，线性误差占 30%。
    // timeWeight 当前没有参与最终公式，保留参数是为了后续按时间段加权扩展。
    return 0.7 * logError + 0.3 * relativeError;
}

QVector<QVector<double>> nmCalculationAutoFitPSO::runSolverDll()
{
    // DLL 求解器路径。
    // 这个函数负责把当前 DataManager 中的项目状态交给底层数值求解器，
    // 并让目标井生成 result log-log 数据。evaluateFitness() 后续会从目标井读取该结果。
    //
    // 如果这里失败，通常需要优先检查：HX_NWTM.dll、license、网格/井数据是否完整、
    // 目标井是否存在，以及 DataManager 中刚写入的参数是否导致求解器异常。
    DEBUG_OUT("SOLVER DLL START");

    if(m_evaluationInProgress > 0) {
        DEBUG_OUT("DLL Solver already running, skipping");
        return QVector<QVector<double>>();
    }

    ++m_evaluationInProgress;
    QVector<QVector<double>> result;
    nmCalculationDllPebiSolverTask* dllTask = nullptr;

    try {
        DEBUG_OUT("Creating DLL solver task");
        dllTask = new nmCalculationDllPebiSolverTask(m_tempDirectory);

        if(m_shouldStop) {
            DEBUG_OUT("Should stop - cleaning up and returning empty result");
            delete dllTask;
            --m_evaluationInProgress;
            return result;
        }

        DEBUG_OUT("Starting DLL solver execution...");

        // 异步执行 DLL 任务。循环等待期间持续 processEvents，保证界面不会完全卡死。
        dllTask->start();

        // 等待完成。最大等待 1 小时，适配大模型慢算；用户停止时会 terminate。
        int waitTime = 0;
        const int maxWait = 3600000;  // 1h超时
        const int checkInterval = 500;

        while(waitTime < maxWait) {
            QApplication::processEvents(QEventLoop::ExcludeUserInputEvents, 100);

            if(!dllTask->isRunning()) {
                DEBUG_OUT("DLL solver task completed");
                break;
            }

            if(m_shouldStop) {
                DEBUG_OUT("DLL solver task terminated by user");
                dllTask->terminate();
                break;
            }

            msleep(checkInterval);
            waitTime += checkInterval;
        }

        // 超时处理
        if(dllTask->isRunning()) {
            DEBUG_OUT("DLL solver task timeout, terminating...");
            dllTask->terminate();
            dllTask->wait(2000);

            delete dllTask;
            dllTask = nullptr;
            --m_evaluationInProgress;
            m_consecutiveFailures++;
            return result;
        }

        // 验证结果数据是否已更新。DLL 任务会把结果写回 DataManager 中的目标井对象。
        nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
        //QVector<nmDataWellBase*> wells = dataManager->getWellDataList();
        nmDataWellBase* pTargetWell = dataManager->findWellByName(m_targetWellName);

        if(!pTargetWell) {
            DEBUG_OUT("No wells found in data manager after DLL execution");
            delete dllTask;
            --m_evaluationInProgress;
            return result;
        }

        // 验证结果数据。evaluateFitness() 最终用的是 logLogResult，
        // 但这里返回 pressureResult 给 validateSolverResult() 做基本求解成功判断。
        QVector<QVector<double>> pressureResult = pTargetWell->getResultPressure();
        QVector<QVector<double>> logLogResult = pTargetWell->getResultLogLog();

        DEBUG_OUT(QString("DLL result verification - Pressure arrays: %1, LogLog arrays: %2")
                  .arg(pressureResult.size()).arg(logLogResult.size()));

        if(pressureResult.size() >= 2) {
            DEBUG_OUT(QString("Pressure result - Time points: %1, Pressure points: %2")
                      .arg(pressureResult[0].size()).arg(pressureResult[1].size()));

            if(pressureResult[0].size() > 0) {
                DEBUG_OUT(QString("Sample pressure data - Time[0]: %1, Time[last]: %2, P[0]: %3, P[last]: %4")
                          .arg(pressureResult[0][0])
                          .arg(pressureResult[0][pressureResult[0].size() - 1])
                          .arg(pressureResult[1][0])
                          .arg(pressureResult[1][pressureResult[1].size() - 1]));
            }
        }

        // 数据有效性检查
        if(pressureResult.size() >= 2 && pressureResult[0].size() > 0 && pressureResult[1].size() > 0) {
            result = pressureResult;
            DEBUG_OUT(QString("Got DLL solver result: %1 points").arg(result[0].size()));
            m_consecutiveFailures = 0;

            // 检查结果是否与之前不同。若连续粒子得到完全相同的压力曲线，
            // 可能说明参数没有正确写入 DataManager，或求解器缓存/状态没有刷新。
            static QVector<double> lastPressureResult;
            bool isDifferentFromLast = false;

            if(lastPressureResult.isEmpty() || lastPressureResult.size() != pressureResult[1].size()) {
                isDifferentFromLast = true;
            } else {
                for(int i = 0; i < qMin(5, pressureResult[1].size()); ++i) {
                    if(qAbs(lastPressureResult[i] - pressureResult[1][i]) > 1e-12) {
                        isDifferentFromLast = true;
                        break;
                    }
                }
            }

            if(isDifferentFromLast) {
                DEBUG_OUT("RESULT VERIFICATION: Got NEW result data from DLL");
                lastPressureResult = pressureResult[1];
            } else {
                DEBUG_OUT("!!! WARNING: Result data appears to be identical to previous run !!!");
            }

        } else {
            DEBUG_OUT("DLL solver result is empty or invalid");
            DEBUG_OUT(QString("Pressure result size: %1, Array sizes: %2, %3")
                      .arg(pressureResult.size())
                      .arg(pressureResult.size() > 0 ? pressureResult[0].size() : 0)
                      .arg(pressureResult.size() > 1 ? pressureResult[1].size() : 0));
            m_consecutiveFailures++;
        }

    } catch(const std::bad_alloc& e) {
        DEBUG_OUT(QString("Memory allocation failed in DLL solver: %1").arg(e.what()));
        m_consecutiveFailures++;
    } catch(const std::exception& e) {
        DEBUG_OUT(QString("Exception in DLL solver: %1").arg(e.what()));
        m_consecutiveFailures++;
    } catch(...) {
        DEBUG_OUT("Unknown exception in DLL solver");
        m_consecutiveFailures++;
    }

    // 清理DLL任务
    if(dllTask) {
        if(dllTask->isRunning()) {
            dllTask->terminate();
            dllTask->wait(3000);
        }

        DEBUG_OUT("Cleaning up DLL solver task...");
        delete dllTask;
        dllTask = nullptr;
    }

    QApplication::processEvents(QEventLoop::AllEvents, 100);
    --m_evaluationInProgress;

    DEBUG_OUT(QString("SOLVER DLL END - ResultPoints: %1")
              .arg(result.isEmpty() ? 0 : result[0].size()));

    return result;
}

//QVector<QVector<double>> nmCalculationAutoFitPSO::runSolverExe()
//{
//	DEBUG_OUT("SOLVER EXE START");
//
//	if(m_evaluationInProgress > 0) {
//		DEBUG_OUT("EXE Solver already running, skipping");
//		return QVector<QVector<double>>();
//	}
//
//	++m_evaluationInProgress;
//	QVector<QVector<double>> result;
//	nmCalculationExeSolverTask* exeTask = nullptr;
//
//	try {
//		DEBUG_OUT("Creating EXE solver task");
//		exeTask = new nmCalculationExeSolverTask(QString());
//
//		if(m_shouldStop) {
//			DEBUG_OUT("Should stop - cleaning up and returning empty result");
//			delete exeTask;
//			--m_evaluationInProgress;
//			return result;
//		}
//
//		DEBUG_OUT("Starting EXE solver execution...");
//
//		// 启动
//		bool executeSuccess = exeTask->execute();
//
//		// 检查执行状态
//		if(!executeSuccess) {
//			DEBUG_OUT(QString("EXE solver execution failed: %1").arg(exeTask->getLastError()));
//			DEBUG_OUT(QString("EXE solver exit code: %1").arg(exeTask->getExitCode()));
//
//			// 清理并返回空结果
//			delete exeTask;
//			exeTask = nullptr;
//			--m_evaluationInProgress;
//			m_consecutiveFailures++;
//			return result;
//		}
//
//		DEBUG_OUT("EXE solver execution completed successfully");
//
//		// 验证结果数据是否已更新
//		nmDataAnalyzeManager* dataManager = nmDataAnalyzeManager::getCurrentInstance();
//		QVector<nmDataWellBase*> wells = dataManager->getWellDataList();
//
//		if(wells.isEmpty()) {
//			DEBUG_OUT("No wells found in data manager after EXE execution");
//			delete exeTask;
//			--m_evaluationInProgress;
//			return result;
//		}
//
//		// 验证结果数据的时间戳或唯一性
//		QVector<QVector<double>> pressureResult = wells[0]->getResultPressure();
//		QVector<QVector<double>> logLogResult = wells[0]->getResultLogLog();
//
//		DEBUG_OUT(QString("Raw result verification - Pressure arrays: %1, LogLog arrays: %2")
//			.arg(pressureResult.size()).arg(logLogResult.size()));
//
//		if(pressureResult.size() >= 2) {
//			DEBUG_OUT(QString("Pressure result - Time points: %1, Pressure points: %2")
//				.arg(pressureResult[0].size()).arg(pressureResult[1].size()));
//
//			if(pressureResult[0].size() > 0) {
//				DEBUG_OUT(QString("Sample pressure data - Time[0]: %1, Time[last]: %2, P[0]: %3, P[last]: %4")
//					.arg(pressureResult[0][0])
//					.arg(pressureResult[0][pressureResult[0].size()-1])
//					.arg(pressureResult[1][0])
//					.arg(pressureResult[1][pressureResult[1].size()-1]));
//			}
//		}
//
//		if(logLogResult.size() >= 3) {
//			DEBUG_OUT(QString("LogLog result - X points: %1, Y1 points: %2, Y2 points: %3")
//				.arg(logLogResult[0].size()).arg(logLogResult[1].size()).arg(logLogResult[2].size()));
//
//			if(logLogResult[0].size() > 0) {
//				DEBUG_OUT(QString("Sample LogLog data - X[0]: %1, X[last]: %2, Y1[0]: %3, Y2[0]: %4")
//					.arg(logLogResult[0][0])
//					.arg(logLogResult[0][logLogResult[0].size()-1])
//					.arg(logLogResult[1][0])
//					.arg(logLogResult[2][0]));
//			}
//		}
//
//		// 数据有效性检查
//		if(pressureResult.size() >= 2 && pressureResult[0].size() > 0 && pressureResult[1].size() > 0) {
//			result = pressureResult;
//			DEBUG_OUT(QString("Got EXE solver result: %1 points").arg(result[0].size()));
//			m_consecutiveFailures = 0;
//
//			// 检查结果是否与之前不同
//			static QVector<double> lastPressureResult;
//			bool isDifferentFromLast = false;
//
//			if(lastPressureResult.isEmpty() || lastPressureResult.size() != pressureResult[1].size()) {
//				isDifferentFromLast = true;
//			} else {
//				for(int i = 0; i < qMin(5, pressureResult[1].size()); ++i) {
//					if(qAbs(lastPressureResult[i] - pressureResult[1][i]) > 1e-12) {
//						isDifferentFromLast = true;
//						break;
//					}
//				}
//			}
//
//			if(isDifferentFromLast) {
//				DEBUG_OUT("RESULT VERIFICATION: Got NEW result data from EXE");
//				lastPressureResult = pressureResult[1];
//			} else {
//				DEBUG_OUT("!!! WARNING: Result data appears to be identical to previous run !!!");
//			}
//
//		} else {
//			DEBUG_OUT("EXE solver result is empty or invalid");
//			DEBUG_OUT(QString("Pressure result size: %1, Array sizes: %2, %3")
//				.arg(pressureResult.size())
//				.arg(pressureResult.size() > 0 ? pressureResult[0].size() : 0)
//				.arg(pressureResult.size() > 1 ? pressureResult[1].size() : 0));
//			m_consecutiveFailures++;
//		}
//
//	} catch(const std::exception& e) {
//		DEBUG_OUT(QString("Exception in EXE solver: %1").arg(e.what()));
//		m_consecutiveFailures++;
//	} catch(...) {
//		DEBUG_OUT("Unknown exception in EXE solver");
//		m_consecutiveFailures++;
//	}
//
//	// 清理EXE任务
//	if(exeTask) {
//		DEBUG_OUT("Cleaning up EXE solver task...");
//		delete exeTask;
//		exeTask = nullptr;
//	}
//
//	QApplication::processEvents(QEventLoop::AllEvents, 100);
//	--m_evaluationInProgress;
//
//	DEBUG_OUT(QString("SOLVER EXE END - ResultPoints: %1")
//		.arg(result.isEmpty() ? 0 : result[0].size()));
//
//	return result;
//}

StopReasonPSO nmCalculationAutoFitPSO::analyzeOptimizationStatus()
{
    // 按优先级判断本轮是否停止。
    // 目标达成和最大迭代是硬条件；真收敛/局部最优需要足够历史数据支撑。
    // 返回值只描述原因，真正的日志和 UI 收尾在 startAutoFitting() 中处理。

    // 1. 检查用户停止
    if(m_shouldStop) {
        return PSO_USER_STOPPED;
    }

    // 2. 检查连续失败
    if(m_consecutiveFailedIterations >= m_maxConsecutiveFailures) {
        return PSO_CONSECUTIVE_FAILURES;
    }

    // 3. 检查是否达到目标精度
    if(m_globalBestFitness < m_targetError) {
        return PSO_TARGET_ACHIEVED;
    }

    // 4. 检查是否达到最大迭代数
    if(m_currentIteration >= m_maxIterations - 1) {
        return PSO_MAX_ITERATIONS;
    }

    // 5. 需要足够的历史数据才能判断收敛
    if(m_convergenceHistory.size() < m_localOptimumWindow) {
        return PSO_CONTINUE_OPTIMIZATION;
    }

    // 6. 检查真正的收敛
    if(m_convergenceHistory.size() >= m_trueConvergenceWindow && checkTrueConvergence()) {
        return PSO_TRUE_CONVERGENCE;
    }

    // 7. 检查局部最优陷阱
    if(checkLocalOptimumTrap()) {
        return PSO_LOCAL_OPTIMUM;
    }

    return PSO_CONTINUE_OPTIMIZATION;
}

bool nmCalculationAutoFitPSO::checkTrueConvergence() const
{
    // 真收敛判断：不是只看“最近没有改进”，而是同时看解质量、误差稳定性、
    // 粒子群多样性、平均速度、长期改进和粒子停滞率。
    // 这样可以减少把局部卡住误判成正常收敛的概率。
    if(m_convergenceHistory.size() < m_trueConvergenceWindow) {
        return false;
    }

    // 1. 检查解质量 - 如果已经接近目标，小改进可能是真收敛
    bool nearTarget = (m_globalBestFitness < m_targetError * m_nearTargetFactor);

    // 2. 检查适应度稳定性 - 长期小幅波动
    double recentVariance = calculateFitnessVariance(10);
    double recentMean = 0.0;
    int windowSize = qMin(10, m_convergenceHistory.size());

    // 计算最近窗口的均值
    for(int i = m_convergenceHistory.size() - windowSize; i < m_convergenceHistory.size(); ++i) {
        recentMean += m_convergenceHistory[i];
    }

    recentMean /= windowSize;

    double relativeVariance = recentVariance / qMax(1e-10, recentMean * recentMean);
    bool stableError = (relativeVariance < m_convergenceVarianceThreshold);

    // 3. 检查粒子群多样性 - 应该收敛到同一区域
    double currentDiversity = calculateSwarmDiversity();
    bool lowDiversity = (currentDiversity < m_diversityThreshold);

    // 4. 检查速度收敛 - 粒子应该几乎停止移动
    double avgVelocity = calculateAverageVelocity();
    bool velocityConverged = (avgVelocity < m_velocityConvergenceThreshold);

    // 5. 检查长期改进趋势
    double longTermImprovement = calculateLongTermImprovement(m_trueConvergenceWindow);
    bool minimalLongTermImprovement = (longTermImprovement < 1e-4); // 0.01%

    // 6. 检查粒子停滞情况
    double stagnationRate = calculateParticleStagnationRate();
    bool mostParticlesConverged = (stagnationRate > 0.8); // 80%以上粒子收敛

    // 真收敛的判断条件
    bool isConverged = nearTarget ||
                       (stableError && lowDiversity && velocityConverged &&
                        minimalLongTermImprovement && mostParticlesConverged);

    if(isConverged) {
        DEBUG_OUT("=== TRUE CONVERGENCE ANALYSIS ===");
        DEBUG_OUT(QString("nearTarget=%1 (error=%2, target*factor=%3)")
                  .arg(nearTarget).arg(m_globalBestFitness, 0, 'e', 4)
                  .arg(m_targetError * m_nearTargetFactor, 0, 'e', 4));
        DEBUG_OUT(QString("stableError=%1 (relativeVariance=%2)")
                  .arg(stableError).arg(relativeVariance, 0, 'e', 6));
        DEBUG_OUT(QString("lowDiversity=%1 (diversity=%2, threshold=%3)")
                  .arg(lowDiversity).arg(currentDiversity, 0, 'f', 6).arg(m_diversityThreshold));
        DEBUG_OUT(QString("velocityConverged=%1 (avgVel=%2)")
                  .arg(velocityConverged).arg(avgVelocity, 0, 'f', 6));
        DEBUG_OUT(QString("longTermImprovement=%1%, stagnationRate=%2%")
                  .arg(longTermImprovement * 100, 0, 'f', 4).arg(stagnationRate * 100, 0, 'f', 1));
    }

    return isConverged;
}

bool nmCalculationAutoFitPSO::checkLocalOptimumTrap() const
{
    // 局部最优判断：当误差离目标还远，但短期几乎不改进，
    // 且粒子群过度聚集/异常分散或大部分粒子停滞时，认为可能卡住。
    if(m_convergenceHistory.size() < m_localOptimumWindow) {
        return false;
    }

    // 1. 检查解质量 - 如果距离目标还很远，停滞就可能是局部最优
    bool farFromTarget = (m_globalBestFitness > m_targetError * m_farTargetFactor);

    // 2. 检查短期改进 - 近期改进非常小
    double shortTermImprovement = calculateLongTermImprovement(m_localOptimumWindow);
    bool poorShortTermImprovement = (shortTermImprovement < 1e-5); // 0.001%

    // 3. 检查粒子多样性 - 可能过早聚集或无效分散
    double currentDiversity = calculateSwarmDiversity();
    bool problematicDiversity = (currentDiversity < m_diversityThreshold * 0.1) ||
                                (currentDiversity > m_diversityThreshold * 5.0);

    // 4. 检查粒子停滞率
    double stagnationRate = calculateParticleStagnationRate();
    bool mostParticlesStagnant = (stagnationRate > 0.9); // 90%以上停滞但解质量差

    // 5. 检查适应度方差 - 可能卡在平坦区域
    double fitnessVariance = calculateFitnessVariance(m_localOptimumWindow);
    bool flatFitnessLandscape = (fitnessVariance < m_convergenceVarianceThreshold * 0.1);

    // 局部最优的判断条件
    bool isLocalOptimum = farFromTarget && poorShortTermImprovement &&
                          (problematicDiversity || (mostParticlesStagnant && flatFitnessLandscape));

    if(isLocalOptimum) {
        DEBUG_OUT("=== LOCAL OPTIMUM ANALYSIS ===");
        DEBUG_OUT(QString("farFromTarget=%1 (error=%2, target*factor=%3)")
                  .arg(farFromTarget).arg(m_globalBestFitness, 0, 'e', 4)
                  .arg(m_targetError * m_farTargetFactor, 0, 'e', 4));
        DEBUG_OUT(QString("poorShortTermImprovement=%1 (improvement=%2%)")
                  .arg(poorShortTermImprovement).arg(shortTermImprovement * 100, 0, 'f', 4));
        DEBUG_OUT(QString("problematicDiversity=%1 (diversity=%2)")
                  .arg(problematicDiversity).arg(currentDiversity, 0, 'f', 6));
        DEBUG_OUT(QString("stagnationRate=%1%, errorVariance=%2")
                  .arg(stagnationRate * 100, 0, 'f', 1).arg(fitnessVariance, 0, 'e', 6));
    }

    return isLocalOptimum;
}

double nmCalculationAutoFitPSO::calculateSwarmDiversity() const
{
    // 粒子群多样性：逐维计算粒子位置标准差，再按该参数搜索范围归一化。
    // 返回值越小，说明粒子越集中；越大，说明搜索仍然分散。
    if(m_swarm.size() < 2) return 0.0;

    int dimensions = getEnabledParameterCount();

    if(dimensions == 0) return 0.0;

    double totalDiversity = 0.0;

    for(int dim = 0; dim < dimensions; ++dim) {
        // 计算该维度上所有粒子的均值
        double mean = 0.0;

        for(int i = 0; i < m_swarm.size(); ++i) {
            mean += m_swarm[i].position[dim];
        }

        mean /= m_swarm.size();

        // 计算该维度上的方差
        double variance = 0.0;

        for(int i = 0; i < m_swarm.size(); ++i) {
            double diff = m_swarm[i].position[dim] - mean;
            variance += diff * diff;
        }

        variance /= m_swarm.size();

        // 归一化到参数范围
        int paramIndex = m_enabledParamIndices[dim];
        double range = m_parameterUpper[paramIndex] - m_parameterLower[paramIndex];
        double normalizedStd = sqrt(variance) / qMax(1e-10, range);

        totalDiversity += normalizedStd;
    }

    return totalDiversity / dimensions;
}

double nmCalculationAutoFitPSO::calculateAverageVelocity() const
{
    // 平均速度：逐维速度绝对值按参数范围归一化。
    // 收敛阶段速度应逐渐变小。
    if(m_swarm.isEmpty()) return 0.0;

    int dimensions = getEnabledParameterCount();

    if(dimensions == 0) return 0.0;

    double totalVelocity = 0.0;

    for(int i = 0; i < m_swarm.size(); ++i) {
        for(int dim = 0; dim < dimensions; ++dim) {
            int paramIndex = m_enabledParamIndices[dim];
            double range = m_parameterUpper[paramIndex] - m_parameterLower[paramIndex];
            double normalizedVel = qAbs(m_swarm[i].velocity[dim]) / qMax(1e-10, range);
            totalVelocity += normalizedVel;
        }
    }

    return totalVelocity / (static_cast<double>(m_swarm.size()) * dimensions);
}

double nmCalculationAutoFitPSO::calculateParticleStagnationRate() const
{
    // 粒子停滞率：当前 fitness 与 pbest 很接近时认为该粒子停滞。
    // 该指标用于辅助判断真收敛或局部最优。
    if(m_swarm.isEmpty()) return 0.0;

    int stagnantParticles = 0;

    for(int i = 0; i < m_swarm.size(); ++i) {
        // 如果当前适应度与个体最优非常接近，认为该粒子停滞
        double improvement = (m_swarm[i].bestFitness - m_swarm[i].fitness) /
                             qMax(1e-10, qAbs(m_swarm[i].bestFitness));

        if(qAbs(improvement) < 1e-6) { // 改进小于0.0001%
            stagnantParticles++;
        }
    }

    return static_cast<double>(stagnantParticles) / m_swarm.size();
}

double nmCalculationAutoFitPSO::calculateFitnessVariance(int windowSize) const
{
    // 最近 windowSize 代全局最优误差的方差。
    // 方差越小，说明全局最优曲线越稳定。
    if(m_convergenceHistory.size() < windowSize) {
        return 1e10; // 数据不足，返回大值
    }

    // 计算最近windowSize次迭代的方差
    double mean = 0.0;
    int startIdx = m_convergenceHistory.size() - windowSize;

    for(int i = startIdx; i < m_convergenceHistory.size(); ++i) {
        mean += m_convergenceHistory[i];
    }

    mean /= windowSize;

    double variance = 0.0;

    for(int i = startIdx; i < m_convergenceHistory.size(); ++i) {
        double diff = m_convergenceHistory[i] - mean;
        variance += diff * diff;
    }

    variance /= windowSize;

    return variance;
}

double nmCalculationAutoFitPSO::calculateLongTermImprovement(int windowSize) const
{
    // 最近 windowSize 代相对改进幅度。
    // 数据不足时返回 1.0，表示“还不能认为没有改进”。
    if(m_convergenceHistory.size() < windowSize) {
        return 1.0; // 数据不足，假设有改进
    }

    double oldFitness = m_convergenceHistory[m_convergenceHistory.size() - windowSize];
    double improvement = (oldFitness - m_globalBestFitness) / qMax(1e-10, qAbs(oldFitness));

    return improvement;
}

QString nmCalculationAutoFitPSO::getStopReasonDescription(StopReasonPSO reason) const
{
    // 将停止枚举转成人类可读文本，用于日志和运行摘要。
    switch(reason) {
        case PSO_TARGET_ACHIEVED:
            return tr("Target error achieved");

        case PSO_TRUE_CONVERGENCE:
            return tr("Algorithm converged to stable solution");

        case PSO_LOCAL_OPTIMUM:
            return tr("Local optimum detected");

        case PSO_MAX_ITERATIONS:
            return tr("Maximum iterations reached");

        case PSO_USER_STOPPED:
            return tr("Stopped by user request");

        case PSO_CONSECUTIVE_FAILURES:
            return tr("Too many consecutive failures");

        case PSO_OPTIMIZATION_FAILED:
            return tr("Optimization failed");

        default:
            return tr("Unknown reason");
    }
}

void nmCalculationAutoFitPSO::updateConvergenceMetrics()
{
    // 每代结束后缓存收敛辅助指标，只保留最近 50 个点，避免长时间运行无限增长。

    // 更新多样性历史
    m_diversityHistory.append(calculateSwarmDiversity());

    // 更新速度历史
    m_velocityHistory.append(calculateAverageVelocity());

    // 更新粒子停滞率历史
    m_particleStagnationHistory.append(calculateParticleStagnationRate());

    // 保持历史长度合理（最多保留50个数据点）
    const int maxHistorySize = 50;

    while(m_diversityHistory.size() > maxHistorySize) {
        m_diversityHistory.remove(0);
    }

    while(m_velocityHistory.size() > maxHistorySize) {
        m_velocityHistory.remove(0);
    }

    while(m_particleStagnationHistory.size() > maxHistorySize) {
        m_particleStagnationHistory.remove(0);
    }
}

void nmCalculationAutoFitPSO::setSimulationMode(bool enabled)
{
    // 仿真模式开关。它用于 UI 演示/调试进度和日志，不调用真实 DLL 求解器。
    // 正式自动拟合流程通常保持 false。
    m_simulationMode = enabled;
    DEBUG_OUT(QString("Simulation mode %1").arg(enabled ? "enabled" : "disabled"));
}

void nmCalculationAutoFitPSO::setSimulationTargetParams(const QVector<double>& targetParams, double targetError)
{
    // 设置仿真模式的“目标参数”和“目标误差”。
    // 仿真过程会把参数和误差平滑地从初始值过渡到这里。
    m_simulationTargetParams = targetParams;
    m_simulationTargetError = targetError;
    DEBUG_OUT(QString("Simulation target params set: %1 parameters, target error: %2")
              .arg(targetParams.size()).arg(targetError, 0, 'e', 4));
}

QVector<QVector<double> > nmCalculationAutoFitPSO::buildSimulatedLogLogData(double progress) const
{
    // 为仿真模式生成一条看起来逐渐贴近目标曲线的 log-log 曲线。
    // progress=0 表示偏离目标较多，progress=1 基本等于目标曲线。
    QVector<QVector<double> > simulated;

    if(m_targetLogLogData.size() < 3) {
        return simulated;
    }

    int count = qMin(m_targetLogLogData[0].size(), qMin(m_targetLogLogData[1].size(), m_targetLogLogData[2].size()));
    simulated.resize(3);
    simulated[0].reserve(count);
    simulated[1].reserve(count);
    simulated[2].reserve(count);

    double clampedProgress = qMax(0.0, qMin(1.0, progress));
    double pressureOffset = 0.35 * (1.0 - clampedProgress);
    double derivativeOffset = 0.28 * (1.0 - clampedProgress);

    for(int i = 0; i < count; ++i) {
        double x = m_targetLogLogData[0][i];
        double y1 = m_targetLogLogData[1][i];
        double y2 = m_targetLogLogData[2][i];

        if(!isFiniteNumber(x) || !isFiniteNumber(y1) || !isFiniteNumber(y2) || x <= 0.0 || y1 <= 0.0 || y2 <= 0.0) {
            continue;
        }

        double localWave = 1.0 + 0.04 * (1.0 - clampedProgress) * qSin(i * 0.17);
        simulated[0].append(x);
        simulated[1].append(qMax(1e-12, y1 * (1.0 + pressureOffset) * localWave));
        simulated[2].append(qMax(1e-12, y2 * (1.0 - derivativeOffset) / localWave));
    }

    return simulated;
}

void nmCalculationAutoFitPSO::runSimulatedFitting()
{
    // 仿真模式入口：模拟完整 PSO 日志、进度、bestCurveUpdated 信号和最终写回。
    // 它不评价粒子、不调用代理模型、不调用 DLL，只用于界面联调或演示自动拟合流程。
    DEBUG_OUT("=== SIMULATED PSO AUTO FITTING START ===");
    m_isRunning = true;
    m_shouldStop = false;
    m_simulationIteration = 0;
    m_simulationMaxIterations = m_maxIterations;
    m_simulationCurrentParticle = 0;
    m_simulationSuccessCount = 0;
    m_simulationFailCount = 0;
    m_simulationIterationStarted = false;

    // 设置起始误差
    m_simulationStartError = 0.7324;
    m_simulationCurrentError = m_simulationStartError;
    m_simulationPreviousIterError = m_simulationStartError;

    int enabledParams = getEnabledParameterCount();

    // 从界面提取的初始值
    m_simulationStartParams = m_initialValues;

    m_globalBestPosition = m_simulationStartParams;
    m_globalBestFitness = m_simulationStartError;
    m_globalBestLogLogData = buildSimulatedLogLogData(0.0);
    emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, 0, m_globalBestFitness);

    // 发送初始化日志
    emit logMessageGenerated(tr("=== PSO Automatic Fitting Started ==="));
    emit logMessageGenerated(tr("Algorithm: Particle Swarm Optimization"));
    emit logMessageGenerated(tr("Enabled parameters count: %1").arg(enabledParams));
    emit logMessageGenerated(tr("Target data validation passed (%1 data points)")
                             .arg(m_targetLogLogData.isEmpty() ? 156 : m_targetLogLogData[0].size()));

    emit logMessageGenerated(tr("=== Evaluating Initial Solution (Elite Protection) ==="));

    QString paramStr = tr("Initial parameters: ");

    for(int i = 0; i < m_simulationStartParams.size(); ++i) {
        paramStr += QString("[%1]=%2 ").arg(i).arg(m_simulationStartParams[i], 0, 'f', 6);
    }

    emit logMessageGenerated(paramStr);

    emit logMessageGenerated(tr("Starting initial solution evaluation..."));

    QApplication::processEvents();
    msleep(500);

    emit logMessageGenerated(tr("evaluateError returned: %1").arg(m_simulationStartError, 0, 'e', 10));
    emit logMessageGenerated(tr("Taking SUCCESS branch (Error < 1e9)"));
    emit logMessageGenerated(tr("Initial solution evaluation successful"));
    emit logMessageGenerated(tr("Initial Error: %1").arg(m_simulationStartError, 0, 'e', 4));
    emit logMessageGenerated(tr("Elite protection activated - initial solution will be preserved if no significant improvement found"));

    emit logMessageGenerated(tr("Swarm initialized: %1 particles, %2 dimensions").arg(m_swarmSize).arg(enabledParams));
    emit logMessageGenerated(tr("=== Starting PSO Main Loop ==="));

    m_simulationStartTime = QTime::currentTime();

    if(!m_simulationTimer) {
        m_simulationTimer = new QTimer(this);
        connect(m_simulationTimer, SIGNAL(timeout()), this, SLOT(onSimulationTimerTick()));
    }

    int intervalMs = m_simulationLogInterval * 1000;
    m_simulationTimer->start(intervalMs);

    DEBUG_OUT(QString("Simulation timer started with interval: %1 ms").arg(intervalMs));
}

void nmCalculationAutoFitPSO::onSimulationTimerTick()
{
    // 仿真模式用 QTimer 分步输出日志，避免一次性刷完，效果更接近真实长任务。

    // 检查是否需要停止
    if(m_shouldStop) {
        m_simulationTimer->stop();

        emit logMessageGenerated(tr("=== User Stop Request Received ==="));
        emit logMessageGenerated(tr("Gracefully stopping PSO optimization..."));
        emit logMessageGenerated(tr("PSO optimization stop request processed"));

        m_globalBestPosition = calculateSimulatedParams(m_simulationIteration);
        m_globalBestFitness = m_simulationCurrentError;

        m_isRunning = false;

        QString message = QString(tr("Stopped by user. Best error: %1, Iterations: %2"))
                          .arg(m_globalBestFitness, 0, 'e', 4).arg(m_simulationIteration);

        emit logMessageGenerated(tr("=== PSO OPTIMIZATION STOPPED BY USER ==="));
        emit progressUpdated(m_simulationMaxIterations, m_globalBestFitness);

        QApplication::processEvents();
        msleep(200);

        emit fittingFinished(true, message);
        return;
    }

    // 检查是否完成所有迭代
    if(m_simulationIteration >= m_simulationMaxIterations) {
        m_simulationTimer->stop();
        finishSimulation();
        return;
    }

    // 检查是否需要开始新的迭代
    if(!m_simulationIterationStarted) {
        startNewSimulationIteration();
        return;
    }

    // 处理当前粒子
    if(m_simulationCurrentParticle < m_swarmSize) {
        emitParticleLog(m_simulationCurrentParticle);
        m_simulationCurrentParticle++;

        // 每处理2个粒子，处理一次事件
        if(m_simulationCurrentParticle % 2 == 0) {
            QApplication::processEvents();
        }
    } else {
        // 当前迭代的所有粒子都处理完了，输出迭代统计
        finishCurrentIteration();
    }
}

void nmCalculationAutoFitPSO::emitSimulationLog(int iteration)
{
    // 旧版仿真日志函数：一次性输出某一代的模拟粒子情况。
    // 当前 runSimulatedFitting() 主要通过 startNewSimulationIteration()/emitParticleLog()
    // 分步输出，但保留本函数便于调试或恢复旧演示节奏。
    m_simulationCurrentError = calculateSimulatedError(iteration);

    if(m_simulationCurrentError < m_globalBestFitness) {
        QVector<double> currentParams = calculateSimulatedParams(iteration);
        m_globalBestFitness = m_simulationCurrentError;
        m_globalBestPosition = currentParams;
    }

    bool shouldOutputDetail = (iteration % qMax(1, m_simulationMaxIterations / 10) == 0) ||
                              (iteration <= 5) ||
                              (iteration >= m_simulationMaxIterations - 2);

    emit logMessageGenerated(tr("--- Iteration %1/%2 ---").arg(iteration).arg(m_simulationMaxIterations));

    if(shouldOutputDetail) {
        emit logMessageGenerated(tr("Current best error: %1").arg(m_globalBestFitness, 0, 'e', 4));
        emit logMessageGenerated(tr("Total evaluations: %1 (successful: %2, failures: %3)")
                                 .arg(iteration * m_swarmSize)
                                 .arg(iteration * m_swarmSize)
                                 .arg(0));
    }

    int particlesToLog = qMin(3, m_swarmSize);
    double previousError = calculateSimulatedError(iteration - 1);

    for(int i = 0; i < particlesToLog; ++i) {
        int particleId = (iteration * 3 + i) % m_swarmSize + 1;

        double particleOldError = previousError * (1.0 + 0.05 * (qrand() % 100) / 100.0);
        double particleNewError = m_simulationCurrentError * (1.0 + 0.03 * (qrand() % 100) / 100.0);

        if(particleNewError < particleOldError) {
            double improvement = (particleOldError - particleNewError) / qMax(1e-10, particleOldError);

            if(improvement > 0.01) {
                emit logMessageGenerated(tr("  Particle %1 improved: %2 -> %3 (%4% improvement)")
                                         .arg(particleId)
                                         .arg(particleOldError, 0, 'e', 3)
                                         .arg(particleNewError, 0, 'e', 3)
                                         .arg(improvement * 100, 0, 'f', 2));
            } else {
                emit logMessageGenerated(tr("  Particle %1: minor improvement %2% (< 1% threshold)")
                                         .arg(particleId).arg(improvement * 100, 0, 'f', 3));
            }
        } else {
            emit logMessageGenerated(tr("  Particle %1: no improvement (Error: %2)")
                                     .arg(particleId).arg(particleNewError, 0, 'e', 3));
        }
    }

    int successCount = m_swarmSize - (qrand() % 2);
    int failCount = m_swarmSize - successCount;
    double successRate = (double)successCount / m_swarmSize * 100;

    if(shouldOutputDetail) {
        emit logMessageGenerated(tr("Current iteration stats: %1 successful, %2 failed out of %3 particles (success rate: %4%)")
                                 .arg(successCount).arg(failCount).arg(m_swarmSize).arg(successRate, 0, 'f', 1));
    }

    if(iteration % 5 == 0 && iteration > 0) {
        double recentImprovement = (previousError - m_simulationCurrentError) / qMax(1e-10, previousError);
        emit logMessageGenerated(tr("  Convergence check: recent improvement = %1% over last 5 iterations")
                                 .arg(recentImprovement * 100, 0, 'f', 4));
    }

    if(iteration > 0 && iteration % 10 == 0) {
        double progress = (double)iteration / m_simulationMaxIterations;
        double diversity = 0.15 * (1.0 - progress * 0.8) + 0.01;
        double avgVel = 0.08 * (1.0 - progress * 0.9) + 0.005;
        double stagnation = 20.0 + progress * 60.0;

        emit logMessageGenerated(tr("  Optimization status: diversity=%1, avgVel=%2, stagnation=%3%")
                                 .arg(diversity, 0, 'f', 4)
                                 .arg(avgVel, 0, 'f', 4)
                                 .arg(stagnation, 0, 'f', 1));
    }

    if(shouldOutputDetail) {
        emit logMessageGenerated(tr("  Iteration %1 completed - Current best: %2")
                                 .arg(iteration).arg(m_globalBestFitness, 0, 'e', 4));
    }
}

double nmCalculationAutoFitPSO::calculateSimulatedError(int iteration)
{
    // 生成随迭代递减的模拟误差，并叠加少量随机扰动。
    // 使用非线性 decay，让前期下降较快、后期逐渐变慢，更像真实优化曲线。
    if(iteration <= 0) return m_simulationStartError;

    if(iteration >= m_simulationMaxIterations) return m_simulationTargetError;

    double progress = (double)iteration / m_simulationMaxIterations;
    double decayFactor = 1.0 - pow(progress, 0.6);

    double error = m_simulationTargetError +
                   (m_simulationStartError - m_simulationTargetError) * decayFactor;

    double noise = (qrand() % 1000 - 500) / 500000.0;
    error = error * (1.0 + noise);
    error = qMax(m_simulationTargetError, error);

    return error;
}

QVector<double> nmCalculationAutoFitPSO::calculateSimulatedParams(int iteration)
{
    // 生成随迭代从初始参数平滑靠近目标参数的模拟参数。
    // 返回向量仍然是“启用参数向量”，可直接传给 applyParametersToDataManager()。
    QVector<double> currentParams;

    if(iteration <= 0) return m_simulationStartParams;

    if(iteration >= m_simulationMaxIterations) return m_simulationTargetParams;

    double progress = (double)iteration / m_simulationMaxIterations;
    double blendFactor = 1.0 - pow(1.0 - progress, 0.6);

    for(int i = 0; i < m_simulationStartParams.size() && i < m_simulationTargetParams.size(); ++i) {
        double startVal = m_simulationStartParams[i];
        double targetVal = m_simulationTargetParams[i];

        double currentVal = startVal + (targetVal - startVal) * blendFactor;

        double noise = (qrand() % 1000 - 500) / 50000.0;
        currentVal = currentVal * (1.0 + noise);

        currentParams.append(currentVal);
    }

    return currentParams;
}

void nmCalculationAutoFitPSO::startNewSimulationIteration()
{
    // 开始仿真模式的一代：更新误差/最优曲线，输出该代标题和累计评价数。
    m_simulationIteration++;
    m_simulationCurrentParticle = 0;
    m_simulationSuccessCount = 0;
    m_simulationFailCount = 0;
    m_simulationIterationStarted = true;

    // 保存上一次迭代的误差
    m_simulationPreviousIterError = m_simulationCurrentError;

    // 计算当前迭代的目标误差
    m_simulationCurrentError = calculateSimulatedError(m_simulationIteration);

    if(m_simulationCurrentError < m_globalBestFitness) {
        QVector<double> currentParams = calculateSimulatedParams(m_simulationIteration);
        m_globalBestFitness = m_simulationCurrentError;
        m_globalBestPosition = currentParams;
        double progress = (double)m_simulationIteration / qMax(1, m_simulationMaxIterations);
        m_globalBestLogLogData = buildSimulatedLogLogData(progress);
        emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, m_simulationIteration, m_globalBestFitness);
    }

    bool shouldOutputDetail = (m_simulationIteration % qMax(1, m_simulationMaxIterations / 10) == 0) ||
                              (m_simulationIteration <= 5) ||
                              (m_simulationIteration >= m_simulationMaxIterations - 2);

    // 输出迭代标题
    emit logMessageGenerated(tr("--- Iteration %1/%2 ---").arg(m_simulationIteration).arg(m_simulationMaxIterations));

    if(shouldOutputDetail) {
        emit logMessageGenerated(tr("Current best error: %1").arg(m_globalBestFitness, 0, 'e', 4));
        emit logMessageGenerated(tr("Total evaluations: %1 (successful: %2, failures: %3)")
                                 .arg((m_simulationIteration - 1) * m_swarmSize)
                                 .arg((m_simulationIteration - 1) * m_swarmSize)
                                 .arg(0));
    }
}

void nmCalculationAutoFitPSO::emitParticleLog(int particleIndex)
{
    // 输出单个仿真粒子的日志。这里随机生成失败、显著改进、微小改进、无改进四类情况，
    // 目的是让 UI 日志表现接近真实 PSO，而不是用于任何真实优化计算。
    int particleId = particleIndex + 1;

    // 为每个粒子生成模拟的误差值
    double particleNewError = m_simulationCurrentError * (0.97 + 0.06 * (qrand() % 100) / 100.0);

    // 随机决定粒子状态
    int randomOutcome = qrand() % 100;

    if(randomOutcome < 3) {
        // 3% 概率：评估失败
        emit logMessageGenerated(tr("  Particle %1: evaluation failed").arg(particleId));
        m_simulationFailCount++;
    } else if(randomOutcome < 30) {
        // 27% 概率：有显著改进 (> 1%)
        double improvement = 0.01 + 0.12 * (qrand() % 100) / 100.0;
        double oldErr = particleNewError / (1.0 - improvement);

        emit logMessageGenerated(tr("  Particle %1 improved: %2 -> %3 (%4% improvement)")
                                 .arg(particleId)
                                 .arg(oldErr, 0, 'e', 3)
                                 .arg(particleNewError, 0, 'e', 3)
                                 .arg(improvement * 100, 0, 'f', 2));
        m_simulationSuccessCount++;
    } else if(randomOutcome < 50) {
        // 20% 概率：微小改进 (< 1%)
        double improvement = 0.001 + 0.008 * (qrand() % 100) / 100.0;

        emit logMessageGenerated(tr("  Particle %1: minor improvement %2% (< 1% threshold)")
                                 .arg(particleId)
                                 .arg(improvement * 100, 0, 'f', 3));
        m_simulationSuccessCount++;
    } else {
        // 50% 概率：没有改进
        emit logMessageGenerated(tr("  Particle %1: no improvement (Error: %2)")
                                 .arg(particleId)
                                 .arg(particleNewError, 0, 'e', 3));
        m_simulationSuccessCount++;
    }
}

void nmCalculationAutoFitPSO::finishCurrentIteration()
{
    // 完成仿真模式的一代：输出统计、收敛检查和进度更新，然后等待下一次 timer tick。
    m_simulationIterationStarted = false;

    bool shouldOutputDetail = (m_simulationIteration % qMax(1, m_simulationMaxIterations / 10) == 0) ||
                              (m_simulationIteration <= 5) ||
                              (m_simulationIteration >= m_simulationMaxIterations - 2);

    // 每5次迭代输出"发现更好解"
    if(m_simulationIteration % 5 == 0 && m_simulationIteration > 0) {
        emit logMessageGenerated(tr("Better solution found: %1").arg(m_globalBestFitness, 0, 'e', 4));
    }

    // 输出当前迭代统计
    double successRate = (double)m_simulationSuccessCount / m_swarmSize * 100;

    if(shouldOutputDetail) {
        emit logMessageGenerated(tr("Current iteration stats: %1 successful, %2 failed out of %3 particles (success rate: %4%)")
                                 .arg(m_simulationSuccessCount).arg(m_simulationFailCount)
                                 .arg(m_swarmSize).arg(successRate, 0, 'f', 1));
    }

    // 每5次迭代输出收敛检查
    if(m_simulationIteration % 5 == 0 && m_simulationIteration > 0) {
        double recentImprovement = (m_simulationPreviousIterError - m_simulationCurrentError) /
                                   qMax(1e-10, m_simulationPreviousIterError);
        emit logMessageGenerated(tr("  Convergence check: recent improvement = %1% over last 5 iterations")
                                 .arg(recentImprovement * 100, 0, 'f', 4));
    }

    // 每10次迭代输出优化状态
    if(m_simulationIteration > 0 && m_simulationIteration % 10 == 0) {
        double progress = (double)m_simulationIteration / m_simulationMaxIterations;
        double diversity = 0.15 * (1.0 - progress * 0.8) + 0.01;
        double avgVel = 0.08 * (1.0 - progress * 0.9) + 0.005;
        double stagnation = 20.0 + progress * 60.0;

        emit logMessageGenerated(tr("  Optimization status: diversity=%1, avgVel=%2, stagnation=%3%")
                                 .arg(diversity, 0, 'f', 4)
                                 .arg(avgVel, 0, 'f', 4)
                                 .arg(stagnation, 0, 'f', 1));
    }

    // 输出迭代完成
    if(shouldOutputDetail) {
        emit logMessageGenerated(tr("  Iteration %1 completed - Current best: %2")
                                 .arg(m_simulationIteration).arg(m_globalBestFitness, 0, 'e', 4));
    }

    // 更新进度
    emit progressUpdated(m_simulationIteration, m_simulationCurrentError);

    QApplication::processEvents();
}

void nmCalculationAutoFitPSO::finishSimulation()
{
    // 仿真模式正常结束：直接采用预设目标参数/目标误差作为最终结果，
    // 发送与真实拟合相同风格的日志、进度和 fittingFinished 信号。
    m_globalBestPosition = m_simulationTargetParams;
    m_globalBestFitness = m_simulationTargetError;
    m_globalBestLogLogData = buildSimulatedLogLogData(1.0);
    emit bestCurveUpdated(m_targetLogLogData, m_globalBestLogLogData, m_simulationMaxIterations, m_globalBestFitness);

    applyParametersToDataManager(m_globalBestPosition);

    emit logMessageGenerated(tr("=== TRUE CONVERGENCE DETECTED ==="));
    emit logMessageGenerated(tr("Algorithm has converged to a stable solution"));
    emit logMessageGenerated(tr("Final error: %1 after %2 iterations")
                             .arg(m_globalBestFitness, 0, 'e', 4).arg(m_simulationMaxIterations));
    emit logMessageGenerated(tr("Solution quality: %1 (1.0 = target achieved)")
                             .arg(m_targetError / qMax(1e-10, m_globalBestFitness), 0, 'f', 3));

    emit logMessageGenerated(tr("Applying optimized parameters to model..."));
    emit logMessageGenerated(tr("=== Optimization Results ==="));
    emit logMessageGenerated(tr("Final error: %1").arg(m_globalBestFitness, 0, 'e', 4));
    emit logMessageGenerated(tr("Total iterations: %1").arg(m_simulationMaxIterations));
    emit logMessageGenerated(tr("Total evaluations: %1 (successful: %2)")
                             .arg(m_simulationMaxIterations * m_swarmSize)
                             .arg(m_simulationMaxIterations * m_swarmSize));

    QString finalParams = tr("Optimized parameters: ");

    for(int i = 0; i < m_globalBestPosition.size(); ++i) {
        finalParams += QString("[%1]=%2 ").arg(i).arg(m_globalBestPosition[i], 0, 'f', 6);
    }

    emit logMessageGenerated(finalParams);

    emit logMessageGenerated(tr("Parameters applied successfully to data manager"));

    emit logMessageGenerated(tr("=== PSO OPTIMIZATION CONVERGED ==="));

    m_isRunning = false;

    emit progressUpdated(m_simulationMaxIterations, m_globalBestFitness);
    QApplication::processEvents();
    msleep(200);

    QString message = QString(tr("PSO optimization converged to stable solution. Best error: %1, Iterations: %2"))
                      .arg(m_globalBestFitness, 0, 'e', 4).arg(m_simulationMaxIterations);
    emit fittingFinished(true, message);
}