feat(nmNum): 增加拟合参数结果展示

- 内存保留每组最优参数，结果表增加拟合参数列。
- 直方图和散点图支持切换拟合参数。
- 基准点使用求解快照，误差轴保留阈值筛选。

Include/nmNum/nmCalculation/nmCalculationDllPebiSolverTask.h

@@ -149,7 +149,7 @@ class NMCALCULATION_EXPORT nmCalculationDllPebiSolverTask : public QThread {
     /** @brief 复用当前独立副本串行计算，保留进度、停止状态及每组失败原因。 */
     bool executeSensitivityBatch();
     /** @brief 固定本组采样参数，在独立输入上拟合并生成最终曲线。 */
-    bool improveSensitivityRun();
+    bool improveSensitivityRun(QVector<double>& fittedParameters);
     /** @brief 执行当前唯一的 PEBI 求解模式。 */
     bool execute();
     /** @brief 成对释放构造时登记的 DataManager 后台使用权。 */

Include/nmNum/nmData/nmDataSensitive.h

@@ -106,6 +106,7 @@ public:
         RunResult() : m_nRunIndex(0), m_dPressureRmse(-1.0), m_bSucceeded(false), m_bCancelled(false), m_bVisible(true) {}
         quint64 m_nRunIndex;
         QVector<double> m_vecParameters;
+        QVector<double> m_vecFittedParameters; // 本组 LM 最优值，与批次拟合参数顺序一致，仅存内存。
         QVector<QVector<double> > m_vecPressure;
         QVector<QVector<double> > m_vecLogLog;
         QVector<QVector<double> > m_vecSemiLog;
@@ -118,9 +119,10 @@ public:
 
     struct NM_DATA_EXPORT BatchResult
     {
-        BatchResult() : m_calcType(CALC_DETERMINISTIC), m_dFilterThreshold(-1.0), m_bCompleted(false), m_bStopped(false) {}
+        BatchResult() : m_calcType(CALC_DETERMINISTIC), m_dFilterThreshold(-1.0), m_nHistogramVariable(-1), m_bCompleted(false), m_bStopped(false) {}
         CalculationType m_calcType;
         QList<VariableSampling> m_variables; // 求解时冻结的配置，独立于之后的参数编辑。
+        QList<nmDataAttribute> m_fitParameters; // 名称、基准单位和求解快照中的模型值，仅存内存。
         QString m_sWellCode;
         QString m_sWellName;
         QString m_sAxisXName;
@@ -132,6 +134,8 @@ public:
         std::vector<RunResult> m_runs;
         // 散点筛选阈值只随当前批次保存在内存中，负值表示尚未筛选。
         double m_dFilterThreshold;
+        int m_nHistogramVariable; // -1 为误差，其余为拟合参数下标，仅存内存。
+        bool statisticValue(const RunResult& run, int variable, double& value) const;
         bool filteredParameterBounds(int index, double& minimum, double& maximum) const;
         bool m_bCompleted;
         bool m_bStopped;

Include/nmNum/nmSubWnd/nmSubWndMain.h

@@ -173,7 +173,7 @@ class NM_SUB_WND_EXPORT nmSubWndMain : public iSubWndBaseFit {
     /** @brief 刷新主界面的误差直方图，修改分箱数时复用当前批次结果。 */
     void updateSensitiveHistogramPlot();
     void onSensitiveHistogramOptions();
-    void onSensitiveHistogramBinsChanged(int bins);
+    void onSensitiveHistogramSettingsChanged(int bins, int variable);
     void onSensitiveScatter();
     /** @brief 当前分析的建网提交结束后，恢复待启动或待捕获的求解流程。 */
     void onPebiGridGenerationFinished(quint64 nGridInputRevision,

Include/nmNum/nmSubWxs/nmWxSensitive.h

@@ -35,12 +35,13 @@ public:
     nmWxSensitivityHistogramOptions(const QSharedPointer<nmDataSensitive::BatchResult>& results,
                                      int bins, QWidget* parent);
 signals:
-    void binsChanged(int bins);
+    void settingsChanged(int bins, int variable);
 private slots:
     void onApply();
     void onAccept();
 private:
     QSpinBox* m_bins;
+    QComboBox* m_variable;
 };
 
 class NM_SUB_WXS_EXPORT nmWxSensitivityScatter : public iDlgBase
@@ -65,6 +66,7 @@ private:
     double m_defaultThreshold;
     bool m_draggingThreshold;
     QComboBox* m_parameter;
+    QComboBox* m_ordinate;
     double m_threshold;
     QLabel* m_legend;
     QAction* m_resultsInPlot;

Src/nmNum/nmCalculation/nmCalculationDllPebiSolverTask.cpp

@@ -28,6 +28,8 @@
 #include "nmDataPvtParaForPebi.h"
 
 #include <QDebug>
+#include <QCoreApplication>
+#include <limits>
 #include <QHash>
 #include <QMutexLocker>
 #include <QScopedPointer>
@@ -991,6 +993,29 @@ nmCalculationDllPebiSolverTask* nmCalculationDllPebiSolverTask::createSensitivit
         task->m_pSensitivityResult->m_sWellCode = task->m_sAutoFitTargetWellCode;
         if(m_pDataManager != nullptr) {
             const QSharedPointer<const nmPebiResultSnapshot> solved = m_pDataManager->getPebiResultSnapshot();
+            // 拟合参数的蓝色基准点来自原模型，不使用设置窗口中可编辑的拟合初值。
+            if(improveSettings && !solved.isNull()) {
+                const nmDataAutomaticFitting& fit = improveSettings->parameters;
+                const bool selected[] = { fit.getPermeabilitySelected(), fit.getSkinSelected(),
+                    fit.getWellboreStorageSelected(), fit.getPorositySelected(),
+                    fit.getFractureConductivitySelected(), fit.getFractureHalfLengthSelected() };
+                const char* names[] = { "Permeability", "Skin", "Wellbore storage", "Porosity",
+                    "Fracture conductivity", "Fracture half length" };
+                const char* units[] = { "mD", "", "m^3/MPa", "", "mD.m", "m" };
+                for(int i = 0; i < solved->getWellCount(); ++i) {
+                    const nmPebiResultWellSnapshot* well = solved->getWellAt(i);
+                    if(well->m_sWellName != improveSettings->targetWell) continue;
+                    double halfLength = std::numeric_limits<double>::quiet_NaN();
+                    if(selected[5]) getSolvedFractureHalfLength(solved->getSensitivityInput(), well->m_sWellCode, halfLength);
+                    const double values[] = { m_pInputSnapshot->m_oStaticPropertyInput.m_dReferencePermeabilityMilliDarcy,
+                        well->m_dSkin, well->m_dWellboreStorage,
+                        m_pInputSnapshot->m_oStaticPropertyInput.m_dReferencePorosity, well->m_dDfc, halfLength };
+                    for(int j = 0; j < 6; ++j) if(selected[j])
+                        task->m_pSensitivityResult->m_fitParameters.append(nmDataAttribute(
+                            QCoreApplication::translate("nmWxAutomaticFitting", names[j]), values[j], units[j]));
+                    break;
+                }
+            }
             for(int i = 0; !solved.isNull() && i < solved->getWellCount(); ++i) {
                 const nmPebiResultWellSnapshot* well = solved->getWellAt(i);
                 if(well->m_sWellCode != task->m_sAutoFitTargetWellCode) continue;
@@ -1065,7 +1090,7 @@ bool nmCalculationDllPebiSolverTask::getSolvedFractureHalfLength(
     return false;
 }
 
-bool nmCalculationDllPebiSolverTask::improveSensitivityRun()
+bool nmCalculationDllPebiSolverTask::improveSensitivityRun(QVector<double>& fittedParameters)
 {
     // 采样值已经写入本任务；候选只覆盖选中的拟合参数，不触碰管理器和正式结果。
     const nmSensitivityFitSettings& settings = *m_pSensitivityFitSettings;
@@ -1172,6 +1197,8 @@ bool nmCalculationDllPebiSolverTask::improveSensitivityRun()
         return fail(error.isEmpty() ? tr("No valid fitting result was obtained.") : error);
     }
     evaluate(best);
+    // 最优参数与最终曲线一同返回，失败组不留下可显示的拟合值。
+    if(m_lastRunSucceeded) fittedParameters = best;
     return m_lastRunSucceeded;
 }
 
@@ -1210,7 +1237,7 @@ bool nmCalculationDllPebiSolverTask::executeSensitivityBatch()
                     calculateSingleRun();
                     result.m_bSucceeded = m_lastRunSucceeded;
                 } else {
-                    result.m_bSucceeded = improveSensitivityRun();
+                    result.m_bSucceeded = improveSensitivityRun(result.m_vecFittedParameters);
                 }
                 if(result.m_bSucceeded) {
                     result.m_vecPressure = m_autoFitResultPressure;
@@ -1235,6 +1262,7 @@ bool nmCalculationDllPebiSolverTask::executeSensitivityBatch()
                 result.m_vecLogLog.clear();
                 result.m_vecSemiLog.clear();
                 result.m_dPressureRmse = -1.0;
+                result.m_vecFittedParameters.clear();
             } else if(!result.m_bSucceeded) {
                 ++failed;
             }

Src/nmNum/nmData/nmDataSensitive.cpp

@@ -5,6 +5,19 @@
 #include <QDateTime>
 #include <QtCore/qnumeric.h>
 
+bool nmDataSensitive::BatchResult::statisticValue(const RunResult& run, int variable, double& value) const
+{
+    // 表格、直方图和散点图共用有效性判断，拟合参数不依赖压力误差是否可用。
+    if(!run.m_bSucceeded || run.m_bCancelled) return false;
+    if(variable == -1) {
+        value = run.m_dPressureRmse;
+        return qIsFinite(value) && value >= 0.0;
+    }
+    if(variable < 0 || variable >= m_fitParameters.size() || variable >= run.m_vecFittedParameters.size()) return false;
+    value = run.m_vecFittedParameters[variable];
+    return qIsFinite(value);
+}
+
 bool nmDataSensitive::BatchResult::filteredParameterBounds(int index, double& minimum, double& maximum) const
 {
     // 所有参数共用同一批误差严格小于阈值的样本，不受曲线可见性影响。

Src/nmNum/nmSubWnd/nmSubWndMain.cpp

@@ -2231,32 +2231,37 @@ void nmSubWndMain::updateSensitiveHistogramPlot()
     }
     iGuiPlot* plot = window->widget()->findChild<iGuiPlot*>();
     const int binCount = window->widget()->property("nmHistogramBins").toInt();
-    QVector<double> errors;
+    // 拟合参数与误差共用分箱绘制，轴名称和单位随本批选择同步切换。
+    const int variable = batch->m_nHistogramVariable;
+    plot->setAxisX(variable < 0 ? tr("Goodness of fit") : batch->m_fitParameters[variable].getName(),
+                  variable < 0 ? "MPa" : batch->m_fitParameters[variable].getUnit());
+    QVector<double> values;
     for(size_t i = 0; i < batch->m_runs.size(); ++i) {
         const nmDataSensitive::RunResult& run = batch->m_runs[i];
-        // 显示勾选不改变统计总体，失败、取消或无有效误差的运行不参与统计。
-        if(run.m_bSucceeded && !run.m_bCancelled && qIsFinite(run.m_dPressureRmse) && run.m_dPressureRmse >= 0.0)
-            errors.append(run.m_dPressureRmse);
+        // 显示勾选不改变统计总体，失败、取消或无有效统计值的运行不参与统计。
+        double value;
+        if(batch->statisticValue(run, variable, value)) values.append(value);
     }
     double minimum = 0.0, maximum = 1.0;
-    if(!errors.isEmpty()) {
-        minimum = maximum = errors.first();
-        for(int i = 1; i < errors.size(); ++i) {
-            minimum = qMin(minimum, errors[i]);
-            maximum = qMax(maximum, errors[i]);
+    if(!values.isEmpty()) {
+        minimum = maximum = values.first();
+        for(int i = 1; i < values.size(); ++i) {
+            minimum = qMin(minimum, values[i]);
+            maximum = qMax(maximum, values[i]);
         }
-        // 单个结果或全部相同的误差也需要非零横轴范围。
+        // 单个结果或全部相同的统计值也需要非零横轴范围。
         if(minimum == maximum) {
-            const double padding = qMax(1e-6, maximum * 0.05);
-            minimum = qMax(0.0, minimum - padding);
+            const double padding = qMax(1e-6, qAbs(maximum) * 0.05);
+            minimum -= padding;
+            if(variable < 0) minimum = qMax(0.0, minimum);
             maximum += padding;
         }
     }
     const double width = (maximum - minimum) / binCount;
     QVector<int> counts(binCount, 0);
-    for(int i = 0; i < errors.size(); ++i) {
-        // 等宽分箱，左闭右开，最大误差归入最后一箱。
-        const int index = qMin(binCount - 1, int((errors[i] - minimum) / width));
+    for(int i = 0; i < values.size(); ++i) {
+        // 等宽分箱，左闭右开，最大值归入最后一箱。
+        const int index = qMin(binCount - 1, int((values[i] - minimum) / width));
         ++counts[index];
     }
     for(int i = plot->m_pPlot->getObjCount() - 1; i >= 0; --i)
@@ -2264,7 +2269,7 @@ void nmSubWndMain::updateSensitiveHistogramPlot()
     plot->clearUndoStack();
     QVector<iCurveDesc> curves;
     int largestCount = 0;
-    if(!errors.isEmpty()) {
+    if(!values.isEmpty()) {
         QVector<QPointF> bars;
         iCurveDesc cumulative;
         cumulative.m_sCurveName = tr("Cumulative probability");
@@ -2281,7 +2286,7 @@ void nmSubWndMain::updateSensitiveHistogramPlot()
             largestCount = qMax(largestCount, counts[i]);
             total += counts[i];
             cumulative.m_vecX.append(center);
-            cumulative.m_vecY.append(100.0 * total / errors.size());
+            cumulative.m_vecY.append(100.0 * total / values.size());
         }
         // 普通曲线入口不支持柱状图，直接使用框架柱状图元，累计概率仍走已有曲线入口。
         nmSensitivityHistogramBars* histogram = new nmSensitivityHistogramBars(width);
@@ -2334,7 +2339,7 @@ void nmSubWndMain::onSensitiveHistogramOptions()
     nmWxSensitivityHistogramOptions* dialog = window->findChild<nmWxSensitivityHistogramOptions*>();
     if(!dialog) {
         dialog = new nmWxSensitivityHistogramOptions(batch, window->widget()->property("nmHistogramBins").toInt(), window->widget());
-        connect(dialog, SIGNAL(binsChanged(int)), this, SLOT(onSensitiveHistogramBinsChanged(int)));
+        connect(dialog, SIGNAL(settingsChanged(int,int)), this, SLOT(onSensitiveHistogramSettingsChanged(int,int)));
         connect(manager, SIGNAL(sigSensitiveResultsChanged()), dialog, SLOT(reject()));
     }
     dialog->show();
@@ -2342,9 +2347,14 @@ void nmSubWndMain::onSensitiveHistogramOptions()
     dialog->activateWindow();
 }
 
-void nmSubWndMain::onSensitiveHistogramBinsChanged(int bins)
+void nmSubWndMain::onSensitiveHistogramSettingsChanged(int bins, int variable)
 {
-    // 分箱设置仅影响本图，散点图继续使用未分箱的逐组误差。
+    // 直方图变量仅存于本批内存结果，散点图独立选择自己的纵轴。
+    nmDataAnalyzeManager* manager = m_pResultRefreshDataManager.data();
+    if(!manager || !manager->getSensitiveData()) return;
+    const QSharedPointer<nmDataSensitive::BatchResult> batch = manager->getSensitiveData()->getResults();
+    if(batch.isNull()) return;
+    batch->m_nHistogramVariable = variable;
     iSubWndFitting* fitting = m_pResultRefreshFitting.data();
     if(!fitting || !fitting->m_pMdiArea) return;
     QMdiSubWindow* window = fitting->m_pMdiArea->findChild<QMdiSubWindow*>("nmSensitivityHistogram");

Src/nmNum/nmSubWxs/nmWxSensitive.cpp

@@ -246,15 +246,22 @@ public:
 
     int columnCount(const QModelIndex& parent = QModelIndex()) const override
     {
-        return parent.isValid() ? 0 : 3;
+        return parent.isValid() ? 0 : 3 + (m_statistics ? m_results->m_fitParameters.size() : 0);
     }
 
     QVariant headerData(int section, Qt::Orientation orientation, int role) const override
     {
         // 表格仅保留显示选择、运行编号与冻结参数。
         if(orientation != Qt::Horizontal || role != Qt::DisplayRole) return QVariant();
-        if(m_statistics) return (QStringList() << "#" << nmWxSensitive::tr("Realization")
+        if(m_statistics) {
+            if(section >= 3 && section - 3 < m_results->m_fitParameters.size()) {
+                const nmDataAttribute& parameter = m_results->m_fitParameters[section - 3];
+                return parameter.getName() + (parameter.getUnit().isEmpty() ? QString()
+                    : " [" + parameter.getUnit() + "]");
+            }
+            return (QStringList() << "#" << nmWxSensitive::tr("Realization")
                 << nmWxSensitive::tr("Goodness of fit [MPa]")).value(section);
+        }
         const QStringList titles = QStringList() << nmWxSensitive::tr("Show") << "#"
                 << nmWxSensitive::tr("Parameters");
         return titles.value(section);
@@ -270,7 +277,8 @@ public:
             if(role != Qt::DisplayRole && role != Qt::ToolTipRole) return QVariant();
             if(index.column() == 0) return QString::number(run.m_nRunIndex + 1);
             if(index.column() == 1) return sensitivityRunParameters(*m_results, run);
-            return hasSensitivityError(run) ? QString::number(run.m_dPressureRmse, 'g', 6)
+            double value;
+            return m_results->statisticValue(run, index.column() - 3, value) ? QString::number(value, 'g', 6)
                     : run.m_bCancelled ? nmWxSensitive::tr("Cancelled")
                     : !run.m_bSucceeded ? nmWxSensitive::tr("Failed") : nmWxSensitive::tr("N/A");
         }
@@ -420,7 +428,7 @@ nmWxSensitivityHistogramOptions::nmWxSensitivityHistogramOptions(
         const QSharedPointer<nmDataSensitive::BatchResult>& results, int bins, QWidget* parent)
     : iDlgBase(parent)
 {
-    // 复用运行表格的参数展示，普通 Monte-Carlo 的统计变量只有误差。
+    // 复用运行表格；第四种在误差后追加拟合参数，选择变量后按原有 Apply/OK 流程生效。
     setAttribute(Qt::WA_DeleteOnClose);
     setObjectName("nmHistogramOptions");
     setWindowTitle(tr("Histogram options"));
@@ -434,9 +442,12 @@ nmWxSensitivityHistogramOptions::nmWxSensitivityHistogramOptions(
     options->addWidget(m_bins);
     options->addStretch();
     options->addWidget(new QLabel(tr("Variable:"), this));
-    QComboBox* variable = new QComboBox(this);
-    variable->addItem(tr("Goodness of fit"));
-    options->addWidget(variable);
+    m_variable = new QComboBox(this);
+    m_variable->addItem(tr("Goodness of fit"), -1);
+    for(int i = 0; i < results->m_fitParameters.size(); ++i)
+        m_variable->addItem(results->m_fitParameters[i].getName(), i);
+    m_variable->setCurrentIndex(m_variable->findData(results->m_nHistogramVariable));
+    options->addWidget(m_variable);
     layout->addLayout(options);
     QTableView* table = new QTableView(this);
     table->setStyleSheet("QTableView { background-color: white; }");
@@ -447,7 +458,8 @@ nmWxSensitivityHistogramOptions::nmWxSensitivityHistogramOptions(
     table->verticalHeader()->hide();
     table->horizontalHeader()->setResizeMode(0, QHeaderView::ResizeToContents);
     table->horizontalHeader()->setResizeMode(1, QHeaderView::Stretch);
-    table->horizontalHeader()->setResizeMode(2, QHeaderView::ResizeToContents);
+    for(int i = 2; i < table->model()->columnCount(); ++i)
+        table->horizontalHeader()->setResizeMode(i, QHeaderView::ResizeToContents);
     layout->addWidget(table);
     QHBoxLayout* buttons = new QHBoxLayout;
     buttons->addStretch();
@@ -465,8 +477,8 @@ nmWxSensitivityHistogramOptions::nmWxSensitivityHistogramOptions(
 
 void nmWxSensitivityHistogramOptions::onApply()
 {
-    // 只提交分箱数，取消不会撤销已经应用的设置。
-    emit binsChanged(m_bins->value());
+    // 分箱数与横轴变量一起提交，取消不会撤销已经应用的设置。
+    emit settingsChanged(m_bins->value(), m_variable->itemData(m_variable->currentIndex()).toInt());
 }
 
 void nmWxSensitivityHistogramOptions::onAccept()
@@ -496,9 +508,11 @@ nmWxSensitivityScatter::nmWxSensitivityScatter(
     options->addWidget(m_parameter, 1);
     options->addSpacing(20);
     options->addWidget(new QLabel(tr("Y Axis:"), this));
-    QComboBox* ordinate = new QComboBox(this);
-    ordinate->addItem(tr("Goodness of fit"));
-    options->addWidget(ordinate, 1);
+    m_ordinate = new QComboBox(this);
+    m_ordinate->addItem(tr("Goodness of fit"), -1);
+    for(int i = 0; i < results->m_fitParameters.size(); ++i)
+        m_ordinate->addItem(results->m_fitParameters[i].getName(), i);
+    options->addWidget(m_ordinate, 1);
     for(size_t i = 0; i < results->m_runs.size(); ++i)
         if(hasSensitivityError(results->m_runs[i])) m_defaultThreshold = qMax(m_defaultThreshold, results->m_runs[i].m_dPressureRmse);
     if(qIsFinite(baseError) && baseError >= 0.0) m_defaultThreshold = baseError;
@@ -526,13 +540,19 @@ nmWxSensitivityScatter::nmWxSensitivityScatter(
     updateLegend();
     m_plot->m_pPlotView->viewport()->installEventFilter(this);
     connect(m_parameter, SIGNAL(currentIndexChanged(int)), this, SLOT(updateParameter()));
+    connect(m_ordinate, SIGNAL(currentIndexChanged(int)), this, SLOT(updateParameter()));
     connect(m_plot->m_pPlotView, SIGNAL(sigAdjustContextMenu(QMenu*)), this, SLOT(adjustContextMenu(QMenu*)));
     updateParameter();
 }
 
 void nmWxSensitivityScatter::updateParameter()
 {
-    // 切换横轴时重建参数坐标，阈值在各参数之间共用。
+    // 两轴切换都从本批结果重建坐标；拟合参数轴不使用误差阈值。
+    m_draggingThreshold = false;
+    const int ordinate = m_ordinate->itemData(m_ordinate->currentIndex()).toInt();
+    const bool errorAxis = ordinate == -1;
+    const double modelY = errorAxis ? m_baseError : m_results->m_fitParameters[ordinate].getValue().toDouble();
+    m_plot->setAxisY(m_ordinate->currentText(), errorAxis ? "MPa" : m_results->m_fitParameters[ordinate].getUnit());
     m_points.clear();
     m_runRows.clear();
     if(m_parameter->currentIndex() < 0) {
@@ -545,17 +565,19 @@ void nmWxSensitivityScatter::updateParameter()
     m_plot->setAxisX(variable.m_varName, unit);
     double modelValue = variable.m_modelValue.getValue().toDouble();
     const bool hasModel = convertSamplingValue(variable.m_varName, modelValue,
-                samplingBaseUnit(variable.m_varName), unit, modelValue) && m_baseError >= 0.0 && qIsFinite(m_baseError);
+                samplingBaseUnit(variable.m_varName), unit, modelValue) && qIsFinite(modelY) && (!errorAxis || modelY >= 0.0);
     for(size_t i = 0; i < m_results->m_runs.size(); ++i) {
         const nmDataSensitive::RunResult& run = m_results->m_runs[i];
-        if(!hasSensitivityError(run) || index >= run.m_vecParameters.size()) continue;
+        double ordinateValue;
+        if(!m_results->statisticValue(run, ordinate, ordinateValue) || index >= run.m_vecParameters.size()) continue;
         double value = run.m_vecParameters[index];
         if(!convertSamplingValue(variable.m_varName, value, samplingBaseUnit(variable.m_varName), unit, value)) continue;
-        m_points.append(QPointF(value, run.m_dPressureRmse));
+        m_points.append(QPointF(value, ordinateValue));
         m_runRows.append(int(i));
     }
     QVector<iCurveDesc> curves;
     for(int i = 0; i < 4; ++i) {
+        if(!errorAxis && (i == 1 || i == 3)) continue;
         iCurveDesc curve;
         curve.m_sCurveName = i == 0 ? tr("sensitivity data") : i == 1 ? tr("Below threshold")
                 : i == 2 ? tr("reference") : tr("Threshold");
@@ -570,7 +592,7 @@ void nmWxSensitivityScatter::updateParameter()
         }
         if(i == 2 && hasModel) {
             curve.m_vecX.append(modelValue);
-            curve.m_vecY.append(m_baseError);
+            curve.m_vecY.append(modelY);
         }
         if(i == 3) {
             curve.m_bLineVisible = true;
@@ -589,30 +611,41 @@ void nmWxSensitivityScatter::updateParameter()
     m_plot->runUpdate();
     m_plot->setReadOnly(true);
     double left = hasModel ? modelValue : m_points.isEmpty() ? 0.0 : m_points.first().x();
-    double right = left, top = hasModel ? m_baseError : 0.0;
+    double right = left;
+    double bottom = hasModel ? modelY : m_points.isEmpty() ? 0.0 : m_points.first().y();
+    double top = bottom;
     for(int i = 0; i < m_points.size(); ++i) {
         left = qMin(left, m_points[i].x());
         right = qMax(right, m_points[i].x());
+        bottom = qMin(bottom, m_points[i].y());
         top = qMax(top, m_points[i].y());
     }
     const double padding = qMax(1e-6, (right > left ? right - left : qMax(1.0, qAbs(left))) * 0.05);
     left -= padding;
     right += padding;
-    top = qMax(1e-6, qMax(top, m_threshold) * 1.1);
+    if(errorAxis) {
+        bottom = 0.0;
+        top = qMax(1e-6, qMax(top, m_threshold) * 1.1);
+    } else {
+        const double margin = qMax(1e-6, (top > bottom ? top - bottom : qMax(1.0, qAbs(top))) * 0.05);
+        bottom -= margin;
+        top += margin;
+    }
     VecDouble ticksX, ticksY;
     for(int i = 0; i <= 5; ++i) {
         ticksX.append(left + i * (right - left) / 5.0);
-        ticksY.append(i * top / 5.0);
+        ticksY.append(bottom + i * (top - bottom) / 5.0);
     }
     m_plot->setAxisScaleX(left, right, ticksX);
-    m_plot->setAxisScaleY(0.0, top, ticksY);
-    updateThreshold();
+    m_plot->setAxisScaleY(bottom, top, ticksY);
+    updateLegend();
+    if(errorAxis) updateThreshold();
 }
 
 void nmWxSensitivityScatter::updateThreshold()
 {
     // 阈值以下的点填黑，并同步图例中的阈值显示。
-    if(m_parameter->currentIndex() < 0) return;
+    if(m_parameter->currentIndex() < 0 || m_ordinate->currentIndex() != 0) return;
     // 拖动时只更新图内反馈，松开后再保存阈值并刷新结果面板。
     if(!m_draggingThreshold && m_results->m_dFilterThreshold != m_threshold) {
         m_results->m_dFilterThreshold = m_threshold;
@@ -669,6 +702,7 @@ void nmWxSensitivityScatter::resetThreshold()
 void nmWxSensitivityScatter::adjustContextMenu(QMenu* menu)
 {
     // 复用绘图右键菜单，保留框架的缩放、复制与导出入口。
+    if(m_ordinate->currentIndex() != 0) return;
     menu->addAction(m_resultsInPlot);
     menu->addAction(tr("Reset line"), this, SLOT(resetThreshold()));
 }
@@ -678,7 +712,7 @@ void nmWxSensitivityScatter::updateLegend()
     // 勾选后在原有两项图例下追加阈值，拖线和重置时实时更新。
     QString text = "<table cellspacing='2'><tr><td>&#9633;</td><td>" + tr("sensitivity data")
             + "</td></tr><tr><td style='color:blue'>&#9632;</td><td>" + tr("reference") + "</td></tr></table>";
-    if(m_resultsInPlot->isChecked())
+    if(m_ordinate->currentIndex() == 0 && m_resultsInPlot->isChecked())
         text += "<div>" + tr("Threshold: %1 MPa").arg(m_threshold, 0, 'g', 6) + "</div>";
     m_legend->setText(text);
     m_legend->adjustSize();
@@ -693,7 +727,8 @@ bool nmWxSensitivityScatter::eventFilter(QObject* object, QEvent* event)
     if(event->type() == QEvent::Paint)
         m_legend->move(view->mapFromScene(m_plot->m_pPlot->getInnerRectF().topLeft()).toPoint() + QPoint(8, 8));
     // 只接管阈值线附近的左键拖动，其余缩放、右键及导出仍由框架处理。
-    if(event->type() == QEvent::MouseButtonPress || event->type() == QEvent::MouseMove || event->type() == QEvent::MouseButtonRelease) {
+    if(m_ordinate->currentIndex() == 0 && (event->type() == QEvent::MouseButtonPress ||
+       event->type() == QEvent::MouseMove || event->type() == QEvent::MouseButtonRelease)) {
         QMouseEvent* mouse = static_cast<QMouseEvent*>(event);
         const QPointF scene = view->mapToScene(QPointF(mouse->pos()));
         const double lineY = view->mapFromScene(QPointF(scene.x(), axisY->getPosForValue(m_threshold))).y();
@@ -713,9 +748,11 @@ bool nmWxSensitivityScatter::eventFilter(QObject* object, QEvent* event)
             const QPointF position = view->mapFromScene(QPointF(m_plot->getAxisOf(0)->getPosForValue(m_points[i].x()), axisY->getPosForValue(m_points[i].y())));
             if(qAbs(position.x() - help->pos().x()) > 6.0 || qAbs(position.y() - help->pos().y()) > 6.0) continue;
             const nmDataSensitive::RunResult& run = m_results->m_runs[m_runRows[i]];
-            QToolTip::showText(help->globalPos(), tr("Run %1\n%2\nGoodness of fit: %3 MPa")
-                        .arg(run.m_nRunIndex + 1).arg(sensitivityRunParameters(*m_results, run))
-                        .arg(run.m_dPressureRmse, 0, 'g', 6), view->viewport());
+            const int ordinate = m_ordinate->itemData(m_ordinate->currentIndex()).toInt();
+            const QString unit = ordinate < 0 ? "MPa" : m_results->m_fitParameters[ordinate].getUnit();
+            QToolTip::showText(help->globalPos(), tr("Run %1\n%2").arg(run.m_nRunIndex + 1)
+                        .arg(sensitivityRunParameters(*m_results, run)) + "\n" + m_ordinate->currentText()
+                        + ": " + QString::number(m_points[i].y(), 'g', 6) + " " + unit, view->viewport());
             return true;
         }
         QToolTip::hideText();