import gc import os import numpy as np import pandas as pd import matplotlib.pyplot as plt from statsmodels.tsa.stattools import acf, pacf from statsmodels.tsa.seasonal import seasonal_decompose from scipy.signal import spectrogram, periodogram def generate_time_series_plots(self): """生成时间序列图""" try: self._log_step("Generating time series plots...") if not hasattr(self, 'data') or self.data is None or len(self.data.columns) == 0: self._log_step("No data available for time series plots", "warning") return None, "No data available", None # 准备数据 n_plots = min(4, len(self.data.columns)) plot_data = self.data.iloc[:, :n_plots].reset_index() # 将 timestamp 转为字符串，确保JSON可序列化 if 'timestamp' in plot_data.columns: plot_data['timestamp'] = plot_data['timestamp'].astype(str) summary = f"Generated {n_plots} time series charts" # charts 模式：仅返回数据，不生成图片；保留绘图版在下方注释 self._log_step("Time series data prepared", "success") return None, summary, plot_data # --- 绘图版保留参考 --- # fig, axes = plt.subplots(2, 2, figsize=(10, 8), dpi=100) # fig.suptitle('Time Series Analysis', fontsize=14) # axes = axes.flatten() # for i in range(n_plots): # try: # col = self.data.columns[i] # axes[i].plot(self.data.index, self.data[col], linewidth=1) # axes[i].set_title(f'{col}') # axes[i].tick_params(axis='x', rotation=45) # axes[i].grid(True, alpha=0.3) # except Exception as plot_err: # self._log_step(f"Plot {col} error: {plot_err}", "warning") # for i in range(n_plots, len(axes)): # fig.delaxes(axes[i]) # plt.tight_layout() # img_path = os.path.join(self.temp_dir.name, 'time_series.png') # plt.savefig(img_path, dpi=100, bbox_inches='tight', format='png') # plt.close(fig) # self._log_step("Time series plots generated", "success") # return img_path, summary, plot_data except Exception as e: self._log_step(f"Time series plots failed: {str(e)[:100]}", "error") return None, f"Error: {e}", None def generate_acf_pacf_plots(self): """生成自相关和偏自相关图""" try: self._log_step("Generating ACF and PACF plots...") if hasattr(self, 'data') and self.data is not None: numeric_cols = self.data.select_dtypes(include=[np.number]).columns n_cols = min(3, len(numeric_cols)) # 计算ACF和PACF数据 acf_pacf_results = {} for col in numeric_cols[:n_cols]: series = self.data[col].dropna() try: acf_vals = np.asarray(acf(series, nlags=min(40, len(series) // 4))) pacf_vals = np.asarray(pacf(series, nlags=min(20, len(series) // 5))) acf_pacf_results[col] = { 'acf': acf_vals.tolist(), 'pacf': pacf_vals.tolist(), } except Exception as e: self._log_step(f"Error calculating ACF/PACF for {col}: {e}", "warning") summary = f"生成 {n_cols} 个变量的ACF和PACF数据" self._log_step("ACF and PACF data generated", "success") return None, summary, acf_pacf_results # --- 绘图版保留参考 --- # fig, axes = plt.subplots(n_cols, 2, figsize=(12, 4 * n_cols)) # fig.suptitle('自相关和偏自相关分析', fontsize=16) # if n_cols == 1: # axes = axes.reshape(1, -1) # for i, col in enumerate(numeric_cols[:n_cols]): # series = self.data[col].dropna() # plot_acf(series, ax=axes[i, 0], lags=min(40, len(series) // 4)) # axes[i, 0].set_title(f'{col} - 自相关函数 (ACF)') # plot_pacf(series, ax=axes[i, 1], lags=min(20, len(series) // 5)) # axes[i, 1].set_title(f'{col} - 偏自相关函数 (PACF)') # plt.tight_layout() # img_path = os.path.join(self.temp_dir.name, 'acf_pacf_plots.png') # plt.savefig(img_path, dpi=150, bbox_inches='tight') # plt.close() # self._log_step("ACF and PACF plots generated", "success") # return img_path, f"生成 {n_cols} 个变量的ACF和PACF图", acf_pacf_results self._log_step("No data available for ACF/PACF plots", "warning") return None, "数据不足，无法生成ACF/PACF图", None except Exception as e: self._log_step(f"ACF/PACF plots failed: {e}", "error") return None, f"ACF/PACF图生成失败: {e}", None def perform_seasonal_decomposition(self): """执行季节性分解""" try: self._log_step("Performing seasonal decomposition...") if hasattr(self, 'data') and self.data is not None: numeric_cols = self.data.select_dtypes(include=[np.number]).columns # 选择第一个数值列进行分解 if len(numeric_cols) > 0: col = numeric_cols[0] series = self.data[col].dropna() # 季节性分解 result = seasonal_decompose(series, model='additive', period=min(24, len(series) // 2)) decomposition_data = pd.DataFrame({ 'observed': result.observed, 'trend': result.trend, 'seasonal': result.seasonal, 'resid': result.resid, }) # 填充NaN以确保JSON序列化 decomposition_data = decomposition_data.astype(object).where( pd.notnull(decomposition_data), None, # type: ignore[arg-type] ) summary = f"季节性分解完成，变量: {col}" self._log_step("Seasonal decomposition completed (data only)", "success") return None, summary, decomposition_data # --- 绘图版保留参考 --- # fig, axes = plt.subplots(4, 1, figsize=(12, 10)) # fig.suptitle(f'{col} - 季节性分解', fontsize=16) # result.observed.plot(ax=axes[0], title='原始序列') # result.trend.plot(ax=axes[1], title='趋势成分') # result.seasonal.plot(ax=axes[2], title='季节成分') # result.resid.plot(ax=axes[3], title='残差成分') # for ax in axes: # ax.tick_params(axis='x', rotation=45) # ax.grid(True, alpha=0.3) # plt.tight_layout() # img_path = os.path.join(self.temp_dir.name, 'seasonal_decomposition.png') # plt.savefig(img_path, dpi=150, bbox_inches='tight') # plt.close() # self._log_step("Seasonal decomposition completed", "success") # return img_path, summary, decomposition_data self._log_step("No numeric columns for decomposition", "warning") return None, "没有数值列可用于季节性分解", None self._log_step("No data available for seasonal decomposition", "warning") return None, "数据不足，无法进行季节性分解", None except Exception as e: self._log_step(f"Seasonal decomposition failed: {e}", "error") return None, f"季节性分解失败: {e}", None def perform_spectral_analysis(self): """执行频谱分析""" try: self._log_step("Performing spectral analysis...") if hasattr(self, 'data') and self.data is not None: numeric_cols = self.data.select_dtypes(include=[np.number]).columns # 计算频谱数据（简化输出，避免数据量过大） spectral_results = {} for col in numeric_cols[:2]: try: series = self.data[col].dropna().values f, t, Sxx = spectrogram(series, fs=1.0, nperseg=min(256, len(series) // 4)) f_p, Pxx_den = periodogram(series, fs=1.0) # 仅保留频谱的均值和形状，避免返回完整矩阵 Sxx_log = 10 * np.log10(Sxx + 1e-12) spectral_results[col] = { 'spectrogram': { 'f': f.tolist(), 't': t.tolist(), 'Sxx_log10_mean': float(np.mean(Sxx_log)), 'Sxx_shape': Sxx.shape, }, 'periodogram': { 'f': f_p.tolist()[:20], 'Pxx_den': Pxx_den.tolist()[:20], }, } except Exception as e: self._log_step(f"Spectral calc failed for {col}: {e}", "warning") summary = "Spectral analysis completed" self._log_step("Spectral analysis completed (data only)", "success") return None, summary, spectral_results # --- 绘图版保留参考 --- # n_cols = min(2, len(numeric_cols)) # fig, axes = plt.subplots(n_cols, 2, figsize=(15, 5 * n_cols)) # fig.suptitle('频谱分析', fontsize=16) # if n_cols == 1: # axes = axes.reshape(1, -1) # for i, col in enumerate(numeric_cols[:n_cols]): # series = self.data[col].dropna().values # f, t, Sxx = spectrogram(series, fs=1.0, nperseg=min(256, len(series) // 4)) # axes[i, 0].pcolormesh(t, f, 10 * np.log10(Sxx), shading='gouraud') # axes[i, 0].set_title(f'{col} - 频谱图') # axes[i, 0].set_ylabel('频率 [Hz]') # axes[i, 0].set_xlabel('时间') # f, Pxx_den = periodogram(series, fs=1.0) # axes[i, 1].semilogy(f, Pxx_den) # axes[i, 1].set_title(f'{col} - 周期图') # axes[i, 1].set_xlabel('频率 [Hz]') # axes[i, 1].set_ylabel('PSD [V**2/Hz]') # axes[i, 1].grid(True, alpha=0.3) # plt.tight_layout() # img_path = os.path.join(self.temp_dir.name, 'spectral_analysis.png') # plt.savefig(img_path, dpi=150, bbox_inches='tight') # plt.close() # self._log_step("Spectral analysis completed", "success") # return img_path, f"频谱分析完成，分析了 {n_cols} 个变量", spectral_results self._log_step("No data available for spectral analysis", "warning") return None, "数据不足，无法进行频谱分析", None except Exception as e: self._log_step(f"Spectral analysis failed: {e}", "error") return None, f"频谱分析失败: {e}", None