def extract_wavelet_features(data, column_name='声波强度 (AE)', scales=np.arange(1, 4), wavelet='cmor'): emr_values = data[column_name].values coefficients, frequencies = pywt.cwt(emr_values, scales, wavelet)
scale_energies = np.sum(np.abs(coefficients)**2, axis=0) scale_means = np.mean(np.abs(coefficients), axis=0) scale_stds = np.std(np.abs(coefficients), axis=0) scale_max = np.max(np.abs(coefficients), axis=0) scale_min = np.min(np.abs(coefficients), axis=0) scale_median = np.median(np.abs(coefficients), axis=0) scale_ptp = np.ptp(np.abs(coefficients), axis=0) total_energy = np.sum(np.abs(coefficients)**2, axis=0) + 1e-10 scale_energy_ratios = np.abs(coefficients)**2 / total_energy features = np.column_stack([scale_energies, scale_means, scale_stds, scale_max, scale_min, scale_median, scale_ptp] + [np.abs(coeff) for coeff in scale_energy_ratios]) feature_names = ['energy', 'mean', 'std', 'max', 'min', 'median', 'ptp'] + [f'energy_ratio_scale_{i+1}' for i in range(scale_energy_ratios.shape[0])] return pd.DataFrame(features, columns=feature_names)
提取小波特征
def process_wavelet_features(data, column_name='声波强度 (AE)', scales=np.arange(1, 4), wavelet='cmor'): wavelet_df = extract_wavelet_features(data, column_name, scales, wavelet) # 合并小波特征和原有DataFrame full_df = pd.concat([data, wavelet_df], axis=1) # 创建时间基特征 full_df['时间 (time)'] = pd.to_datetime(full_df['时间 (time)']) full_df['hour'] = full_df['时间 (time)'].dt.hour full_df['day_of_week'] = full_df['时间 (time)'].dt.dayofweek full_df['rolling_mean'] = full_df[column_name].rolling(window=10, min_periods=1).mean() full_df['rolling_max'] = full_df[column_name].rolling(window=10, min_periods=1).max() full_df['rolling_min'] = full_df[column_name].rolling(window=10, min_periods=1).min() return full_df