不知道这个效果满不满足你的要求？ #卡尔曼滤波 damping提供了预测的偏差，预测偏差越大，则卡尔曼滤波的修正功能越强，曲线就越平滑 #卡尔曼滤波器的transition_covariance提供了转换矩阵的协方差，这个越大曲线就越接近于真实值 from pykalman import KalmanFilter def kalman_filter(observations,damping=1): observation_covariance = damping initial_value_guess = observations[0] transition_matrix = 1 transition_covariance = 10 kf = KalmanFilter( initial_state_mean=initial_value_guess, initial_state_covariance=observation_covariance, transition_matrices=transition_matrix, observation_covariance=observation_covariance, transition_covariance=transition_covariance, transition_offsets=None) pre, cov = kf.smooth(observations) return pre import matplotlib.pyplot as plt plt.figure(figsize=(20,10)) legend('原始','滤波后') data = attribute_history(stock, N, '1d', ['close']) dr = np.log(data.close/data.close[0]) #对收盘价做一个标准化处理，并求其log,series newdr = kalman_filter(dr,10) # FFT低通滤波 plt.plot(dr.tolist()) plt.plot(newdr.tolist())