Python中如何利用Pandas计算平稳时间序列季节性分量的周期

方案1：完全匹配原有逻辑的无循环实现 #

如果你确认季节性分量是完全精确重复的，直接用向量化操作查找匹配位置即可，无需任何显式循环和条件判断：

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# 原有数据生成代码
seasonal = [-0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726, -0.0012477419628991032, -0.0042654910887713745, -0.006234490214646844, 0.0007106773261453963, 1.1533604530851796e-08, -0.004141904258934777, 0.0006148978972421542, 0.0017480068715999646, 0.0011169491792932956, 0.002641724820318341, 0.005415250461344693, 0.003642109435703726]
indice = pd.date_range("2019-07-31 23:55:00", periods=len(seasonal), freq="T")
seasonal = pd.Series(data=seasonal, index=indice)

# 替换原有循环的核心代码
first_val = seasonal.iloc[0]
match_positions = np.where(seasonal == first_val)[0]
periodo = match_positions[1] - match_positions[0]

print("Periodo: {}".format(periodo))
seasonal.plot()
plt.show()

这个实现和你原有逻辑完全等价，输出结果也完全一致，所有操作都是向量化执行，性能比循环更高。

方案2：通用季节性周期检测（推荐） #

实际生产环境中，序列分解得到的季节性分量往往存在微小的计算误差，不会完全精确重复，此时可以使用时间序列领域的标准方法：自相关系数（ACF）检测周期。自相关系数衡量序列当前值和滞后N阶的值的相关性，季节性周期对应的滞后阶数会出现明显的峰值：

from statsmodels.tsa.stattools import acf

# 计算前24阶的自相关系数（可根据你预期的最大周期调整nlags参数）
acf_values = acf(seasonal, nlags=24)
# 跳过滞后0阶（自相关恒为1），取最大相关对应的滞后阶数即为周期
periodo = np.argmax(acf_values[1:]) + 1
print("Periodo: {}".format(periodo))

这个方法鲁棒性更强，就算后续数据有噪声、分量不是完全相等，也能正确识别出周期。

内容的提问来源于stack exchange，提问作者Johannes C.