ARIMA模型参数选择

• 检查序列是否平稳
  • 若不平稳，使用差分平稳化序列，确定差分阶数d
• ARMA定阶
  • 通过PACF确定AR的阶数p
  • 通过ACF确定MA的阶数q
  • 
• 根据参数p,d,q建立模型ARIMA(p,d,q)

#  ARIMA模型
#  平稳性

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

# 构造时间时间序列
df_obj = pd.DataFrame(np.random.randn(1000, 1), 
                       index=pd.date_range('20170101', periods=1000),
                       columns=['data'])
df_obj['data'] = df_obj['data'].cumsum()
print(df_obj.head())

# 一阶差分处理
df_obj['diff_1'] = df_obj['data'].diff(1)

# 二阶差分处理
df_obj['diff_2'] = df_obj['diff_1'].diff(1)

# 查看图像
df_obj.plot(subplots=True, figsize=(18, 12))

## ACF 和 PACF
from scipy import  stats
import statsmodels.api as sm

sm.graphics.tsa.plot_acf(df_obj['data'], lags=20)
sm.graphics.tsa.plot_pacf(df_obj['data'], lags=20)