ARIMA滚动预测

import numpy as np
from sklearn import preprocessing
import os
import matplotlib.pyplot as plt
from statsmodels.tsa.arima_model import ARIMA
import pandas as pd
from statsmodels.tsa.arima_model import ARIMAResults
import pandas as pd
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '0'
list_1 = []
list_1 = pd.read_csv(r'C:\Users\Linger\Desktop\data_10-12\data10_10-12.csv')
list_1 = np.array(list_1)
list_1 = list_1.astype('float32')
list_1=np.reshape(list_1,[-1,1])


train = list_1[1344:1536]
test = list_1[1536:2496]
history3 = [x for x in train]
prediction = list()
for i in range(10):
    history3 = list_1[1344+(96*i):(96*i+1536)]
    for t in range(96):
        model = ARIMA(history3, order=(2, 0, 0))
        model_fit = model.fit(disp=0)
        output3 = model_fit.forecast()
        yhat3 = output3[0]
        prediction.append(yhat3)
        obs3 = test[i*96+t]
        history3 = list(history3)
        history3.append(obs3)
        print(t, 'predicted=%f, expected=%f' % (yhat3, test[i*96+t]))
    print("输出predictions信息", prediction)



from sklearn.externals import joblib


#loaded = ARIMAResults.load('model4.pkl')
prediction=np.reshape(prediction,[-1, 1])
test3=np.reshape(test, [-1, 1])
yp = np.hstack((prediction,test3))
py = np.hstack((test3,prediction))

Y = yp[:, 1]
P = py[:, 1]

mape=0
for i in range(len(test)):
    if prediction[i]!=0:
        mape=mape+abs(prediction[i]-test[i])/test[i]
mape=mape/len(test)

print("test MAPE：")
print(mape)
# print("test rmse：")
# print(RMSE)
# print("test mae：")
# print(MAE)

# sess.close()
plt.plot(Y)
plt.plot(P, color='red')
plt.show()