請問可以問你完整的程式碼怎麼打的嗎?
我自己跑出來是長下面那樣

我order也是選(1,0,1)
感覺你的預測結果誤差比較小

import pandas as pd
import matplotlib.pyplot as plt
import statsmodels.api as sm
import statsmodels.tsa as tsa
import numpy as np
import statsmodels.graphics.tsaplots as ts
from sklearn.utils.testing import ignore_warnings
from sklearn.exceptions import ConvergenceWarning
from sklearn.metrics import mean_squared_error


@ignore_warnings(category=ConvergenceWarning)
def resolve(data, test25, ans):
    result_predict = list()
    result_answer = list()
    for i in range(len(test25.values)):
        for j in range(9):
            data.append(float(test25.values[i][j + 2]))
        v = np.array(data)
        #myorder = sm.tsa.arma_order_select_ic(v, max_ar=5, max_ma=4, ic='aic')
        #lag_ar = myorder.aic_min_order[0]
        #lag_ma = myorder.aic_min_order[1]
        lag_ar = 1
        lag_ma = 1
        lmodel = tsa.arima_model.ARIMA(v, order=(lag_ar, 0, lag_ma))
        lfitted = lmodel.fit(display=-1)
        pre = lfitted.predict(len(v) + 1, len(v) + 1)
        answer = float(ans[i][0])
        print('iter: %d, predict: %.3f, answer: %.3f' % (i, pre[0], answer))
        data.append(answer)
        result_predict.append(pre[0])
        result_answer.append(answer)
    return result_predict, result_answer


df = pd.read_csv('hw1/train.csv')

pm25 = df[df['測項'].str.contains('PM2.5')]

data = list()
for x in pm25.values:
    for y in range(24):
        data.append(float(x[y + 3]))

test = pd.read_csv('hw1/test.csv', header=None)
test.columns = ['id', 'type', '0', '1', '2', '3', '4', '5', '6', '7', '8']
test25 = test[test['type'].str.contains('PM2.5')]

ans = pd.read_csv('hw1/ans.csv')
ans.columns = ['id', 'val']
ans.set_index('id', inplace=True)

predict, expect = resolve(data, test25, ans.values)

mse = mean_squared_error(expect, predict)
print('mean square error: %.3f' % mse)

final = pd.DataFrame()

final['original'] = expect
final['predict'] = predict
final.plot(figsize=(20, 10))
plt.show()

這樣mse是53.683，不過得跑好幾分鐘。

好的，我研究看看 謝謝!