时间序列预测入（二）

# -*- coding: utf-8 -*-
"""
Created on Fri Mar 22 21:03:34 2019

@author: Administrator
"""
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from datetime import datetime
from datetime import timedelta

num = 14
filenames = []
basepath = ‘D:\\pworkspace\\data\\Metro_train\\‘
for i in range(1, num+1):
    if i < 10:
        filenames.append(basepath + ‘record_2019-01-0‘ + str(i) + ‘.csv‘)
    else:
        filenames.append(basepath + ‘record_2019-01-‘ + str(i) + ‘.csv‘)

flag = True      
for filename in filenames:
    df = pd.read_csv(filename)
    df[‘time‘] = df[‘time‘].str[:-4] + ‘0:00‘
    df[‘time‘] = pd.to_datetime(df[‘time‘])
    df0 = df[df[‘stationID‘] == 0].copy()
    del df
    user_in = df0[df0[‘status‘] == 1]
    user_out = df0[df0[‘status‘] == 0]
    user_in = user_in.groupby(‘time‘)
    user_out = user_out.groupby(‘time‘)
    user_in = user_in.count()
    user_out = user_out.count()
    user_in[‘count‘] = user_in[‘userID‘]
    user_out[‘count‘] = user_out[‘userID‘]
    user_in = user_in.drop([‘lineID‘, ‘stationID‘, ‘deviceID‘,
                            ‘status‘, ‘payType‘, ‘userID‘], axis=1)
    user_out = user_out.drop([‘lineID‘, ‘stationID‘, ‘deviceID‘,
                            ‘status‘, ‘payType‘, ‘userID‘], axis=1)
    if flag:
        user_in_all = user_in
        #user_out_all = user_out
        flag = False
    else:
        user_in_all = pd.concat([user_in_all,user_in], axis=0)
        #user_out_all = pd.concat([user_out_all,user_out], axis=0)
        

#start = datetime(2019,1,1,0,0,0)
#timelist = [ str(start + timedelta(seconds=600*i)) for i in range(24 * 6 * 2)]

startdate = datetime(2019,1,1,0,0,0)
enddate = startdate + timedelta(days=num-1, minutes=50, hours=23)

all_time_data = pd.DataFrame({‘time‘ : pd.date_range(start=str(startdate), 
                               end=str(enddate),  freq=‘10T‘)})
all_time_data[‘count‘] = 0
all_time_data.index = all_time_data[‘time‘]
all_time_data = all_time_data.drop(‘time‘, axis=1)

user_in_all = pd.merge(all_time_data, user_in_all, right_on=‘time‘, 
         left_index=True, how=‘outer‘)
user_in_all[np.isnan(user_in_all[‘count_y‘])] = 0
user_in_all[‘count_x‘] = user_in_all[‘count_x‘] + user_in_all[‘count_y‘]

user_in_all[‘count‘] = user_in_all[‘count_x‘]
user_in_all = user_in_all.drop([‘count_x‘, ‘count_y‘], axis=1)
user_in_all.plot(figsize=(15,8))
plt.show()

ts = user_in_all[‘count‘]
ts_ewma = pd.DataFrame(ts).ewm(span=60).mean()
ts_ewma.plot(figsize=(15,8))
plt.show()

from statsmodels.tsa.stattools import acf, pacf, adfuller
from statsmodels.stats.diagnostic import acorr_ljungbox
from statsmodels.tsa.arima_model import ARIMA
#import statsmodels.api as sm

ts_diff_1 = ts_ewma.diff(1).dropna(axis=0, how=‘any‘)
ts_diff_1 = ts_diff_1[‘count‘]

# ADF平稳性检验
adfuller(ts_diff_1, autolag=‘AIC‘)
# 白噪声检验
acorr_ljungbox(ts_diff_1, 1)
# ACF PACF
lag_acf = acf(ts_diff_1, nlags=50)
lag_pacf = pacf(ts_diff_1, nlags=50)
plt.figure(facecolor=‘white‘, figsize=(15, 8))
plt.plot(lag_acf)
plt.show()
plt.figure(facecolor=‘white‘, figsize=(15, 8))
plt.plot(lag_pacf)
plt.axhline(y=-1.9/np.sqrt(len(ts_diff_1)), linestyle=‘--‘, color=‘gray‘)
plt.axhline(y=1.9/np.sqrt(len(ts_diff_1)), linestyle=‘--‘, color=‘gray‘)
plt.show()

model = ARIMA(ts_diff_1, order=(6, 0, 0))
ts_predict = model.fit().predict()
rmse = np.sqrt(sum((ts_predict - ts_diff_1)**2) / ts_diff_1.size)

plt.figure(facecolor=‘white‘, figsize=(15, 8))
plt.plot(ts_predict, lw=0.5, color=‘blue‘, label=‘Predict‘)
plt.plot(ts_diff_1, lw=0.5, color=‘red‘, label=‘Original‘)
plt.legend(loc=‘lower right‘)
#plt.ylim((-1000, 1000))
plt.show()