Python实现多变量序列堆叠式LSTM模型，并实现未来多时刻预测

在时间序列相关的很多建模工作中，LSTM模型是经常会使用到的，从提出到现在LSTM模型已经有了很多的扩展、变种和应用，今天我们简单地实现基于LSTM模型来对多个变量的数据进行建模预测，在简单地预测中只能做单步预测，这里实现了多步的预测分析。

具体实现如下：

#!usr/bin/env python
#encoding:utf-8
from __future__ import division'''
__Author__:沂水寒城
功能： 基于堆叠式LSTM神经网络的多变量序列预测模型
'''import os
import json
import sys
import math
import time
import keras
import random
import pymysql
import datetime
import platform
import numpy as np
import pandas as pd
from keras.utils import plot_model
from  keras.layers.pooling import MaxPooling1D
from keras.layers import LSTM,Dense,Input,add,Flatten
import matplotlib.pyplot as plt
from keras.models import Sequential,Model
from sklearn.preprocessing import MinMaxScaler,LabelBinarizer,LabelEncoder,Imputer
from keras.layers.core import Activation,Dropout,Dense,Flatten
from sklearn.model_selection import train_test_split
from scipy.stats import pearsonr,spearmanr,kendalltau 
from keras.callbacks import EarlyStopping,ModelCheckpoint,Callback
from keras.layers.normalization import  BatchNormalization  
from keras.optimizers import SGD, Adadelta, Adagrad,RMSprop  
from keras.layers.advanced_activations import PReLU  
from utils import predictFuture
from sklearn.metrics import explained_variance_score, mean_absolute_error, mean_squared_error, r2_score
from keras.layers.convolutional import Conv1D,ZeroPadding1D,Conv2D,MaxPooling2D,AveragePooling2D,Convolution2D,ZeroPadding2D
from keras.layers import Dense, Dropout, Embedding, LSTM, Bidirectional
from keras.layers import TimeDistributeddef sliceWindow(data,step):'''移动滑窗创建数据集'''X,y=[],[]for i in range(0,len(data)-step,1):end=i+steponeX,oney=data[i:end,:],data[end, :]X.append(oneX)y.append(oney)return np.array(X),np.array(y)def dataSplit(dataset,step):'''数据集分割'''datasetX,datasetY=sliceWindow(dataset,step)train_size=int(len(datasetX)*0.70)X_train,y_train=datasetX[0:train_size,:],datasetY[0:train_size,:]X_test,y_test=datasetX[train_size:len(datasetX),:],datasetY[train_size:len(datasetX),:]X_train=X_train.reshape(X_train.shape[0],step,-1)X_test=X_test.reshape(X_test.shape[0],step,-1)print('X_train.shape: ',X_train.shape)print('X_test.shape: ',X_test.shape)print('y_train.shape: ',y_train.shape)print('y_test.shape: ',y_test.shape)return X_train,X_test,y_train,y_testdef seq2seqModel(X,step):'''序列到序列堆叠式LSTM模型'''model=Sequential()model.add(LSTM(256, activation='relu', return_sequences=True,input_shape=(step,X.shape[2])))model.add(LSTM(256, activation='relu'))model.add(Dense(X.shape[2]))model.compile(optimizer='adam', loss='mse')return modelif __name__=='__main__':#数据集加载with open('dataset.txt') as f:data_list=[one.strip().split(',') for one in f.readlines()[1:] if one]dataset=[]for i in range(len(data_list)):dataset.append([float(O) for O in data_list[i][1:]])dataset=np.array(dataset)step=7X_train,X_test,y_train,y_test=dataSplit(dataset,step)model=seq2seqModel(X_train,step)model.fit(X_train,y_train,epochs=50,verbose=0)# test=[#       [30.0,58.0,7.0,24.0,0.9,83.0,103.0],#       [43.0,72.0,6.0,23.0,1.1,85.0,103.0],#       [66.0,105.0,6.0,22.0,1.3,134.0,103.0],#       [54.0,94.0,7.0,27.0,1.1,125.0,103.0],#       [64.0,90.0,6.0,19.0,1.2,127.0,103.0],#       [59.0,92.0,6.0,20.0,1.1,126.0,103.0],#       [61.5,91.0,6.0,19.5,1.15,134.0,103.0]#      ]# #真实值   [38.0,66.0,5.0,17.0,1.2,138.0,103.0]# test=np.array(test)# test=test.reshape((1,step,7))# y_pre=model.predict(test,verbose=0)# print('y_pre: ',y_pre)future=predictFuture(model,dataset,7,step,60)for one in future:print('one: ',one)dataset=[]for i in range(len(data_list)):dataset.append([float(O) for O in data_list[i][1:]])D1=[one[0] for one in dataset]D2=[one[0] for one in future]plt.plot(list(range(len(D1))),D1,label='F1 True')plt.plot(list(range(len(D1),len(D1)+len(D2))),D2,label='F1 Future Predict')plt.legend()plt.title('Data True and Predict Cruve')plt.savefig('demo.png')

我们在代码中提供了单个样本的测试样例，如下所示：