基于TensorFlow实现LSTM多变量预测

LSTM基本概念

LSTM预测执行流程

假设时间步长为2，则预测t时刻的label值时，需要依赖t-2时刻和t-1时刻的数据，所以可以根据上述图片中描述的数据进行预测。

LSTM预测代码实现

数据划分代码

import numpy as np
import pandas as pd
from sklearn.preprocessing import MinMaxScaler

# 设置时间步长
n_past = 100

# 读取数据集
data = pd.read_csv("data/data.csv")

# 划分数据集和测试集
test_split = round(len(data)*0.1)
data_training = data[:-test_split]
data_testing = data[-test_split:]

#归一化(此处还可以使用标准化)
scaler = MinMaxScaler(feature_range=(0,1))
df_for_training_scaled = scaler.fit_transform(data_training)
df_for_testing_scaled = scaler.transform(data_testing)

# 按照步长划分数据集
# createXY()函数传入参数为:数据集、预测下一个目标值时查看过去多少条数据
def createXY(dataset,n_past):
    dataX = []
    dataY = []
    for i in range(n_past, len(dataset)):
        dataX.append(dataset[i - n_past:i, 0:dataset.shape[1]])
        dataY.append(dataset[i,-1])
    return np.array(dataX),np.array(dataY)

模型构建代码

# 构建模型
def build_model(optimizer):
    grid_model = Sequential()
    grid_model.add(LSTM(50,input_shape=(n_past,9),return_sequences=True))
    grid_model.add(LSTM(50))
    grid_model.add(Dropout(0.1))
    grid_model.add(Dense(1))
    grid_model.compile(loss = 'mse',optimizer = optimizer)
    return grid_model

# 
grid_model = KerasRegressor(build_fn=build_model,verbose=1,validation_data=(testX,testY))

# 超参数搜索,设置超参数搜索范围区间
parameters = {'batch_size' : [16,32],
              'epochs' : [20,30],
              'optimizer' : ['adam','Adadelta'] }

grid_search  = GridSearchCV(estimator = grid_model,
                            param_grid = parameters,
                            cv = 2)

grid_search = grid_search.fit(trainX,trainY)

grid_search.best_params_
my_model=grid_search.best_estimator_.model

my_model.save('saved_model_LSTM/')

prediction=my_model.predict(testX)

# 反归一化
prediction_copies_array = np.repeat(prediction,9, axis=-1)
pred = scaler.inverse_transform(np.reshape(prediction_copies_array,(len(prediction),9)))[:,-1]

original_copies_array = np.repeat(testY,9, axis=-1)
original = scaler.inverse_transform(np.reshape(original_copies_array,(len(testY),9)))[:,-1]

该模型使用的超参数搜索方法为网格搜索法（Grid Search），读者可综合性能需求选取其他的超参数搜索方法。