tcn时空卷积神经网络数据预测

前言

本文主要是时空卷积神经网络（Temporal convolutional network，TCN）进行时序数据分析，预测

代码

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from tcn.tcn import TCN
"""
pip3 install keras-tcn
"""
filepath = '12.csv'
data = pd.read_csv(filepath)
dataset_train = pd.read_csv('12.csv',usecols=[1])

#  利用Keras创建TCN模型
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
# 初始化顺序模型
regressor = Sequential()
regressor.add(TCN(nb_filters=10, input_shape=(3, 1),kernel_size= 4, dilations=[1, 2, 4, 8]))  # ,return_sequences=True
# 定义线性的输出层
regressor.add(Dense(units = 1))
# 模型编译：定义优化算法adam， 目标函数均方根MSE
regressor.compile(optimizer = 'adam', loss = 'mean_squared_error')

# 模型训练
history = regressor.fit(X_train, y_train, epochs = 100, batch_size = 100, validation_split=0.001)
regressor.summary()
# 模型预测
predicted_stock_price = regressor.predict(X_test)

from sklearn import metrics
y = real_stock_price
y_hat = predicted_stock_price
MSE = metrics.mean_squared_error(y, y_hat)
RMSE = metrics.mean_squared_error(y, y_hat)**0.5
MAE = metrics.mean_absolute_error(y, y_hat)
MAPE = metrics.mean_absolute_percentage_error(y, y_hat)

print("Test RMSE:",RMSE)
print("Test MSE:",MSE)
print("Test MAE:",MAE)
print("Test MAPE:",MAPE)

损失函数图像
预测数据分析图像

Test RMSE: 1.3329355073539122
Test MSE: 1.7767171
Test MAE: 0.76241606
Test MAPE: 0.0063908156

总结

• 使用tcn对数据进行预测，可以和lstm等模型进行比较分析
• tcn也可以进一步优化参数

备注:

需要源代码和数据集,或者想要沟通交流,请私聊,谢谢.