基于注意力机制的CNN-LSTM短期股票价格预测

CNN-LSTM Short-term Stock Price Prediction Based on Attention Mechanism

根据股价存在的高频性、长记忆性及不确定性,文章给出了基于注意力机制的卷积神经网络一长短期记忆网络股票价格序列预测方法。首先使用CNN来对数据序列进行卷积操作,以提取其特征分量。然后,利用长短期记忆网络(LSTM)对所抽取出的特征分量做序列预测。最终,注意力模块通过神经网络来自动化拟合权重分配,并对LSTM各个时间节点的隐含层输出向量与对应的权重相乘并求和,为重要的特征分量赋予更大的权重,以此作为模型最终的特征表达。

According to the high frequency,long-term memory and uncertainty of stock price,this paper presents a prediction method of stock price series based on convolutional neural network and long-term and short-term memory network based on attention mechanism.Firstly,CNN is used to convolve the data sequence to extract its feature components.Then,the long-term and short-term memory network(LSTM)is used to predict the extracted feature components.Finally,the attention module automatically assigns fitting weights through neural network,multiplies and sums the output vectors of the hidden layer of each time node of LSTM with the corresponding weights,and gives more weights to the important feature components,which is the final feature expression of the model.