基于注意力机制的LSTM多因子股指预测研究

A Study of LSTM Multi-Factor Stock Index Prediction Based on Attention Mechanism

摘要: 股票价格指数是根据选定的股票样本，通过加权计算得出反映股市整体走势的指标。本文使用结合注意力机制的LSTM模型对股票价格指数进行建模和预测。LSTM是循环神经网络的一种变体，能够有效地处理长期依赖关系。注意力机制模仿了人类视觉注意力的功能，处理信息时能够动态地聚焦于最为重要的部分。首先，将原始数据归一化处理，划分为训练集和测试集。再将训练集数据转换为监督型数据并用来训练模型。最后，利用训练好的模型进行预测并与测试集数据进行对比，以评估模型的效果。研究结果表明，基于注意力机制的LSTM多因子预测模型的MAE为35.3749，RMSE为43.4630。相对于LSTM单因子模型和基于注意力机制的LSTM单因子预测模型，MAE分别降低了18.49%和61.47%，RMSE分别降低了10.55%和55.19%。

Abstract: A stock price index is an indicator that reflects the overall trend of the stock market based on a weighted calculation of a selected sample of stocks. In this paper, the stock price index is modelled and predicted using an LSTM model incorporating an attention mechanism, a variant of recurrent neural networks that can effectively handle long-term dependencies. The attention mechanism mimics the function of human visual attention and is able to dynamically focus on the most important parts when processing information. Firstly, the raw data is normalised and divided into a training set and a test set. The training set data is then converted to supervised data and used to pair train the model. Finally, the trained model is used to make predictions and compared with the test set data to evaluate the effectiveness of the model. The results of the study show that the MAE of the LSTM multifactor prediction model based on the attention mechanism is 35.3749 and the RMSE is 43.4630. The MAE is reduced by 18.49% and 61.47%, and the RMSE is reduced by 10.55% and 55.19%, respectively, with respect to the LSTM one-factor model and the LSTM one-factor prediction model based on the attention mechanism.