基于小波与注意力机制的金融序列BiLSTM预测
BILSTM Prediction of Financial Time Series Based on Wavelet and Attention Mechanism
徐东泽 1肖琴1
作者信息
- 1. 上海应用技术大学理学院,上海 201418
- 折叠
摘要
为了提高金融序列的预测精度,以上证指数的收盘价为例,将小波分析与双向长短期记忆神经网络(BiLSTM)和注意力机制(Attention)相结合,构建新的金融时间序列的预测模型.小波分析能去除时间序列中的杂质,BiLSTM网络能修复上下文信息的遗漏,注意力机制能选出重点.通过与没有小波分析的数据或有小波和LSTM等模型的预测效果做对比,结果表明,小波分析加上Attention机制构成的BiLSTM神经网络模型对金融时间序列数据有很好的预测精度,能够有效的预测金融时间序列的趋势.
Abstract
To improve the prediction accuracy of the financial series,taking the closing price of Shanghai Compos-ite Index as an example,a new prediction model of financial time series was constructed by combining wavelet analysis with BiLSTM and Attention mechanism.Wavelet analysis can remove the impurities in the time series,the BiLSTM network can repair the omission of context information,and the attention mechanism can select the key points.By comparing with the prediction effect of LSTM and other models without wavelet analysis,the result show that the neu-ral network model composed of BiLSTM and Attention has good prediction accuracy for financial time series data and can effectively predict the trend of financial time series.
关键词
小波分解/神经网络/双向长短期记忆神经网络Key words
Wavelet decomposition/Neural network/Bidirectional short-term and long-term memory neural net-work引用本文复制引用
基金项目
上海市自然科学基金(16ZR1447200)
国家自然科学基金青年科学基金(11701379)
出版年
2024
NETL
NSTL
万方数据