基于经验模态分解与投资者情绪的长短期记忆网络股票价格涨跌预测模型  

Long short-term memory network model for stock price trend prediction based on empirical mode decomposition and investor sentiment

作　　者：翁晓健 林旭东[1] 赵帅斌 WENG Xiaojian;LIN Xudong;ZHAO Shuaibin(College of Mathematics and Information,South China Agricultural University,Guangzhou Guangdong 510642,China)

机构地区：[1]华南农业大学数学与信息学院,广州510642

出　　处：《计算机应用》2022年第S02期296-301,共6页journal of Computer Applications

摘　　要：针对传统的基于统计学的回归股票预测模型难以表征多个变量之间的关系,预测出的股票价格趋势误差较大,提出一种基于经验模态分解(EMD)与投资者情绪的长短期记忆(LSTM)神经网络股票价格涨跌预测模型。首先,将股票收盘价通过EMD分解得到若干个具有不同时间尺度的局部特征信号的本征模函数(IMF);其次,通过引入改进的股票领域情感词典,对东方财富网股吧的帖子,进行上一个股票交易日收盘后和下一个股票交易日开盘前的投资者情感分析,得到下一个股票交易日的投资者情绪指标;最后,将基础的股票基本行情数据、经过EMD得到的IMF以及投资者情绪指标加入LSTM神经网络预测下一个交易日的股票涨跌。仿真实验结果表明,在2019年1月至2021年9月的牧原股份(002714)股票数据上,与单独使用LSTM模型相比,改进后的LSTM模型的预测准确率提高了12.25个百分点,在预测为涨的F1值和预测为跌的F1值上分别提高了1.2个百分点和25.21个百分点。由此可见,基于EMD与投资者情绪的LSTM股票价格涨跌预测模型有效提高了预测精度,为股票市场的涨跌预测提供了一种有效的实验方法。Aiming at the fact that the traditional regression stock prediction models based on statistics are difficult to characterize the relationship between multiple variables,and the predicted stock price trend has a large error,a Long ShortTerm Memory(LSTM)neural network model for stock price prediction based on Empirical Mode Decomposition(EMD)and investor sentiment was proposed.Firstly,the stock closing price was decomposed through EMD to obtain several Intrinsic Mode Function(IMF)with local characteristic signals of different time scales.Secondly,the posts on the East Money Net Forum after the close of the previous stock trading day and before the opening of the next stock trading day were analyzed by the improved stock domain sentiment dictionary for sentiment analysis.Finally,the basic stock market data,IMF obtained through EMD,and investor sentiment indicators were added to the LSTM neural network to predict the stock price rise and fall in the next trading day.Simulation results show that in the stock data of Muyuan Stock(002714)from January 2019 to September 2021,the improved LSTM model improves the prediction accuracy by 12.25 percentage points,increases the F1value predicted to fall and the F1 value predicted to rise by 1.2 percentage points and 25.21 percentage points respectively compared to using the LSTM model alone.Therefore,the improved LSTM network model can effectively improve the prediction accuracy and provide an effective experimental method for the prediction of the stock market.

关 键 词：股票预测模型 机器学习 投资者情绪 经验模态分解 长短期记忆神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]