基于生成对抗网络的股票收盘价预测方法  

Stock closing price prediction method based on generative adversarial network

作　　者：彭乾 张龑 PENG Qian;ZHANG Yan(School of Computer Science and Information Engineering,Hubei University,Wuhan 430062,China)

机构地区：[1]湖北大学计算机与信息工程学院,湖北武汉430062

出　　处：《湖北大学学报（自然科学版）》2024年第6期743-753,共11页Journal of Hubei University：Natural Science

基　　金：国家自然科学基金(61977021)资助。

摘　　要：针对股票市场的非线性、不平稳性和数据的复杂性带来的预测挑战,本研究提出一种基于生成对抗网络的改进型股票价格预测模型(MAC-WGAN-GP)。该模型通过融合CNN-BiLSTM模型和多头注意力机制作为生成器,用以更准确地生成股票收盘价预测;而判别器则采用多层卷积神经网络,负责判定生成的收盘价与实际值的差异。为了提高模型的预测性能和稳定性,本研究还结合经验模态分解(EMD)和技术指标,以从原始股票收盘价数据中提取更有效的特征。实验采用了中国银行、工商银行、建设银行和农业银行4个不同的股票数据集进行验证,展示了MAC-WGAN-GP模型在MSE、RMSE、MAE和R2等4个评价指标上相比于基线模型的改进,证明其在股票预测任务中的有效性和高拟合能力。In addressing the predictive challenges arising from the nonlinearity,instability,and complexity of stock market data,this thesis proposed an enhanced stock price prediction model based on Generative Adversarial Networks(GANs),termed MAC-WGAN-GP.This model integrated a CNN-BiLSTM architecture and a multi-head attention mechanism as the generator,aimed at generating more accurate predictions of stock closing prices.Meanwhile,a multi-layer convolutional neural network served as the discriminator,responsible for determining the discrepancy between the generated closing prices and the actual values.To enhance the predictive performance and stability of the model,this study also incorporated empirical mode decomposition(EMD)and technical indicators to extract more effective features from the original stock closing price data.Experiments conducted on four different stock datasets,including those of the Bank of China,Industrial and Commercial Bank of China,China Construction Bank,and Agricultural Bank of China,demonstrated the improvement of the MAC-WGAN-GP model over the baseline model in terms of evaluation metrics such as mean squared error(MSE),root mean squared error(RMSE),mean absolute error(MAE),and R-squared(R2),confirming its effectiveness and high fitting capability in stock prediction tasks.

关 键 词：股票价格预测 生成对抗网络 多头注意力机制 CNN-BiLSTM 经验模态分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]