摘要
传统的股票指数预测方法是在含噪声、非平稳以及非线性的原始股指序列数据上实施的,这将导致预测精度的下降。为了解决这个问题,提出了一种基于时频融合卷积神经网络的股指预测方法。首先通过引入变分模态分解(VMD)将原始序列数据分解到频域特征上,使得分解后的股指数据具有低信噪比,同时具有更明显的趋势性和平稳性。进一步结合时序卷积神经网络(TCN),构建了时频融合的卷积神经网络模型。最后在6个实际数据集上与8个基准方法进行比较,实验结果表明该方法具有更高的预测精度和更好的解释性。
The traditional stock index forecasting methods were conducted on the noisy,non-stationary and non-linear original stock index time series data,which would degrade the prediction accuracy.In order to deal with this issue,a novel stock index prediction method was proposed by incorporating the time-frequency features and the convolutional neural network.Firstly,the original time series data were decomposed into time-frequency features by employing the variational mode decomposition(VMD).The decomposed series data had a low signal-to-noise ratio and also stationarity with a clear trend.Then,by combining with temporal convolutional network(TCN),a fusion of time-frequency-based convolutional neural network model was proposed.Finally,compared with eight baseline methods on six real-world datasets,the experimental results showed that our method had higher prediction accuracy and better interpretability.
作者
姜振宇
黄雁勇
李天瑞
蔡福旭
JIANG Zhenyu;HUANG Yanyong;LI Tianrui;CAI Fuxu(School of Statistics, Southwestern University of Finance and Economics, Chengdu 611130, China;School of Computing and Artificial Intelligence, Southwest Jiaotong Univeristy, Chengdu 611756, China;Putian College Affiliated Hospital, Putian 351100, China)
出处
《郑州大学学报(理学版)》
北大核心
2022年第2期81-88,共8页
Journal of Zhengzhou University:Natural Science Edition
基金
教育部人文社会科学青年基金项目(21YJCZH045)
中央高校基本科研业务专项资金项目(JBK2101001)。
关键词
股票指数预测
时频融合
变分模态分解
时序卷积网络
stock market index prediction
fusion of time-frequency
variational mode decomposition
temporal convolutional network