基于小波变换与时序特征提取的股价数据增强方法研究

打开文本图片集

中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2026）05-0083-07

Research on Stock Price Data Augmentation Method Based on Wavelet Transform and Temporal Feature Extraction

HE Yi'， CHEN Dingyi1，2

（1.SchoolofElectricalandInformationEngneering，Hubei UnversityofAutomotiveTechnology，iyan442，Cina; 2.BusinessSchool，TheUniversityofQueensland，Brisbane 4ooo，Australia）

Abstract：This paper proposes astock price data augmentation method based on wavelet tansform and temporal feature extraction toaddresstheproblems ofnoise interferenceandinsuffcientsample size instock marketdata.The methoddecomposes stock price data into low-frequencyand high-frequencycomponents through Discrete Wavelet Transform （DWT），and adds atenuationfactorstotheigh-frequencypartogeneratediversifeddatasamples.Meanhile，itcombinesConvolutionalNeural Network（CNN）and Bidirectional Long Short-Term Memory （BiLSTM） network toeffectivelyextract local features and longtermdependenciesistockpricesequences.ExperimentalverificationonACL18，CMN-US，andCMN-CNatasetssosthat this methodsignificantlyimproves theaccuracyandstabilityof stockpriceprediction，especiallyinscenarios withlargestock market fluctuations.For instance，onthe ACL18 dataset，the accuracy of the StockNet model increases from 58.23% to 59.80% 5 which shows its good application prospect in stock price trend prediction.

Keywords：wavelet transform; CNN; stock price trend prediction;BiLSTM

0 引言

股价趋势预测对投资决策和风险管理具有重要意义。（剩余10922字）