基于動(dòng)態(tài)提示池的股票趨勢(shì)預(yù)測(cè)終身學(xué)習(xí)算法

打開(kāi)文本圖片集

中圖分類號(hào)：TN919-34；TP391 文獻(xiàn)標(biāo)識(shí)碼：A 文章編號(hào)：1004-373X（2025）10-0063-08

DOI：10.16652/j.issn.1004-373x.2025.10.011 引用格式：，，，等.基于動(dòng)態(tài)提示池的股票趨勢(shì)預(yù)測(cè)終身學(xué)習(xí)算法[J].現(xiàn)代電子技術(shù)，2025，48（10）：63-70.

Abstract：Stockdatabelongs tostreamingdatawithadistributionthatchangesovertime，making itextremelychallenging topredictstocktrends.Existingforecastingmethodsadapttothelatestdatadistributionbyretrainingmodelsonarollngbasis， neglectingrepetitivepaternsinhistoricaldata，resultingincatastrophicforgetingandadecreaseinmodelprediction performance.Toaddressaboveissue，aPoolTrainalgorithmisproposed.Inthisalgorithm，theknowledge learnedfromeach retrainingofthemodelisstoredinadynamichintpool，alowingittorememberoldknowledgewhilelearningnewtasks. Acordingtotheknowledgeinthedynamicselectioncombinationhintpol，thecommon hintscancompletediferentdata distribution tasks.TheexperimentalresultsontheCSI3OOdatasetshowthat，incomparisonwiththecurrentoptimalalgorithm DDG-DA，thePooTrainalgorithmcanimprovetheinformationcoefficent （IC），informationcoeffientratio（ICIR），rank informationcoefcient（RankIC），andrankinformationcoeffcientratio（RankICIR）by11.5%，11.41%，0.2%，and34.69%， respectively.Itshowsthattheproposedalgorithmcanrealizebeterresultsinpredictingstock trends，providingvaluable reference information for investors.

Keywords：stock trendprediction；dynamiccuepol;lifelong learning;rollingtraining;corelationcoefficent;information coefficient

0 引言

股票趨勢(shì)反映著國(guó)家的宏觀經(jīng)濟(jì)政策，影響著投資者的經(jīng)濟(jì)利益，預(yù)測(cè)股票趨勢(shì)已成為機(jī)器學(xué)習(xí)領(lǐng)域的研究熱點(diǎn)之一]。（剩余13418字）