现代电子技术2025，Vol.48Issue(10)：63-70,8.DOI:10.16652/j.issn.1004-373x.2025.10.011
基于动态提示池的股票趋势预测终身学习算法
Stock trend forecasting lifelong learning algorithm based on dynamic prompt pool
摘要
Abstract
Stock data belongs to streaming data with a distribution that changes over time,making it extremely challenging to predict stock trends.Existing forecasting methods adapt to the latest data distribution by retraining models on a rolling basis,neglecting repetitive patterns in historical data,resulting in catastrophic forgetting and a decrease in model prediction performance.To address above issue,a PoolTrain algorithm is proposed.In this algorithm,the knowledge learned from each retraining of the model is stored in a dynamic hint pool,allowing it to remember old knowledge while learning new tasks.According to the knowledge in the dynamic selection combination hint pool,the common hints can complete different data distribution tasks.The experimental results on the CSI300 dataset show that,in comparison with the current optimal algorithm DDG-DA,the PoolTrain algorithm can improve the information coefficient(IC),information coefficient ratio(ICIR),rank information coefficient(Rank IC),and rank information coefficient ratio(Rank ICIR)by 11.5%,11.41%,0.2%,and 34.69%,respectively.It shows that the proposed algorithm can realize better results in predicting stock trends,providing valuable reference information for investors.关键词
股票趋势预测/动态提示池/终身学习/滚动训练/相关系数/信息系数Key words
stock trend prediction/dynamic cue pool/lifelong learning/rolling training/correlation coefficient/information coefficient分类
信息技术与安全科学引用本文复制引用
周文瑞,孟林建,綦小龙,刘艳芳,乎西旦·居马洪,林玲..基于动态提示池的股票趋势预测终身学习算法[J].现代电子技术,2025,48(10):63-70,8.基金项目
伊犁师范大学校级重点项目(2023YSZD006) （2023YSZD006）
伊犁师范大学重大专项(2024ZDZX004) （2024ZDZX004）
新疆维吾尔自治区自然科学基金项目(2021D01C466) （2021D01C466）
国家自然科学基金项目(62266046) （62266046）
