基于GGInformer模型的金融数据特征提取及价格预测  

Feature Extraction and Price Prediction of Financial Data Based on GGInformer Model

作　　者：任晟岐 宋伟[1] REN Shengqi;SONG Wei(School of Computer and Artificial Intelligence,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学计算机与人工智能学院,河南郑州450001

出　　处：《郑州大学学报（理学版）》2024年第5期62-70,共9页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家高能物理科学数据中心开放课题(HT-HEPS-T7-01050200-21-0008);河南省科技攻关计划国际合作项目(172102410065);河南省高等学校重点科研项目(22A520010)。

摘　　要：为了解决金融时序预测任务中出现的特征参数冗余问题,用遗传算法对金融数据进行特征提取,通过三组对比实验进行验证分析。实验结果显示,加入了遗传算法的预测模型比未加入遗传算法的模型在三种数据集上的MSE均有所降低。最终结果证明遗传算法可以有效解决金融产品价格预测过程中的特征冗余问题。为了解决非线性的长序列金融数据预测效果差的问题,通过结合GRU网络和Informer模型构建了GGInformer模型来对金融产品价格进行预测。模型在三种外汇产品数据集上与其他四种预测基准方法进行了对比实验,实验结果与可视化分析表明,所提模型在金融产品交易价格的预测结果上有明显优势,可以提高预测的精度。In order to solve the problem of redundant feature parameters in financial time series prediction tasks,genetic algorithm was selected to extract features from financial data.Three sets of comparative ex-periments were conducted to verify and analyze the results.The experimental results showed that the pre-diction model with genetic algorithm added had lower MSE results than the model without genetic algo-rithm on three datasets.The final results showed that genetic algorithms could effectively solve the prob-lem of feature redundancy in the process of predicting financial product prices.In order to solve the prob-lem of poor prediction performance of nonlinear long sequence financial data,a GGInformer model was constructed by combining GRU network and Informer model to predict financial product prices.The mod-el was compared with four prediction benchmark methods on three foreign exchange product datasets.The experimental results and visual analysis showed that the model had significant advantages over other pre-diction models in predicting the trading prices of financial products,and could improve the accuracy of prediction.

关 键 词：遗传算法 特征提取 金融产品价格预测 Informer模型 GRU网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]