基于多目标演化算法和改进概率分类的重尾时间序列预测  

PREDICTION OF HEAVY TAILED TIME SERIES BASED ON MULTI-OBJECTIVE EVOLUTIONARY ALGORITHM AND IMPROVED PROBABILISTIC CLASSIFICATION

作　　者：邹小云[1] 林文学[2] Zou Xiaoyun;Lin Wenxue(Department of Scientific Research,Hubei Polytechnic Institute,Xiaogan 432000,Hubei,China;School of Extended Education,Hubei Polytechnic Institute,Xiaogan 432000,Hubei,China)

机构地区：[1]湖北职业技术学院科研处,湖北孝感432000 [2]湖北职业技术学院继续教育学院,湖北孝感432000

出　　处：《计算机应用与软件》2020年第12期273-279,共7页Computer Applications and Software

基　　金：湖北省教育厅科学技术研究项目(B2017519);湖北职业技术学院校级课题“教育智能化背景下高职院校应用数学教学模式创新研究与实践”(2019A05)。

摘　　要：金融、通信和气象等领域中高频时间序列的边际分布均为重尾分布,而传统时间序列预测算法大多将数据流考虑为正态分布,导致传统算法无法适用于重尾分布的时间序列。针对这种情况,提出一种基于演化算法和改进概率分类器的重尾时间序列预测算法。将预测提前量和预测准确率作为两个优化目标,利用演化算法对两个目标进行独立优化。对高斯过程分类进行改进,使其支持重尾时间序列的分类问题,并且优化了时间效率。实验结果表明,该算法有效地提高了时间序列的预测准确率。In the financial,communication and meteorological fields,the side distributions of high frequent time series are heavy tailed distributions,but most of the traditional time series prediction algorithms treat the time series as normal distribution,so that they are not suitable for heavy tailed distributed time series.In view of this,we propose a prediction algorithm of time series based on evolutionary algorithm and improved probabilistic classifier.We treated the prediction earliness and prediction accuracy as two optimization objectives and used evolutional algorithm to optimize both of objectives jointly.The classification of Gaussian processes was improved to support the classification of heavy tailed time series and optimize the time efficiency.The experimental results indicate that our algorithm improves the prediction accuracy of time series effectively.

关 键 词：多目标优化 风险预测 重尾分布 时间序列分类 概率分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]