在线临近支持向量回归机算法  

Online proximal support vector machine regression

作　　者：吴青 李明明 李飞燕 WU Qing;LI Mingming;LI Feiyan(School of Automation,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)

机构地区：[1]西安邮电大学自动化学院,陕西西安710121

出　　处：《西安邮电大学学报》2022年第5期88-97,共10页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省重点研发计划项目(2022GY-050)。

摘　　要：对于时间序列预测问题,临近支持向量回归机的求解矩阵的规模会随着样本数目的增加而急剧增大。针对这一问题,提出了一种在线临近支持向量机回归算法。利用在线学习来更新样本数据以及核矩阵,通过删除旧样本添加新样本来保证训练样本数目的恒定,以降低存储空间和计算代价。实验结果表明,与已有的相关算法相比,所提算法不仅具有良好的预测精度,而且在进行多步预测时,具有较高的预测效率。For the time series prediction,in order to overcome the problem that scale of the solution matrix of the proximal support vector machine regression will sharply increase with the increase of the number of samples,the online proximal support vector machine regression(OL-PSVR)is proposed.Updating the sample data and the kernel matrix through online learning,removing old samples and adding new samples to guarantee the number of training samples is constant,so as to reduce the storage spaces and computational costs.Experiment results show that the proposed algorithm not only has good prediction accuracy,but also has relatively faster prediction efficiency compared with the existing related algorithms in multi-step prediction.

关 键 词：时间序列预测 多步预测 临近支持向量机 在线学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]