全文下载:2012050510
邵伟明, 田学民*
(中国石油大学(华东) 信息与控制工程学院,山东 青岛 266580)
摘要:为提高软测量模型的全局推广能力,提出一种基于快速留一交叉验证法(FLOO-CV)的在线递推最小二乘支持向量机(LSSVM)建模方法。在前向学习过程中,设计一种基于FLOO-CV预报误差的模型更新阈值,该阈值无需人工设定,且能够根据过程特性自适应改变;后向学习时采用FLOO-CV删除对模型整体性能影响最小的冗余样本,最大程度地保留模型的推广性能。工业聚丙烯熔融指数的软测量建模研究表明,该方法能够在提高模型泛化能力的同时,有效降低模型更新频率。
关键词:软测量,在线递推,LSSVM,FLOO-CV
中图分类号:TP 301.6文献标志码:A
Online Recursive LSSVM Modeling Method Based on FLOO-CV
SHAO Wei-ming, TIAN Xue-min
(CollegeofInformationand ControlEngineering,ChinaUniversityofPetroleum,Qingdao266580,China)
Abstract: Aiming at enhancing soft sensor model's generalization ability, an online recursive LSSVM modeling method based on FLOO-CV is presented. In forward learning, an adaptive FLOO-CV prediction error-based threshold without any manual work for updating the model is proposed and FLOO-CV is also utilized in backward learning to delete redundant samples that put minimal influence on the global model, which retains the model's generalization ability maximally. The foregoing scheme is applied to build an industrial polypropylene unit's melt index model. The results indicate that the proposed method can not only improve model's prediction accuracy but also efficiently decrease model′s update frequency.
Key words: soft sensor, online recursive, LSSVM, FLOO-CV
收稿日期:2012-08-04
基金项目: 国家自然科学基金项目(51104175),山东省自然科学基金项目(ZR2011FM014)。
作者简介: 邵伟明(1986—),男,博士研究生.*通信联系人.