Indexed by:期刊论文

First Author:Zhang X.

Correspondence Author:Zhao, J.; Research Center of Information and Control, Dalian University of Technology, Dalian 116024, China; email: zhaoj@dlut.edu.cn

Co-author:Zhao J.,Wang W.

Date of Publication:2010-10-01

Journal:ICIC Express Letters

Included Journals:EI、Scopus

Document Type:J

Volume:4

Issue:5 B

Page Number:1767-1772

ISSN No.:1881803X

Abstract:The key of gas scheduling in modern steel enterprises is to accurately predict gasholder level on-line. Because of the frequent and great fluctuation of level, it is difficult to be precisely predicted by the manual experience or the mechanism model based method. In this study, a class of MKL based on reduced gradient algorithm is extended using LSSVR, and the MKLLSSVR is developed. The MKLLSSVR overcomes the poor generalization of single kernel based LSSVR and the long training process of traditional MKL based SVR, and can rapidly give the optimal linear combination of kernels to base the resulting regressor for better interpretation. The simulation results with the real gasholder level data show the MKLLSSVR can achieve better prediction performance compared to the traditional MKL and single kernel based LSSVR. ? 2010 ICIC International.

Translation or Not:no