摘要
针对非线性多输入多输出(MIMO)系统的黑箱辨识问题,提出一种基于ε不敏感损失函数的多输出支持向量回归机(SVR)模型,并给出了偏置的有效求取算法.在一个优化问题中,该模型能最小化所有输出带正则项的结构风险总和,并能为不同输出选择不同的核函数及模型参数.将多输出SVR模型应用于非线性MIMO系统的辨识,仿真结果表明,该模型克服了传统支持向量回归机必须为每个输出单独建模这一缺陷,并能提升系统的整体辨识能力.
A multi-output support vector regression machine model based on e insensitivity loss function is proposed for the black-box identificaton of the nonlinear MIMO system, and the effective algorithm for calculating the bias of the regression function is presented. The model can minimize the regularized structure risk summation of all outputs, and it can select different kernel functions and model parameters for different outputs. The model is applied to the nonlinear MIMO system identification. The simulation results show that the model overcomes the traditional SVR＇s limitation of modeling for each output individually, and can improve the integral identification ability of the MIMO system.
出处
《控制与决策》
EI
CSCD
北大核心
2008年第7期813-816,822,共5页
Control and Decision
基金
国家自然科学基金重点课题(60736026)
教育部新世纪优秀人才支持计划项目
