稀疏在线无偏置最小二乘支持向量机的预测控制  

Predictive control using sparse online non-bias LSSVM

作　　者：周欣然[1,2] 滕召胜[2] 蒋星军[3] 

机构地区：[1]中南大学信息科学与工程学院,长沙410075 [2]湖南大学电气与信息工程学院,长沙410082 [3]湖南广播电视大学计算机系,长沙410004

出　　处：《电子测量与仪器学报》2011年第4期331-337,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(编号:60872128)资助项目;湖南省科学技术厅(编号:2009FJ3077)资助项目

摘　　要：针对非线性预测控制中的预测模型,设计了稀疏在线无偏置最小二乘支持向量机(SONB-LSSVM),并提出了基于SONB-LSSVM的有约束单步预测控制算法。在每个控制周期,该SONB-LSSVM递推地学习新样本,并删除贡献最小样本。该样本删除技巧能提高学习样本集的多样性和代表性;与ONB-LSSVM相比,SONB-LSSVM的泛化性能受输入信号频率影响较小。控制量由Brent优化方法计算。由于SONB-LSSVM能及时学习过程动态新特性,该预测控制方法具有良好的自适应能力.液位控制仿真表明,在多种波形的期望输出并有扰动情况下该预测控制方法都是有效的。Aiming at predicting model of nonlinear predictive control,a sparse online non-bias least square support vector machine（SONB-LSSVM） is designed,and a constrained single-step-ahead predictive control（PC） is proposed utiliz-ing SONB-LSSVM.During per controlling period,the SONB-LSSVM studies new sample and removes the least important one recursively.The skill for deleting sample can improve diversity and representative capacity of the training sample set;generalization of SONB-LSSVM is less affected by the input signal frequency compared with ONB-LSSVM.The control values are computed via Brent optimization method.Because SONB-LSSVM can study new dynamic properties of process in time,the predictive control strategy possesses excellent adaptation.Simulation results of liquid-level process control show the validity of the predictive control in various waveform expected output case with disturbance existing.

关 键 词：预测控制 非线性系统 最小二乘支持向量机 学习算法 稀疏性 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP18[自动化与计算机技术—控制科学与工程]