基于多项式核RVM的非线性模型预测控制  

Nonlinear Model Predictive Control Based on RVM with Polynomial Kernel

作　　者：陈佳[1] 颜学峰[1] 钟伟民[1] 钱锋[1] 

机构地区：[1]华东理工大学化学工程联合国家重点实验室,上海200237

出　　处：《控制工程》2008年第2期158-160,195,共4页Control Engineering of China

基　　金：国家杰出青年科学基金资助项目(60625302);国家973计划基金资助项目(2002CB3122000);国家863计划基金资助项目(20060104Z1081);上海市科委重大基础研究基金资助项目(05DJ14002);上海市自然科学基金资助项目(05ZR14038)

摘　　要：针对非线性、不确定性对象不易建模的特点,提出了基于多项式核关联向量机(RVM)的解析型非线性预测控制方法,该方法采用多项式核RVM进行模型辨识,得到的对象模型作为预测模型。由于RVM具有较好的非线性建模能力,弥补了SVM参数设定难和稀疏性不强等弱点;同时,多项式形式的模型表达式使二次型优化目标函数可以通过函数解析方法求得最优控制输入,即简化了滚动优化模块,增强了控制的实时性。通过对一个标准的非线性Benchmark问题进行仿真实验,结果表明该方法具有良好的控制性能。In view of the difficulty in nonlinear and uncertain plant modelling, an analytic nonlinear predictive control method based on poly-nomial kemel relevance vector machine （RVM） is proposed. Predictive model is obtained from identification of plant model with polynomial kernel RVM. RVM can make up the shortcoming of SVM such as parameter setting and sparsity beacuse of its good ability of nonlinear modelling. Also quadratic form objective function can lye solved with a function analitical method because of the polynomial model expression. So the rolling optimization module is simplified and the control real-time performance is strengthened. The predictive control algorithm is demonstrated on a recognized benchmark problem, and the simulation results show that it can be well applied to nonlinear systems with good performance.

关 键 词：多项式核函数 关联向量机 非线性模型 预测控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]