基于决策函数及PSO优化的SVM预测控制应用研究  

Application Study of SVM Predictive Control Based on Decision Functions Simplification and PSO Optimization

作　　者：王杰[1] 陈锴鹏[1] 

机构地区：[1]郑州大学电气工程学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2013年第2期53-56,共4页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(60905039)

摘　　要：SVM处理大样本问题时性能明显不如神经网络,因此笔者利用矩阵变换进行决策函数的简化来提升SVM的训练速度,对SVM建模时非必需的支持向量进行约简,并引入一个松弛变量来提升约简效果.实验证明,约简后支持向量个数减少三分之一以上.SVM所建立的模型进行线性化之后应用于预测控制当中,采用PSO算法来选择最优的SVM参数和计算预测控制的最优控制律.通过对水泥回转窑窑尾烟室温度的数据进行实验仿真,结果表明该方法可以提高系统响应速度,减小系统响应的超调量.For large-scale samples, SVM does not perform as well as neural networks and this paper tries to improve the training speed by simplify the decision functions by matrix transform. We simplify the unnecessary support vectors in SVM modeling and introduce a relaxation factor in order to improve the effects of simplifica- tion. Experiment shows that the number of support vectors is redused by at least one third. Using the model nonlinear model built by SVM after linearizing as the predictive model of predictive control. PSO was used to select the best SVM parameters and computing the optimal control law of predictive control. The method can accelerate the response and shorten the overshoot through a simulation of a cement rotary kiln.

关 键 词：支持向量机 决策函数 粒子群算法 预测控制 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]