核学习自适应预测控制器的在线更新方法

针对非线性过程控制器的设计问题,将基于稀疏核学习的一种具有解析形式的自适应预测控制算法与选择性递推核学习相结合.该在线核学习模型可以通过递推算法进行节点增长和删减的有效更新.因此,所提出的控制器复杂度可控,且能学习过程的时变等特性,从而获得更好的性能.通过一非线性时变过程的仿真研究,验证了所提出的核学习控制器较传统的PID和无在线更新的核学习控制器等具有更好的自适应能力和鲁棒性.

Online adaptation of kernel learning adaptive predictive controller

To design controllers for nonlinear processes, a sparse kernel learning adaptive predictive controller with an analytical form is extended to the updated form using the selective recursive kernel learning method. The online kernel learning model can be efficiently updated with node increment and decrement via recursive learning algorithms. Consequently, the proposed kernel controller can restrict its complexity and adaptively trace the time-varying characteristics of a process to achieve better performance. Simulation of the proposed kernel controller for a nonlinear time-varying process is performed. In comparing with the traditional PID controller and the related kernel controller without online updating, this controller exhibits more satisfactory adaptation and robustness.