基于即时学习的非线性系统自适应PID控制

Adaptive PID control for nonlinear systems based on lazy learning

中文关键词  广义最小方差  即时学习  k矢量近邻  PID控制器

英文关键词  general minimum variance  lazy learning  k-vector nearest neighbors  PID controller

中文摘要

      当使用先进策略整定PID控制器参数时, 往往要依赖于系统所辨识的模型, 而模型的精度与优化算法的计算效率直接影响到系统的控制效果. 本文利用即时学习算法的本质自适应特点(建模数据在时间与空间上相邻性),来提高辨识模型的精度，并基于广义最小方差的性能指标, 用等价多项式的方法, 推导出PID形式的控制律, 从而避免其他优化算法带来的计算量, 提高了控制精度与计算效率. 仿真结果验证了该方法的有效性.

英文摘要

      When applying advanced strategies to determine the parameters of a PID controller, a model need to be identified for the controlled system. The predicted precision and the computation efficiency of the identifying algorithm directly affect the control performance of the system. To improve the precision, lazy learning algorithm, which has a essentially adaptive characteristics(i.e.,in which the data used for modeling are not only neighbors in time domain, but also neighbors in space domain), is used to identify the model of system. By employing the generalized minimum variance as the performance function and using the polynomial method, the control-law is derived for the PID controller, in which its parameters are tuned online in the process of the lazy learning identification. Simulation results show good performances of this algorithm.