基于单网络评判学习的非线性系统鲁棒跟踪控制

针对一类具有不确定性的连续时间非线性系统,提出一种基于单网络评判学习的鲁棒跟踪控制方法.首先建立由跟踪误差与参考轨迹构成的增广系统,将鲁棒跟踪控制问题转换为镇定设计问题.通过采用带有折扣因子和特殊效用项的代价函数,将鲁棒镇定问题转换为最优控制问题.然后,通过构建评判神经网络对最优代价函数进行估计,进而得到最优跟踪控制算法.为了放松该算法的初始容许控制条件,在评判神经网络权值更新律中增加一个额外项.利用Lyapunov方法证明闭环系统的稳定性及鲁棒跟踪性能.最后,通过仿真结果验证该方法的有效性和适用性.

Robust tracking control for nonlinear systems based on critic learning formulation with single network

For a kind of continuous-time nonlinear systems with uncertainties, a robust tracking control method is established based on critic learning formulation with single network. Firstly, an augmented system consisting of the tracking error and the reference trajectory is established, and the robust tracking control problem is transformed into a stabilization design problem. By adopting a cost function with a discount factor and a special utility term, the robust stabilization problem is transformed into an optimal control problem. %modified Then, the optimal cost function is estimated by building a critic neural network, and consequently the optimal tracking control algorithm can be derived. In order to relax the initial admissible control conditions in the proposed algorithm, an extra term is added to the weight updating law of the critic neural network. Furthermore, the stability of the closed-loop system and the robust tracking performance are proved using the Lyapunov approach. Finally, the effectiveness and applicability of the developed approach are demonstrated via simulation results.