神经网络在鲁棒镇定控制器设计中的应用

Neural network algorithm for stabilizing control of robustness in systems

采用神经网络法求解最优的鲁棒控制器,解决了不确定参数线性离散系统的在线优化控制问题,保证了系统具有良好的动、静态性能.从一个保证系统稳定的引理出发,通过二次规划确定鲁棒镇定控制器的形式,采用神经网络计算控制器参数,该网络具有全局收敛性,不用训练数据,易于用电子线路实现,且满足实时控制的需要.并进行了仿真验证.

In order to solve a class of online optimal control problems of linear discrete-time systems with uncertain parameters and to ensure perfect dynamic static behavior, a neural network approach is proposed for solving optimal stabilizing controller with robustness. From a lemma which can guarantee global asymptotic stability of the system, the form of optimal stabilizing controller with robustness is established by quadratic programming. A neural network is proposed for computing the parameters of this controller. It has global convergence property and does not use the data for training, so it is convenient for implementation with electron circuitry and satisfies the requirement for real-time control. The advanced performance of the proposed network is demonstrated through simulation.