控制系统的神经网络鲁棒设计方法

拓展了基于一个给定的模型考虑系统的摄动问题的这种传统的鲁棒控制设计方法，给定多个模型，设计多个鲁棒控制器，将有更大的容许摄动范围，针对多个不同控制器用传统方法很难同时实现的状况，用具有定性和定量多模态控制能力的神经网络同时实现了多个传统控制器的功能，方法简单，实现容易。将该方法用于飞机增稳系统的控制设计，仿真结果证明控制效果良好。

ROBUST DESIGNING APPROACH OF CONTROL SYSTEMS BASED ON NEURAL NETWORKS

In this paper the authors extend the idea that the conventional robust control designing method is based on one given model of system. The wider permitting variation will be obtained if the system is controlled by using more controllers designed based on multiple models of the system, but these controllers can not be carried out by conventional methods simultaneously. How to center the functions of these conventional controllers on an integrated controller is the main work of this paper. An approach is presented to realize the functions of these traditional controllers using one neuro controller. This scheme is based on neural networks with the abilities of qualitative reasoning, quantitative numeric calculation, learning and remembering. This method, combining conventional control theories and the advantages of neural networks can stabilize the system with wide parameter variation and structure variation. Finally, this method is used to the design of airplane stabilizing systems and the result of simulation shows excellent performance. Furthermore this approach is simple and easy to realize.