基于反步法的CSTR神经网络非奇异自适应控制

Neural Adaptive Singularity-Free Controller for CSTR System Based on Backstepping

针对一类参数不确定连续搅拌釜式反应釜(CSTR:Continues Stirred-Tank Reactors)系统中的参数不确定性,研究了一种基于反步(Backstepp ing)方法的神经网络自适应控制器。该控制器采用多层神经网络,可较好地逼近系统的复杂非线性动态,网络权值能在系统先验知识不多的情况下在线调整,更新律可用Lya-punov综合法在线获得;通过构造类加权形式Lyapunov函数,使控制器能有效处理自适应控制中可能的奇异性问题。系统仿真验证了方法的有效性和可行性。结果表明:该控制器能保证闭环系统全局稳定,并对系统参数的不确定性和有界干扰具有一定的鲁棒性。

For a class of CSTR （Continues Stirred-Tank Reactors） system with uncertain parameters, the procedure is developed for design of neural adaptive controller based on backstepping approach. It is one of the applications for muhi-layer neural networks to approximate the complex nonlinear dynamics. The NN （ Neural Network ） weights are turned on-line with no prior training needed by Lyapunov approach. The feature of the presented scheme is that by modifying a special quasi-weighted Lyapunov function, the possible control singularity issue in the design of NN adaptive controller is dealt with effectively. Finally, a simulation example is given to demonstrate the efficiency and feasibility of presented method and the result shaws that the controller is robust to some parameter uncertainties and external bounded disturbance, and it can guarantee the global boundedness for all closed-loop signals.