基于神经网络方法的不确定非线性微分-代数子系统的鲁棒反推镇定控制

Robust backstepping stabilization control for uncertain nonlinear differential-algebraic equations subsystems based-on artificial neural networks

对于指数1且关联可测的不确定非线性微分-代数子系统,将反推方法和神经网络相结合,研究了其鲁棒渐近镇定控制问题.基于反推方法来构造镇定控制器,利用3层的神经网络来逼近每一步控制器构造过程中的不确定项.提出一种新的自适应算法对神经网络权值进行在线调节,并适当选取每一步虚拟控制器的参数,最终得到的控制器使得闭环系统是渐近稳定的.

For a class of uncertain nonlinear differential-algebraic equations subsystems whose index is one and in- terconnection is locally measurable, the problem of robust stabilization is considered by combining the backstepping method and artificial neural networks. The robust stabilization controller is proposed based on backstepping approach by using three-layer artificial neural networks to approximate the uncertain terms arisen in the procedure of control- ler design. The weights of neural networks are updated online with a new self-adaptive algorithm. By choosing the gain parameters of the virtual controllers step-by-step, a stabilization controller is obtained through which the closed- loop systems are made asymptotically stable.