非脆弱递归滑模动态面自适应神经网络控制

Non-fragile recursive sliding mode dynamic surface control with adaptive neural network

针对一类非匹配不确定非线性系统的跟踪控制问题,提出了一种递归滑模动态面自适应控制算法.采用神经网络(neural network,NN)在线逼近系统不确定项,通过设计递归滑模动态面有效综合反推步骤中每步跟踪误差之间相互影响和制约的关系.该方法避免了反推法存在的"微分爆炸"问题,克服了传统动态面方法对其低通滤波器时间常数和神经网络自适应参数摄动脆弱的缺点.稳定性分析证明了该方法能够保证闭环系统所有状态半全局一致最终有界,且跟踪误差可以收敛至原点的任意小邻域.

A non-fragile recursive sliding mode dynamic surface adaptive control method is proposed for a class of uncertain, mismatched nonlinear system. By employing the neural network （NN） to approximate the system uncertainty and designing the recursive sliding mode dynamic surface to synthesize the interaction of the tracking error in each step of backstepping scheme, we make the method to get rid of the ＇explosion of complexity＇ associated with the backstepping con- trol and to avoid being fragile to the perturbation in both the filter time constant and adaptive parameters of neural network in thetraditional dynamic surface control. Stability analysis verifies the semi-global, uniform, and ultimate boundedness （SUUB） for all the states of the closed-loop system, and guarantees the tracking error to converge to an arbitrarily small neighborhood of the origin.