基于二阶Adaline网络的滑模变结构控制器

针对一般神经网络滑模控制系统在线训练时间长、实时性差的问题,提出一种自适应神经滑模变结构控制系统。该系统主要由二阶Adaline网络辨识器、滑模控制器和变结构二阶学习算法估值器组成。其中二阶Adaline网络是对原Adaline网络的改进,在其输入样本向量中增广了有关样本的二次项,相当于用时变二次多项式去逼近非线性系统,既有原Adaline网络训练速度快的优点,又具有一定非线性逼近能力;滑模控制器根据辨识误差自适应调整趋近律中的增益参数;而变结构二阶学习算法首先对递推近似Newton法进行简化,然后将其改造成一种新型的变结构算法,具有较快的收敛速度。针对同一非线性系统,采用常规神经滑模变结构控制存在较小的抖振,平均调节误差为0.0913;而采用自适应神经滑模变结构控制很快趋于平稳,几乎消除高频抖振,平均调节误差为0.0109。仿真结果证明了该方案的有效性。

Sliding Mode Variable Structure Controller Based on Second Order Adaline Network

A new system is put forward for general common neural network sliding mode control system having long training time online and bad real time quality.It is mainly composed of second order Adaline network identifier,sliding mode controller and variable structure second order learning algorithm estimator.And second order Adaline network is the improvement of original network and is equioalent to approximate nonlinear system using time-varying quadratic polynomial.Because the second-order items of related samples are augmented in the vector of input samples,it has quick training velocity as the advantage of original Adaline network and definite nonlinear approximation ability.Sliding mode controller adaptively adjusts gain parameter of approach law according to identification error.The above variable structure second order learning algorithm not only simplifies approximation recursive Newton algorithm but also reconstructs it into a new variable structure algorithm,and has relatively quick convergent velocity.For the same nonlinear system,less chattering is occurred and average adjusting error is 0.093 1 when the conventional neural sliding mode variable structure control is used.But high frequency chatting is almost eliminated,the response is quickly seffled and average adjusting error is 0.010 9 when new type neural sliding mode variable structure control is used.The simulation result shows that the scheme is effective.