基于神经网络的全局滑模变结构控制

Global Sliding Mode Variable Structure Control Based on Neural Network

针对一类非线性不确定离散时间系统,提出了一种基于神经网络趋近律的全局滑模变结构控制方法。分别用两个前馈神经网络(FNNs)自适应调整趋近律中的参数ε和δ,克服了常规变结构控制方法中需要预先设定趋近律中参数的限制。在用径向基神经网络(RBFNN)对系统进行模型估计的同时,基于平移滑平面的设计方案,实现了系统的全局鲁棒滑模控制。仿真结果表明控制系统具有良好的跟踪性能,该方案使系统一开始就处于滑平面上,消除了系统抖振,具有较强的鲁棒性。

A global sliding mode variable structure control method based on neural network reaching law for a class of nonlinear uncertain discrete-time systems was proposed. Parameters, ε and δ, which were determined previously in the conventional reaching law, were regulated adaptively by two feed-forward neural networks （FNNs） respectively. System model was estimated by radial basis function neural network （RBFNN）; as well as, system global robust sliding mode control was realized based on the shifting sliding surface. Simulation results show that good tracking performance is obtained; meanwhile system state is always sliding on the sliding surface, so system chattering is eliminated and good robustness is achieved.