Sliding Mode Trained Neural Control for Single and Coupled Inverted Pendulum System

This study applies a sliding-mode-based neural network to control inverted pendulum systems. Neural network weights are updated using a cost function which denotes distance from the sliding manifold. Thus, minimizing the cost function equals reaching the sliding surface. Sliding mode based neural network also makes the system robust to uncertainties in parameters and dynamical uncertainties. Chattering effect is solved by modifying the cost function. Simulations are fulfilled for a SISO and a MIMO model of an inverted pendulum. The results of simulations reveal the effectiveness of proposed method.