Neural Network Learning Control of Multi-input System with Unknown Dynamics

There are few studies on the optimal control of the multi-input system with different input dynamics in the literature. For this problem, the learning Nash controllers are obtained with a simplified-reinforcement learning (SRL) scheme and Nonzero-sum game theory. A neural network (NN) identifier is first established to approximate the unknown multi-input system. Then SRL NNs are used to approximate the optimal performance index of each input, which is used to learn the optimal control policies for the multi-input system. The weights of the NN architecture are tuned with a novel algorithm, and the parameter convergences are analyzed to be uniformly ultimately bounded. Finally, one two-input nonlinear system is introduced to verify the proposed learning control scheme.