Training method for a sliding mode controller and quantified robustness against uncertainty

摘要

Proposes a design method of a sliding mode controller by training a neural network. The singular solution of the optimal control problem is applied to training the neural network. We focus on the robustness of the trained controller against uncertainties, and we propose a method to quantify the robustness of any kind of controllers by training another neural network. Moreover, methods to train a quantified robust controller are proposed, and they can also improve the robustness. Some numerical simulations show the effectiveness of proposed methods.