Particle Swarm optimization-Based Neuro-Dynamic Programming for Nonzero-Sum Games of Multi-Player Nonlinear Systems

Abstract

This paper focuses on an integral reinforcement learning (IRL)-based optimal control scheme using particle swarm optimized neural networks for nonzero-sum games of multi-player nonlinear systems with unknown drift dynamics. By combining IRL with neuro-dynamic programming method, the identification procedure is obviated. The optimal control policy of each player is acquired by solving the coupled Hamilton-Jacobi equation via the particle swarm optimized critic neural network, which avoids the difficulty in selecting the initial weight vector manually. The closed-loop system is ensured to be stable according to the Lyapunov’s direct method. The effectiveness of the developed scheme is demonstrated by numerical simulations.