Event-triggered Integral Reinforcement Learning for Nonzero-Sum Games

Abstract

AbstractThe current methods for optimal tracking control problem in nonzero-sum games with unknown drift dynamics are time-triggered, which will not be applicable in the environment with limited transmission bandwidth and computing resources. Therefore, this paper proposes an event-triggered internal reinforcement learning method to solve the Hamilton-Jacobi-Bellman equation. A weight updating rule based on gradient descent method is used to update the weight of critic neural network without the need of initial admissible control. Then, in order to reduce the cost of computation and communication, an event triggering control law is designed. Based on Lyapunov theory, the uniform ultimate boundedness (UUB) properties of the tracking error and the critic neural network estimation error have been proved. Finally, a numerical simulation example is given to verify the feasibility of the proposed method.KeywordsNonzero-sum gamesOptimal tracking controlIntegral reinforcement learningEvent-triggeredNeural network