Optimal Cooperative Control of Multiagent System Based on Generalized Policy Iteration
Haixing Li, Yong Zhang, Ke Wang, Chaoxu Mu · 2023
In this paper, an adaptive dynamic programming algorithm (ADP) based on generalized policy iteration (GPI) is proposed for the multiagent system (MAS) cooperative control problem. The main idea is to solve the approximate optimal solution of the Hamilton-Jacobi-Bellman equation using ADP technique by constructing the performance index function and the critic-actor dual network structure. Among them, neural network-based critic-actor structure is used to obtain iterative control policy and value function. Next, the convergence and optimality of the GPI algorithm are proved. Finally, the effectiveness of the algorithm is verified by a discrete multiagent system with a leader-follower structure.