Data‐Driven Consensus for Markov Jump Singularly Perturbed Multi‐Agent Systems: A Novel Parallel Hybrid Learning Scheme

ABSTRACT

In this article, a novel parallel hybrid learning method is applied to tackle the optimal consensus problem of singularly perturbed multi‐agent systems with Markov jump parameters. First, a distributed optimization scheme applied in multi‐agent systems is proposed. The optimal consensus problem of multi‐agent systems is reconstructed via internal information interactions under a zero‐sum game framework. Considering the parameters jumping characteristics in Markov jump singularly perturbed multi‐agent systems, a subsystem transformation approach is introduced. Furthermore, a model‐based parallel hybrid learning method is developed for multi‐agent systems to obtain fast convergence speed and meet the stability requirement for initial policy. Simultaneously, to dispense with the limitations of complex system model parameters, a model‐free parallel hybrid learning framework is proposed, which solely relies on state and control input data from each agent and its neighboring agents to generate an optimal control policy. Finally, a simulation example is given to demonstrate the efficacy of the proposed algorithm.