Reinforcement learning-based double-time-scale multi-agent system control method and system

Abstract

The application relates to the technical field of multi-agent systems, and provides a double-time-scale multi-agent system control method and system based on reinforcement learning, which comprises the following steps: based on a communication topology graph, tracking errors between agents are established, dynamic information of the tracking errors is derived, malicious attacks with control inputs are combined, integral sliding mode controllers are used to offset coupling terms, a tracking error dynamics model is obtained, H ∞ A control method is converted into a zero-sum game problem between agents, a Hamilton function related to a performance index is defined, the Hamilton function is partially differentiated, a tracking strategy is derived, the tracking strategy is brought into a Lyapunov equation, a game algebraic Riccati equation of the tracking error is derived, after splitting, an integral reinforcement learning algorithm is used, and optimal feedback gains of the multi-agent system are calculated through a strategy iteration mode. The malicious attacks are eliminated while the optimal consistency of the multi-agent system is ensured.

Images