Answers are grounded in this paper and cite their sources.
Integrating intermittent and stochastic renewable energy sources and electric vehicles (EVs) into microgrids presents significant challenges to frequency stability. These challenges are compounded in interconnected multi-microgrid systems, where uncertainties amplify, making coordinated frequency control essential. Conventional model-based analytical methods often struggle to manage stochastic disturbances and varying operational conditions. While supervised and unsupervised learning are unsuitable for such decision-making control tasks, existing reinforcement learning approaches still face shortcomings in areas such as coordination among microgrids, exploration efficiency, safety, and convergence speed. This paper proposes a coordinated frequency control framework for multi-microgrid systems (MMSs) using multi-agent reinforcement learning (MARL) guided by model predictive control (MPC) expert knowledge. An MMS model incorporating diesel generators and EVs is established, based on which MPC generates constraint-satisfying expert knowledge. The MARL framework features two key designs: phased training combining imitation learning and hybrid training, and a composite reward function integrating local frequency regulation, inter-microgrid coordination, and penalties for excessive frequency deviations. Simulations across various scenarios demonstrate the superiority of the proposed method over comparison methods in control accuracy, response speed, and disturbance rejection.
No takes yet. Share an insight, caveat, or question.
Lou et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: