Multiple microgrids intelligent energy management with capacity constraint using hybrid deep neural network and reinforcement learning

• Hybrid model integrates DNN and RL for energy management in interconnected microgrids. • Shapley value ensures fair energy distribution, lowering average retail rates for MGs. • DNN preserves MG data confidentiality while optimizing energy distribution in the grid. • Model-free RL reduces retail costs by dynamically adjusting energy rates in microgrids. • Scalable framework manages complex energy distribution with multiple MGs and networks. This paper presents a unified hybrid energy management framework for interconnected microgrid (MG) systems, combining deep neural networks (DNN), model-free reinforcement learning (RL), and cooperative game theory. The proposed method preserves MG privacy through deep learning models that infer behavior without accessing internal data. A model-free reinforcement learning strategy enables the system operator to dynamically adjust retail pricing in response to real-time system conditions. To ensure equitable cost distribution, a Shapley value-based mechanism allocates profits fairly, even to MGs that do not receive capacity allocation. The framework supports bidirectional energy exchange under Point of Common Coupling (PCC) capacity constraints. Simulation results on a large-scale testbed reveal that the model reduces average retail prices by 9.37% compared to the case without capacity constraints and by 3.81% relative to fixed-capacity scenarios. The results validate the framework’s effectiveness in balancing operational cost, pricing flexibility, and cooperative fairness among distributed MGs.