Integrated Manufacturing-Microgrid Control Using Multi-Agent Deep Reinforcement Learning

This paper presents a novel approach to integrated manufacturing-microgrid control using multi-agent deep reinforcement learning. We address the fundamental challenges of coordinating production schedules with variable renewable energy sources through a distributed control framework based on Partially Observable Markov Decision Process (POMDP). Our framework combines state-of-the-art reinforcement learning algorithms with domain-specific constraints to handle uncertainties in both manufacturing operations and renewable energy generation. Experimental results demonstrate that our Soft Actor-Critic (SAC) implementation achieves superior reward performance compared to Proximal Policy Optimization (PPO), while PPO shows better stability in production rates. The algorithms exhibit distinct battery management strategies, with PPO utilizing a wider State of Charge (SOC) range for maximum capacity utilization and SAC maintaining a more conservative range for enhanced stability. These results demonstrate the viability of our approach for sustainable manufacturing systems requiring both high performance and operational stability.