Decentralized Energy Management in Electrical and Thermal Microgrids Utilizing Reinforcement Learning

This paper proposes a fully decentralized reinforcement learning–based energy management framework for hybrid electrical–thermal microgrids with distributed energy resources. Uncertainties in renewable energy generation, variations in load demand, and the nonlinear nature of battery systems make it difficult to achieve optimal energy management in microgrids. Additionally, using centralized controller techniques in large-scale systems increases computational complexity and makes controller procedure implementation more challenging. This study proposes a fully decentralized multi-agent architecture in which the stochastic performance of agents in the microgrid is modeled using Markov decision processes. This model treats consumers, batteries, and distributed thermal and electrical resources as intelligent, self-governing agents that learn from their surroundings and converge to their best policies through decentralized exploitation. The proposed model-free learning-based approach is designed to not only maximize the profits of producers but also minimize the costs for consumers and reduce the microgrid's reliance on the main grid. Finally, using real-world data from renewable power plants and electricity market data, the performance of the proposed method is evaluated through simulation and accuracy assessment.

کلیدواژه ها:

Decentralized energy management ، Microgrid ، distributed resources ، reinforcement learning ، Markov decision process