محمد حسين نوائي سرشكه

The widespread use of renewable energy resources an‎d distributed generation units has established multi-microgrid systems as a primary solution for achieving economic an‎d sustainable operation in distribution networks, while simultaneously underscoring the need for real-time, data-driven energy management. In recent years, deep reinforcement learning (DRL) algorithms have attracted considerable attention due to their capability to make real-time decisions for microgrid energy management problems; however, training these algorithms requires a large volume of operational data, an‎d collecting such data can compromise the privacy an‎d security of individual microgrids. To address this issue, this study proposes a federated hierarchical energy management framework for a multi-microgrid system within a cloud–fog–edge computing architecture, in which only the knowledge derived from operational experience is shared among microgrids, thereby avoiding the direct exchange of raw operational data. First, each microgrid independently trains its own reinforcement learning agent using only local operational data. Then, the updat‎ed parameters of the local agents are periodically transmitted to a central server, where a global model is obtained through weighted aggregation an‎d subsequently sent back to the microgrids to updat‎e their local agents. In this way, operational data an‎d detailed information on generation an‎d consumption remain at the local level, which helps protect confidentiality an‎d enhances information security. Simulation results show that, compared with a conventional hierarchical structure, the proposed federated hierarchical framework preserves decision-making quality an‎d operational costs, while achieving a significant improvement in privacy protection. Furthermore, by reducing the amount of exchanged data, the proposed method decreases communication ban‎dwidth consumption an‎d improves the scalability of the system.

سيستم مديريت انرژي , شبكه هوشمند , سيستم چند ريزشبكه‌اي , يادگيري تقويتي عميق , يادگيري فدرال

Energy Management System , Smart Grid , Multi-Microgrid System , Deep Reinforcement Learning , Federated Learning

Mohammad Hosein Navaei Sereshkeh

Dr.Shahram Jadid