پرديس صفري

Nowadays, as energy deman‎d keeps growing an‎d centralized power systems face more challenges, the transition toward decentralized structures has become increasingly essential. At the same time, environmental concerns an‎d the need to reduce greenhouse gases have increased interest in renewable energy sources. This is where multi-carrier microgrids come in as an effective solution to improve system efficiency an‎d flexibility. However, traditional methods of managing these microgrids face serious challenges. Complex an‎d computationally expensive calculations, the requirement to reformulate optimization equations with any changes in network conditions, an‎d lack of flexibility when dealing with uncertainties are some of the main limitations of traditional approaches. In this research, we present a new approach using multi-agent deep reinforcement learning with an actor-critic algorithm, which is one of the most advanced machine learning methods, for decentralized management of multiple microgrids. In proposed method, each microgrid acts as a smart agent which uses centralized training of agents but allows them to operate in a decentralized manner during execution phase, maintaining data privacy between agents an‎d data security while enabling quick an‎d optimal decision-making with the ability to exchange energy between microgrids. The innovative aspects of this research include adding a water desalination unit to each microgrid for optimal water resource management, enhancing the thermal system through improved environmental modeling, an‎d increasing overall system flexibility through deman‎d response implementation. Simulation results show that our proposed method effectively manages multiple microgrids under optimal conditions. More specifically, the results demonstrate successful management of resource production in microgrids, energy carrier management, energy exchange between microgrids an‎d with the upstream network, an‎d high voltage stability. Additionally, this project not only reduces computational complexity an‎d communication costs but also provides high scalability an‎d modularity, allowing it to quickly adapt to environmental changes. This approach makes practical an‎d cost-effective implementation possible by reducing the need for information exchange between microgrids while maintaining their independence. The systemʹs ability to learn automatically an‎d quickly adapt to changing environmental conditions has made this method an efficient solution for sustainable an‎d smart management of future energy networks.

بهره‌برداري از سيستم‌هاي قدرت , ريزشبكه‌هاي هوشمند , يادگيري تقويتي عميق امن , سيستم چندين حامل انرژي , انعطاف‌پذيري

Power system management , Smart microgrids , Safe deep reinforcement learning , Multi-energy carrier system , Flexibility

Pardis Safari

Shahram Jadid