Optimal Design and Sizing of Microgrid with Hybrid Renewable Energy Sources Using Multi-Agent Deep Reinforcement Learning Techniques

The growth of electrical energy has given rise to apprehensions surrounding voltage fluctuations, system stability, and energy dissipation. Therefore, carefully considering the location and size of the Renewable Energy Source (RES) and Battery Energy Storage System (BESS) is of the utmost significance. This paper proposes a technique, Multi-Agent Deep Reinforcement Learning (MADRL), for determining the location and sizing of RES and BESS. The method is applied to the IEEE-33 nodes system, considering residential, commercial, and combination loads and the electric vehicle charging station (EVCS). The technique identifies by minimizing losses and average voltage, and this paper focuses on three algorithms: Multi-Agent Soft Actor-Critic (MASAC), Multi-Agent Deep Deterministic Policy Gradient (MADDPG), and Advantage Actor-Critic (MAA2C). The MASAC algorithms with residential load effectively identified the optimal installation locations and sizes.