Multi-Objective Energy Management Optimization on Grid-Integrated Microgrid Using Multi-Agent Deep Reinforcement Learning for Enhanced System Stability in HRES and BESS

The integration of BESS and RES is crucial for modern power grids, yet their optimal placement and sizing remain complex challenges due to the interdependent goals of minimizing power losses and maintaining voltage stability. This paper proposes a novel multi-agent deep reinforcement learning framework to jointly optimize the location and sizing of BESS, photovoltaic, and wind resources in a grid-connected distribution network. The primary objectives are to minimize active power loss and enhance voltage profile stability, as directly measured by the Voltage Stability Index and Voltage Deviation Index. To demonstrate the framework’s efficacy, we implement and compare three advanced MADRL algorithms Multi-Agent Deep Deterministic Policy Gradient (MADDPG), Multi-Agent Soft Actor-Critic (MASAC), and Multi-Agent Advantage Actor-Critic (MAA2C) on a modified IEEE 33-bus test system under residential, commercial, and mixed load profiles. The key contribution of thiswork is a scalable, model-free coordination strategy that directly addresses the coupled location and sizing problem for multiple DER types, overcoming the limitations of traditional iterative or centralized optimization methods. Comprehensive simulations reveal that the MADDPG-based agent consistently achieves the strongest technical performance across all test scenarios. Specifically, it reduces energy losses by up to 45.23% and improves voltage stability indices by over 15% compared to baseline scenarios, while maintaining a favourable balance between technical superiority and economic feasibility. The results underscore the potential of the proposed MADDPG-driven framework as a robust and effective tool for real-world distribution system planning, enabling enhanced operational resilience through intelligent DER coordination.