Multi-Agent Deep Reinforcement Learning- Based Autonomous Control of Grid-Forming Inverters in Renewable-Dominated Power Systems

Abstract: The rapid transition towards renewable energy sources such as solar and wind power has significantly transformed modern power systems. However, the large-scale integration of these intermittent and stochastic energy sources introduces severe challenges related to grid stability, frequency regulation, voltage control, and power quality. Traditional power systems rely on synchronous generators to provide inertia and maintain system stability, but with increasing penetration of inverter-based renewable sources, this inherent stability is gradually decreasing.

Grid-Forming Inverters (GFIs) have emerged as a promising solution to address these challenges by emulating the behavior of conventional synchronous machines. These inverters are capable of regulating voltage and frequency while supporting grid stability under dynamic conditions. However, conventional control strategies such as droop control, PI/PID controllers, and model-based techniques exhibit limitations in handling highly nonlinear, uncertain, and time- varying operating conditions.

In this context, Artificial Intelligence (AI), particularly Deep Reinforcement Learning (DRL), provides a powerful framework for developing adaptive and intelligent control strategies. This paper proposes a Multi-Agent Deep Reinforcement Learning (MADRL)-based autonomous control approach for grid-forming inverters in renewable- dominated power systems. In the proposed framework, each inverter operates as an independent intelligent agent that learns optimal control actions through continuous interaction with the environment.

The multi-agent structure enables decentralized control, coordination among multiple inverters, and scalability for large power systems. The proposed approach enhances voltage stability, frequency regulation, power sharing, and fault tolerance under varying operating conditions. Simulation results demonstrate that the MADRL-based control significantly outperforms conventional methods in terms of dynamic response, stability margins, and overall system efficiency.

Keywords: Grid-Forming Inverter, Deep Reinforcement Learning, Multi-Agent Systems, Smart Grid, Renewable Energy Integration, Autonomous Control, Power System Stability.