Decentralized Actor-Critic with Graph Neural Networks for Real-Time Reactive Power Coordination in Active Distribution Networks

The increasing integration of distributed renewable energy resources into the active power grid creates a lot of voltage fluctuation difficulties. This causes an imbalance in the active power grid system. To address these complicated issues, this paper suggests a novel Decentralized Actor-Critic with Graph Neural Networks (DAC-GNN) model by considering the future scope of the existing research papers based on the demand for an efficient decentralized reactive power coordinating system. Instead of traditional centralized approaches, DAC-GNN uses graph-based representations to increase the use of the power grid’s topological structure; this allows the agents to make well-informed decisions based only on local and nearby information. This completely decentralized system solves the dependency on global state information and the essential stability and communication issues noted in previous studies. This proposed decentralized deep reinforcement learning (DRL) model based on DAC-GNN is designed to effectively coordinate reactive power modifications throughout the active grid, which preserves voltage stability and reduces power losses. The study simulation was also conducted based on the guidelines of the existing reference paper by using IEEE testing distribution feeders to verify the system’s stability under various grid conditions, including high renewable involvement and dynamic loads. Therefore, the comparative evaluations show that the proposed DAC-GNN model effectively outperforms the intelligent algorithm with its effective outcomes regarding voltage management, power loss reduction and scalability. Overall, these results confirm the effectiveness of the proposed approach in addressing modern grid demands and providing an effective solution for real-time voltage and reactive power coordination in active distribution networks.