Attention Enabled Multi-Agent DRL for Decentralized Volt-VAR Control of Active Distribution System Using PV Inverters and SVCs

This paper proposes attention enabled multi-agent deep reinforcement learning (MADRL) framework for active distribution network decentralized Volt-VAR control. Using the unsupervised clustering, whole system can be decomposed into several sub-networks according to voltage and reactive power sensitivity relationships. Then, distributed control problem of each sub-network is modeled as Markov games solved by improved MADRL algorithm, where an adaptive agent. An mechanism developed help agent focus on specific information that mostly related reward. All agents are centrally trained offline learn optimal coordinated strategy executed in a manner make online decisions with only local information. Compared other approaches, proposed method effectively deal uncertainties, achieve fast decision makings, significantly reduce communication requirements. Comparison results model-based data-driven methods IEEE 33-bus 123-bus systems demonstrate benefits approach.