Localized Volt-Var Optimal Control by Centralized OPF and Decentralized Neural Network Training

The centralized optimization for distributed energy resources (DERs) dispatch is well-known for its communication requirements, and the traditional local volt-var control cannot perform system-wide coordination. An advanced local volt-var optimization method using centralized OPF and decentralized learning methods for controllable devices in the distribution system is proposed to eliminate the communications while guaranteeing a global optimization control performance. Two different sensitivity factors for photovoltaics (PVs) and load-tap changers (LTCs) are adopted to linearize the centralized optimal power flow (OPF), and results obtained from the coordinated simulation are then utilized for individual device training. Controllable devices aim to mimic the centralized OPF result with their local measurements, providing system coordination without communication. The proposed approach leverages available grid and controllable devices, eliminating reliance on communications while being adaptive and robust to volatile operating conditions.