Data-Driven Two-Stage Voltage/VAR Control Using PV Inverters in Network Recovery

In recent years, the power outages caused by catastrophic weather events have become the main concern. Although loads can be effectively restored by network reconfiguration and line repair, topology variation has a significant impact on bus voltages in active distribution networks (ADNs). Additionally, temporal variations and intermittency of renewable power outputs also bring challenges to voltage control. To address these issues, this paper proposes a data-driven voltage/VAR control (VVC) method using PV-associated inverters for the optimal recovery process of active distribution networks (ADNs). Mathematical methods are widely used to solve the VVC problem, which may be inefficient or infeasible when uncertainties and binary recourse variables are involved. Thus, this paper coordinates a mathematical method and a deep reinforcement learning (DRL) method to solve a local VVC problem using inverters responding to uncertain PV power output and different network topologies. Routing and scheduling of repair crews are optimized by a mathematical method, and the DRL agent learns to determine the reactive power output of inverters in terms of repair procedures and uncertainties. A case study conducted on a 33-bus system demonstrates the effectiveness of the proposed method on load restoration and VVC.