Artificial Neural Network-Based Adaptive Voltage Regulation in Distribution Systems using Data-Driven Stochastic Optimization

Modern distribution networks have a high integration level of distributed energy resources (DERs). Due to the stochastic nature of renewable energy production and user load consumption, it is challenging for distribution system operators (DSOs) to maintain the voltages within safe bounds. Centralized, decentralized, and distributed operational schemes have been used to tackle these challenges, however centralized and distributed methods require extensive communication infrastructure. This paper utilizes an offline, centralized data-driven conservative convex approximation of chance constrained optimal power flow to compute PV inverter reactive power set-points with consideration of PV and load uncertainties. Then, an artificial neural network (ANN) controller is developed for each PV inverter in order to mimic the centralized PV inverter control set-points, in a decentralized fashion. Numerical tests using real-world data on a benchmark feeder demonstrate that ANN controllers can attain near-optimal performance in voltage regulation and loss improvements while satisfying the probabilistic constraints.