Generative adversarial network assisted stochastic photovoltaic system planning considering coordinated multi-timescale volt-var optimization in distribution grids

Abstract

In this paper, a novel photovoltaic (PV) system planning framework is proposed for distribution grids. The main contribution of this work is that the volt-var control (VVC) capability of PV inverters is duly considered during the planning stage to reduce the expected power loss cost and meanwhile counteract uncertain voltage fluctuation and deviation caused by random PV production and load demand. Specifically, the proposed PV system planning framework is formulated as a two-stage stochastic optimization model, where the first stage is to determine the planning decisions of PV systems before the uncertainty realization, and after the uncertainty is observed, the second stage coordinates PV inverters and existing legacy VVC devices (capacitor banks) to minimize the power loss and voltage fluctuation in a multi-timescale manner. To solve the formulated complex optimization problem, the original model is decoupled and solved using an iterative solution method where commercial solvers can be directly used to obtain the optimal solutions. Besides, a data-driven based scenario selection method based on Generative Adversarial Network (GAN) is proposed to capture the uncertainties of PV production and load demand with the full diversity of behaviours. Finally, the proposed framework is tested on IEEE 37-node and 123-node test systems to demonstrate the effectiveness of the proposed method. This paper can provide important insights into investment strategies of renewable resources designed by power grid authorities, which can benefit from the proposed framework to maintain the reliable system operation and reduce the renewable energy curtailment.