Generation Method of Multi-Regional Photovoltaic Output Scenarios-Set Using Conditional Generative Adversarial Networks

Abstract

The uncertainty of photovoltaic (PV) output power has an increasing impact on power balance with the increase of installed capacity. The construction of day-ahead PV output scenarios-set is an important basis for the stochastic optimal scheduling of the power system. For the uncertainty modeling of multi-regional day-ahead PV output, a scenarios-set generation method based on improved conditional generation adversarial network (CGAN) is proposed. This method learns the potential spatio-temporal characteristics of the output power of PV clusters distributed in different regions by convolutional neural networks. Moreover, a mapping relationship between the input PV prediction results and the output scenarios-set is established. Thereafter, the scenarios-set with correlation characteristics for day-ahead multi-regional PV clusters is generated simultaneously. By comparing with the traditional Latin hypercube sampling (LHS) method, the results of the proposed method show the comprehensive advantages in terms of the uncertainty range and the spatial correlation coefficient.