Generative adversarial framework‐based one‐day‐ahead forecasting method of photovoltaic power output

Abstract

: Accurately forecasting the energy yields of photovoltaic (PV) systems is of great benefit to the power optimization of smart grid. However, most existing data-driven paradigm-based forecasting methods often suffer from the diversity of power fluctuation patterns, which results in poor performance while dealing with distributed PV systems that are more sensitive to external factors. To overcome this problem, this study first formulates an input data scheme based on historical PV power data, in which a novel decomposition algorithm is proposed to separate the structural information and fluctuation components in the power data. Then, a forecasting model based on the generative adversarial framework is developed, whose generative module can obtain the ability to learn the desired feature distribution with the discriminator, rather than selecting features via a certain trained preference. Moreover, the proposed method takes into account some meteorological elements to further characterise the power fluctuations. Extensive self-evaluation and comparison with several state-of-the-art methods highlight the superiority of the proposed method.