Photovoltaic output potential assessment via transformer-based solar forecasting and rooftop segmentation methods

Abstract

Given the escalating carbon emission crisis, there is an urgent need for large-scale adoption of renewable energy generation to replace traditional fossil fuel-based energy generation for a smooth energy transition. In this regard, distributed photovoltaic power generation plays a crucial role. Predicting the GHI in advance to predict the power of photovoltaic power generation has become one of the methods to solve the grid-connected stability in recent years, which enables the grid staff to dispatch and plan in advance through the forecast results, reduce fluctuations, and maintain grid stability. In this study, we present a deep learning-based method to assess photovoltaic output potential by solar irradiance forecasting and rooftop segmentation. First, we utilize a multivariate input Transformer model that incorporates various data to predict GHI; Second, using remote sensing images to train Swin-Transformer to identify the potential area of rooftop photovoltaic panel; Finally, the potential assessment was achieved by calculating the array output through the GHI and area data we generated in the first two parts. Our evaluation methodology and results provide technical support for the transition of energy structure.