Graph Neural Network based Short-term Solar Irradiance Forcasting Model Considering Surrounding Meteorological Factors

Abstract

Accurate short-term solar irradiance forecasting can achieve precise solar photovoltaic (PV) power forecasting and ensure the safe and stable operation of power grid. However, the existing solar irradiance forecasting methods only based on the historical power data and meteorological information of the local PV power station itself, which is difficult to obtain sufficiently accurate forecasting results. In this paper, we propose a short-term irradiance forecasting model based on Graph Neural Network (GNN) considering surrounding meteorological factors to further improve the accuracy. Firstly, the spatio-temporal correlation stations are constructed according to geographical location and meteorological information, and simulate the spatio-temporal correlation data around the target station by utilizing the satellite image-irradiance mapping model. Secondly, based on the complex network theory, a new index is proposed to evaluate the connectivity of the graph structure, which improves the predictive ability of the GNN model. Finally, the spatio-temporal correlation around the target site is mined through GNN model to achieve the short-term irradiance forecasting. The results show that the proposed method further improves the forecasting accuracy compared with models that don’t consider surrounding meteorological factors. The reliability of the graph connectivity is directly proportional to the forecasting accuracy, which verifies the effectiveness of the proposed index.