A Feature Importance Analysis Based Solar Irradiance Mapping Model Using Multi-channel Satellite Remote Sensing Data

Abstract

Solar irradiance is a crucial parameter that impacts the accuracy of photovoltaic (PV) generation prediction. However, due to the equipment deployment limitations and malfunction factors, accurate irradiance data with sufficient historical accumulation and wide spatial distribution are usually unavailable. Therefore, satellite remote sensing shows its unique significance as an additional stable irradiance observation source. This paper develops a feature importance analysis-based solar irradiance mapping model to calculate ground solar irradiance only using satellite data. In this paper, the K-means method is employed to classify the weather condition. The correlation of all channels data of FY4 satellite with the solar radiation of PV site is analyzed using the Pearson correlation coefficient under different sky conditions. The XGBoost feature important algorithm is applied to analyze the importance of different channel features, which optimizes and determines the input of the mapping model. The gradient boost regression model (GBR) is chosen as the mapping model to calculate solar irradiance with the combination channels satellite data obtained according to feature important analysis. The simulation results show that the proposed model performs best compared with other regression methods.