Geographically agnostic machine learning model for forecasting solar irradiance

Abstract

Solar Photovoltaic (PV) systems are essential in combating the energy crisis. The intermittent nature of solar isolation is a significant setback. Forecasting of solar irradiance offers a viable solution. The Artificial Intelligence (AI) models highlighted in the literature for solar irradiance forecasting are tailored to specific locations. Models designed for specific places cannot accurately predict solar irradiance in other locations. These models employ diverse techniques and are typically evaluated using at least 4 input parameters, thereby increasing model complexity. These limitations are addressed in this work. The authors present a geographically agnostic Machine Learning (ML) model that forecasts solar irradiance using a universal dataset that spans extreme climatic conditions. An ML-based model is developed with time, temperature, and dew factors as input parameters. Time series analysis is performed separately across distinct areas within all climatic zones to categorize solar irradiance data. Different ML algorithms are ensembled to formulate the model. Simulation results show acceptable values of the Root Mean Square Error (RMSE) and the coefficient of determination (R²), thus validating the performance of the proposed model. The proposed geographically agnostic model distinguishes itself by its innovative approach and promising performance, with a moderate RMSE.