Solar Irradiance Forecasting using Dynamic Mode Decomposition

Abstract

Reliable solar irradiance prediction is necessary for an easier transition from dependence on fossil fuels to renewable energy sources. The features of solar irradiance, such as its non-linearity and high variability, make predicting it a challenging task. This challenge is traditionally addressed by using regression and other ensemble models that require significantly large historical data to adequately train and rely on domain-specific knowledge. In this study, a data-driven framework that employed dynamic mode decomposition for solar irradiance forecasting was proposed. The efficiency of the dynamic mode decomposition-based framework was verified by employing it for short-term forecasting using two distinct datasets from geographically diverse locations. The comparative advantage over traditional regression was confirmed using performance assessment measures, including mean absolute error, mean bias error, and root mean square error. The resulting forecasts significantly outperformed the benchmark models, demonstrating that the proposed model could effectively forecast short-term solar irradiance with improved accuracy.