Time Series Analysis for Solar Irradiance Forecasting: A Systematic Review

Accurate solar irradiation forecasting is essential to optimize renewable energy systems, ensure grid reliability, and drive the shift to sustainable energy solutions worldwide. This study conducted a structured analysis on articles, conference papers, and references discussing the application of time series analysis for predicting solar irradiance. Our review concentrates on the forecasting methods, forecast resolution, evaluated meteorological and astronomical parameters, and the results. One topic discussed in this review is the comparative analysis between stand-alone machine learning and deep learning approaches. In addition, we also examine the hybrid methods, which include hybrid convolutional neural network (CNN)-based, long short-term memory (LSTM)-based, gate recurrent unit based, echo state network based, weather clustering, and decomposition techniques. Moreover, Kalman filter-based as well as statistical and probabilistic methods are also considered in several references that we explore. In references that we studied, various forecast resolutions are considered depending on the applications requirements. This starts from intra-hour, hourly, daily and multi-horizon resolutions. Various meteorological and astronomical parameters are also considered to assist accurate forecasting, where the most frequently used parameters are solar irradiance, temperature, humidity and wind speed.