A review on data-driven methods for solar energy forecasting

Abstract

Solar photovoltaic energy has emerged as one of the fastest-growing electricity-generation technologies, making substantial contributions to carbon-free energy production. To fully harness its potential and ensure efficient grid integration, accurate solar energy forecasting techniques are essential. This review systematically analyzes global contributions to solar energy forecasting research through an in-depth bibliometric study of 1323 research articles published between 2013 and 2022. A detailed examination of 75 influential articles from this collection provides insights into the evolution and current state of forecasting approaches. We assess the use of statistical, machine-learning, deep-learning, and hybrid models and evaluate their performance across various temporal horizons and geographical contexts. Our analysis reveals a notable shift from statistical models toward machine-learning and deep-learning approaches, particularly from 2018 to 2022. Hybridization of models consistently reduces forecasting error by over 20 % compared to single-model approaches in most reviewed cases. We also explore aspects such as model complexity, data sources, forecasting accuracy, influence of meteorological parameters, and data-processing techniques. Our findings underscore a global transition toward deep-learning-based hybrid models that demonstrate superior scalability and accuracy. These models are increasingly adopted across various spatial and temporal forecasting scenarios, paving the way for standardized methodologies and helping address regional disparities in solar energy forecasting research and implementation.

Topics & Concepts