A review of machine learning and deep learning models, features, and applications for solar PV forecasting

Abstract

Solar energy has become an important source of renewable energy towards supporting the increased electricity demand in the world and minimizing reliance on fossil energy. Nevertheless, solar irradiance is intermittent and unreliable, which necessitates the precise prediction of solar energy to promote effective integration and energy management into the grid. This review bridges a gap in the research to unify the different machine learning (ML) and deep learning (DL) methods to predict solar power and solar resource, with the need to have a comparative evaluation of the performance, interpretability, and adaptability of the methods. This paper presents a systematic review of the recent developments in ML and DL based models applied in the analysis of solar power and solar resources forecasting, such as standalone, hybrid, and ensemble models. The research topic is to determine the appropriate parameters of input, feature selection techniques, and forecasting horizons that improve the accuracy and precision of the model. The methodology that was adopted is a comprehensive comparative analysis of the previous research with an emphasis on the strengths, weaknesses, and possibilities of the various predictive models. The findings prove that hybrid and ensemble models are invariably more successful compared to traditional models, in terms of accuracy and reliability. The review shows that the synergy of AI-based solutions can determine significant improvements in terms of accuracy of forecasting, energy scheduling, and grid stability. Taking all of this into consideration, the provided review fits into the smart prediction model evolution for effective and environmentally friendly use of solar energy. The key contributions of this paper and the main highlights are as follows: Introduces the general classification of ML and DL methods applied on PV power and solar resource prediction. Compares single, hybrid, and collaborative forecasting destructions at diverse timescales. Discusses how feature selection and meteorological parameters can be used to improve the accuracy of the model. Recognizes interpretability, data availability, as well as generalization challenges in AI-based forecasting. Recommends future research initiatives, such as transfer learning, probabilistic forecasting, and explainable AI usages. Proposed a graphical taxonomy to have a better conceptual knowledge and viable model choice. Provides an updated synthesis (2020–2025) of literature to ensure contemporary relevance to solar PV forecasting research. Introduces the general classification of ML and DL methods applied on PV power and solar resource prediction. Compares single, hybrid, and collaborative forecasting destructions at diverse timescales. Discusses how feature selection and meteorological parameters can be used to improve the accuracy of the model. Recognizes interpretability, data availability, as well as generalization challenges in AI-based forecasting. Recommends future research initiatives, such as transfer learning, probabilistic forecasting, and explainable AI usages. Proposed a graphical taxonomy to have a better conceptual knowledge and viable model choice. Provides an updated synthesis (2020–2025) of literature to ensure contemporary relevance to solar PV forecasting research.