A review of data-driven deep learning models for solar and wind energy forecasting

Abstract

Numerous papers using advanced Artificial Intelligence (AI) based models - such as deep neural networks (DNN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks - have been reported for solar and wind forecasting. However, a systematic quantitative comparison of these diverse studies remains underexplored. This paper presents a comprehensive review and critical comparison of data-driven forecasting methods based on key parameters, including forecasting horizon, input features, geographical location, forecasting accuracy, training/testing period data length, pre-processing techniques, model architecture, activation functions, training algorithms, and the simulation platforms. Special emphasis is placed on data preparation strategies and model optimization techniques that significantly influence forecasting performance and model robustness. The scope is focused on purely data-driven AI and hybrid approaches, excluding physical and statistical models. An exploration of the strengths and weaknesses of these methods underscores the significance of hybrid models, particularly those combining DNN. A key contribution of this study lies in its structured synthesis of performance outcomes from various reported works, methodically arranged by increasing testing data duration. This organization aids in identifying consistently reliable and high-performing models. The findings highlight the superior accuracy and adaptability of hybrid AI models, offering practical guidance for researchers, developers, and stakeholders in renewable energy forecasting and planning.