Advanced Time-Series Forecasting: Evaluating Deep Learning Approaches for Solar Irradiance Prediction

Abstract

The global demand for energy is substantial and continues to rise each year. In response to the escalating challenges of climate change and the global energy crisis, renewable energy sources, particularly solar power, offer a promising solution to meet growing energy needs. Consequently, effective operation and maintenance of Photovoltaic (PV) systems are crucial, as they directly influence their economic sustainability and longevity. This study aims to evaluate and compare the forecasting performance of four time-series models—Holt-Winters, Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), and Gated Recurrent Unit (GRU)—in predicting solar irradiance based on real-world meteorological data. A comparative analysis is conducted to identify the most effective model under the tested conditions and to establish a performance baseline for future hybrid or ensemble model development. Experimental results reveal that the RNN model achieves the best overall performance, with the lowest Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) values (0.1296, 0.3600, and 0.2634, respectively), outperforming LSTM, GRU, and Holt-Winters on this dataset. These findings are particularly valuable for energy system planners, researchers in renewable energy forecasting, and data scientists involved in solar grid integration, as they offer insights into model selection for enhancing forecasting accuracy and grid reliability.