Global solar radiation prediction using machine learning approaches

Abstract

Global Solar Radiation (GSR) stands as a crucial renewable energy source for electricity and heat generation without emitting greenhouse gases. Fuelled by escalating fossil fuel prices, the necessity to curb greenhouse gases (GHG) emissions, and the rapid advancement of solar technology; the role of GSR becomes pivotal in shaping the energy landscape. So, it becomes imperative to understand the variability and availability of GSR on various time scales in the temporal domain. This research conducts an in-depth comparative analysis of various machine learning models, including Long-Short Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), Convolutional Neural Networks (ConvNet), Mul-tilayer Perceptron (MLP), Generalized Additive Model (GAM), Gaussian Process Regression (GPR), and Linear Regression (LR) for GSR prediction to recommend the best method/s for the purpose. Employing robust evaluation metrics such as: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Biased Error (MBE), and the coefficient of determination (R2), the study examines the predictive capabilities of these models. The numerical experimental results show that BiLSTM emerges as the standout performer, having minimal deviation from actual values and slightly positive bias. Its remarkable R2 value (99.26%) highlights its predictive capability.