Solar Radiation Prediction: A Comprehensive Analysis of Machine Learning Models and Interpretability Techniques

Solar energy is an imminent renewable energy resources. Solar radiation is one of the important factors of solar energy systems. Obtaining sun radiation data for many regions is still difficult due to either absence of solar radiation measuring equipment's or due to their high cost and limited availability. This study contributes to the current machine learning approaches to find out the solar radiation value for given climatic parameters. The main goal of this study is to analyze a high-accuracy solar radiation forecasting model using regression based models as Artificial Neural Network (ANN), Neural Network Autoregressive with Exogenous Input (NNARX), XgBoost, Multilayer perceptron (MLP) and AdaBoost. The results are evaluated using MAE, MSE, R2 and RMSE. Based on the evaluation metrics the best performing algorithm for this study is XgBoost. The results are interpreted here for each algorithms based on explainable AI - LIME (Local Interpretable Model agnostic Explanations). Additionally, time series analysis was done for this study to have a better understanding of the solar radiation pattern over time.