Prediction of solar radiation as a function of particulate matter pollution and meteorological data using machine learning models

A clear understanding of global solar radiation data-based location is required for designing efficient solar energy systems less affected by environmental and meteorological conditions. Hourly solar radiation was predicted as a function of meteorological data (wind speed, temperature, relative humidity, hourly sunshine) and dust particulates (PM 2.5 and PM 10 ) in Al Ahmadi, Kuwait, using the Nonlinear Autoregressive Exogenous (NARX) model. The models were trained on data from 2020 and tested on data from 2021 to assess accuracy. The prediction results were then compared to actual measurements. Relative important features behind the prediction of solar radiance were studied using eXtreme Gradient Boosting (XGBoost) and Random Forest (RF) models. The results showed the coefficient of determination (R 2 ) and the mean absolute percentage error (MAPE) between the actual and predicted values for training data sets are 95.97 % and 0.0044, respectively using the NARX model. Relative important features suggest that along with temperature, PMs have an importance similar to that of meteorological data in the prediction.