Long-Term Prediction of Solar Radiation Using XGboost, LSTM, and Machine Learning Algorithms

The rapid development and application of different solar energy-based technologies have prompted research into the accurate prediction of solar energy availability. In the existing literature, different models and methods have been proposed for the accurate prediction of solar energy, however, there is still a need for the development of more accurate algorithms to forecast solar energy. Also, it is unclear in existing literature if a model can accurately and consistently predict solar energy resources when applied to different data from various locations (with varying parameters). Therefore, in this study, four different artificial intelligence models are compared for long-term solar radiation prediction. These models include long-short term memory, XGboost, multilinear regression, and decision tree regression. The models have been selected based on their high performance for similar tasks in existing works of literature. The performance of these models when applied for solar radiation prediction considering 1-minute time intervals in a location in Senegal, West Africa is presented. Furthermore, these models are applied to two different datasets (measured with two different instruments) with varying input parameters. The overarching goal of this research is to determine if a model can consistently outperform others (when predicting solar radiation), despite the variation in input parameters and locations. The prediction of global horizontal irradiance and diffused irradiance is estimated in this study. The evaluation metric used in observing the performance of the models are R, root mean square error (RMSE), and mean absolute error (MAE).