Modeling Global Solar Radiation Using Machine Learning with Model Selection Approach: A Case Study in Tanzania

Abstract

The measurement of solar radiation is essential information for different applications. Thus, predicting future values of this data is also important, from load control in electrical networks to forecasts in the agricultural area. However, obtaining the necessary data for such predictions requires stations with specific equipment, which is generally expensive and scarce. This conflict between the need for data and the difficulty in obtaining it produces a scenario where new and cheaper technologies must be used to mitigate existing problems. In this context, machine learning methods have been satisfactorily used to predict solar radiation in different locations. This work presents a study of machine learning models in predicting solar radiation values in the Dar es Salaam region, Tanzania. This is the first machine learning for solar radiation prediction in Tanzania to the best of the authors’ knowledge. Here is presented a framework where three different machine learning techniques are used and compared to predict these data: support vector regression, extreme gradient boosting, and ridge regression. The results show that the three models can predict solar radiation with a good error rate and that the use of all available variables reduces the error considerably.