Machine learning models to predict diffuse solar radiation based on diffuse fraction and diffusion coefficient models for humid-subtropical climatic zone of India

In the present work, twelve machine learning (ML) models are developed for assessment of monthly average diffuse solar radiation (DSR) with solitary input forecaster as clearness index. Two categories of ML models were demarcated (i.e. diffusion coefficient and diffuse fraction) with six models for each group. The correctness of models was examined as a function of some frequently used statistical pointers. A comparision was also done between developed ML models and some well-recognised models available from previous works. The results show that ML models perform very well in comparision to models available in the literature. The top-performing models in category 1 are the k-nearest neighbours (KNN) model for both training and testing data. In category 2, for training data random forest (RF) model perform well while for testing data support vector regression (SVR) models perform well. The performance can be slightly improved by using two or more input parameters such as temperature difference, relative humidity and relative sunshine along with clearness index as input. Thus, ML models can be used to estimate DSR in the humid-subtropical climate of India.