Development of machine learning models based on air temperature for estimation of global solar radiation in India

The designing of solar thermal systems need accurate information on global solar radiation (GSR). In the present study, six machine learning models, for example, random forest, k-nearest neighbors, Gaussian process regression, support vector machine, multilayer perception, and XGBoost, are developed for GSR prediction with only air temperature as input for different climatic zones of India. The performance of machine learning models is also compared with some well-known empirical models. The results show that generally, the performance of the machine learning models is better than empirical models, though, for a few climatic zones, empirical models give a better prediction. The top-performing models are k-nearest neighbors and XGBoost. Thus, we highly recommend temperature-based models to predict GSR in the regions of India where only air temperature data are available. The accurate information of future GSR can be easily obtained by combining future air temperature forecasts with KNN/XGBoost models. These models can be extremely helpful in designing solar thermal systems in those regions where solar radiation facility is not available.

Automated summary

The study found that machine learning models outperform empirical models for GSR prediction in different climate zones of India, though empirical models perform better in some areas. Temperature-based models, particularly k-nearest neighbors and XGBoost, are recommended for GSR prediction in regions with only air temperature data available. Combining future air temperature forecasts with KNN/XGBoost models can provide accurate GSR information for designing solar thermal systems in areas without solar radiation facilities.