Predictive Model of Solar Irradiance Using Artificial Intelligence

Abstract

Solar power in India is growing at a tremendous pace. India's solar power capacity is 20 GW and has grown 8-fold since 2014. Assessing the solar potential in India is thus the need of the hour. The objective of this study is to make an optimized prediction model of the monthly potential of solar irradiance of the Indian Subcontinent, by utilizing hour-wise unstructured voluminous (80 million line item) satellite-based data from 609 locations for 15 years. The variables chosen are temperature, pressure, relative humidity, month, year, latitude, longitude, altitude, DHI, DNI, and GHI. Combining predictive models using combinations of SVM, ANN, and RF for factors affecting solar irradiance. This model's performance has been evaluated by its accuracy. Accuracy for DHI, DNI, GHI values on testing data evaluated through the SVM model is 95.11%, 93.25%, and 96.88%, respectively, whereas accuracy evaluated through the ANN model is 94.18%, 91.60%, and 95.90%, respectively. The achieved high prediction accuracy makes the SVM, ANN, and RF model very robust. This model with a sustainable financial model can thus be used to identify major locations to set up solar farms in the present and future and the feasibility of its establishment, wherever local meteorological data measuring facilities are not available in India. Along with the air temperature, air pressure, and humidity predictive interrelation model created to aid the irradiance model this can be used for climate predictions in the Indian sub-continental region.