Improvements to Solar Energy Generation Predictions via Temporal Resolution of Meteorological Data

Solar energy generation using photovoltaic technologies has emerged as one of the core renewable energy sources.  Therefore, its prediction, which highly relies on meteorological information, is of paramount importance for various applications including power grid planning, integrating solar power effectively into the power grid, resource allocation, and sustainable energy management.  This study presents the results of our investigation on the impact of the temporal resolution of meteorological data on solar energy generation predictions using machine learning (ML) models and a methodology that we developed to improve these predictions.  Two independent solar energy generation datasets (one for urban rooftop solar panels in San Antonio, Texas, U.S., and the other one for a solar farm in India) along with the corresponding meteorological datasets that include solar irradiation, wind, temperature, and humidity information were utilized.  Our results demonstrate that the temporal resolution of the meteorological data has a profound influence in predicting solar energy during rapid meteorological changes, particularly during sunrise and sunset times when the time rate of change of the meteorological data may be relatively large.  We developed a simple, yet very effective, method to enhance machine learning model predictions through modifying the temporal resolution of the available meteorological data.  We will present, using quantitative metrics, the predictive capabilities of our ML-based model, the impact of the temporal resolution of meteorological data on prediction accuracy, and the method that we developed to enhance prediction accuracy.  This research was supported by the funds provided by CPS Energy (San Antonio, Texas, USA) through UTSA-TSERI.