Localized solar power prediction based on weather data from local history and global forecasts

TL;DR

This paper introduces a hybrid weather-based model for solar power prediction that combines local historical data and global forecasts, achieving around 80% accuracy for 18-hour ahead predictions to improve PV output estimation.

Contribution

It presents a novel time series model integrating local and global weather data for more accurate solar power forecasting, addressing limitations of previous methods.

Findings

Achieves approximately 80% accuracy in 18-hour ahead solar power forecasts.

Utilizes NOAA HRRR global weather model data in the prediction process.

Demonstrates improved prediction performance by combining local and global weather information.

Abstract

With the recent interest in net-zero sustainability for commercial buildings, integration of photovoltaic (PV) assets becomes even more important. This integration remains a challenge due to high solar variability and uncertainty in the prediction of PV output. Most existing methods predict PV output using either local power/weather history or global weather forecasts, thereby ignoring either the impending global phenomena or the relevant local characteristics, respectively. This work proposes to leverage weather data from both local weather history and global forecasts based on time series modeling with exogenous inputs. The proposed model results in eighteen hour ahead forecasts with a mean accuracy of $\approx$ 80\% and uses data from the National Ocean and Atmospheric Administration's (NOAA) High-Resolution Rapid Refresh (HRRR) model.