Short Term Forecasting of Solar Irradiance using Data Analytics and Probabilistic Cloud Cover Forecasts

Solar loading refers to the intensity of solar radiation incident on a surface or region. It can be quantified in terms of solar irradiance. Accurate forecasting of solar irradiance has wide-ranging applications, but is of particular interest in the field of solar photovoltaic (PV) electrical generation, both for optimized site selection (long term) and for power grid load balancing (short term). In the current study, we focus on short term (several hours) forecasting of solar irradiance. Many previous studies have demonstrated the utility of advanced statistical techniques (“data analytics” or “machine learning”) in improving the accuracy of such forecasts. Various data analytics techniques have been applied to solar irradiance forecasting, although most studies have found some implementation of artificial neural networks (ANN) to exhibit the best performance. We build on previous studies that performed short term solar irradiance forecasting by incorporating results of our group's previous efforts on predicting short-term cloud cover probabilities into our data analytics forecasting framework. Leveraging hourly “ground truth” solar irradiance data from the national solar radiation database (NSRDB), we generate solar loading or irradiance forecasts in hourly increments ranging from t+1 to t+5 hours.