UC San Diego

The development and utilization of solar energy has resulted in increased interest in solar irradiance forecasting. Ground level insolation has a natural variability due to atmospheric processes that are directly tied to the local meteorological conditions. Independent System Operators (ISOs) find that forecasting errors for small timescales are highly dependent on the characteristics and dynamics of the local cloud cover. This work seeks to explore the use of Genetic Programming to develop forecasting programs that surpass the performance of persistence forecasting. Specifically, our interest lies in forecasting a 30-second average Direct Normal Irradiance with a time horizon of five minutes. The GP-produced forecasting programs will be compared to the performance of persistence forecasting in the terms of Root Means-Squared Errors (RMSE). These proof-of-concept experiments have demonstrated that GP is a promising approach, producing forecasting programs with a 10% performance improvement over persistence forecasts