Short-Term PV Output Forecasts with Support Vector Regression Optimized by Cuckoo Search and Differential Evolution Algorithms

Abstract

Renewable energy sources have gained momentum in electric power systems. Without the ability to precisely forecast their power production, integrating these sources to the electric power grid may affect grid stability. Support Vector Regression (SVR) is proven to be able to capture and deal with nonlinearity in forecasting problems. However, determining the appropriate parameters for SVR is the key issue in attaining an accurate SVR forecasting model. The objective of this paper is to forecast the output power of solar photovoltaic (PV) systems using support vector regression, of which its parameters are optimized by Cuckoo Search (CS) and Differential evolution (DE) algorithms. Real-world solar data from the 6.4kW rooftop solar PV unit located at the Advance Research Institute (ARI) of Virginia Tech in Arlington, Virginia, are used as the basis of the forecast. Six input variables are used for model development, namely day, month, hour, global normal radiation, temperature and wind speed. Model performance is evaluated using statistical indicators. Results indicate that SVR with Radial Basis function optimized by CS and DE give the most accurate forecasts.