Hour-Ahead Bivariate and Multivariate Solar PV Power Forecasting for Effective Grid Integration
Abstract
For effective large grid integration and market participation, solar PV plants need accurate power forecasting. System operators use this forecasting to maintain power system security and reliability with economic dispatch. Machine learning model like Support Vector regression (SVR) is popularly used in short-term forecasting for better accuracy than other artificial intelligence (AI) models. However, SVR models need large amount of data and very complex computations to achieve better accuracy. To address these problems, this research aims to develop an SVR model improved by integrating non-linear least square Gauss-Newton method (SVR+GN) for hour-ahead solar PV power forecasting with five-minute resolution. Performance of this model was verified in bivariate and multivariate mode by comparing with SVR, non-linear and persistence models on sunny, partly sunny and overcast days. Improved SVR model needed less data and simple computations for training, while achieved higher accuracy which is useful for efficient grid integration and market participation of solar PV.