Application of Extreme Learning Machine for short term output power forecasting of three grid-connected PV systems

Abstract

The power output (PO) of a photovoltaic (PV) system is highly variable because of its dependence on solar irradiance and other meteorological factors. Hence, accurate PO forecasting of a grid-connected PV system is essential for grid stability, optimal unit commitment, economic dispatch, market participation and regulations. In this paper, a day ahead and 1 h ahead mean PV output power forecasting model has been developed based on extreme learning machine (ELM) approach. For this purpose, the proposed forecasting model is trained and tested using PO of PV system and other meteorological parameters recorded in three grid-connected PV system installed on a roof-top of PEARL laboratory in University of Malaya, Malaysia. The results obtained from the proposed model are compared with other popular models such as support vector regression (SVR) and artificial neural network (ANN). The performance in terms of accuracy and precision of the prediction models is conducted with standard statistical error indicators including: relative root mean square error (RMSE), mean absolute percentage error (MAPE), mean absolute bias error (MABE) and coefficient of determination (R2). The comparison of results obtained from the proposed ELM model to other models showed that ELM model enjoys higher accuracy and less computational time in forecasting the daily and hourly PV output power.