Improvement of a Simple Ensemble Prediction System for Wind Power Forecasting

Abstract

Improvement of a simple and practical Ensemble Prediction System (EPS) based on multi-models for wind power forecasting was performed by optimizing various ensemble member combinations. Three types of ensemble methods were compared; multi-models, multi-physics and hybrid ensemble. Multi-models were composed of four different numerical weather prediction models; WRF-NMM, WRF-ARW with small and large domains, UM-RDAPS and WW3. WRF-ARW was also used in multi-physics by designing two different planetary boundary layer and surface layer scheme, respectively. As a result, simulated wind speed from the multi-models EPS showed smaller root-mean-square error (RMSE) and a higher spread (SPRD) than those of the multi-physics EPS. Among the five different multi-models, three models such as KIER1, KIER2 and KMA-WW3 were reliable in terms of RMSE and correlation coefficient (CORR) for all typhoon cases. Therefore, a new EPS constructed by simple averaging the wind speeds taken from the three models. The new EPS showed a higher correlation and agreement with a value of 1.4% and 2.1% compared to those in old EPS, respectively.