Adaptive Wind Speed Forecasting Based on Optimized Ensemble Numerical Weather Prediction and Temporal Feature Selection

Abstract

The huge uncertainty of wind energy coupled to the highly wind penetrated power system calls for accurate wind speed forecasting (WSF). However, the existing numerical weather prediction (NWP) involved WSF adopts the non-adaptive NWP correction strategy and stochastic feature selection method without temporal evaluation, limiting the potential promotion for ultra-short-term WSF precision. This paper introduces a novel adaptive ultra-short-term WSF method incorporated with NWP from multiple meteorological organizations. An optimized ensemble (OE) approach produces the combined NWP with adaptive time-varying weights. Temporal feature selection (TFS) is operated on the ensemble NWP based on innovative temporal-related evaluation criterion and observation domain, which decides the feature vectors for WSF in each time step. A realworld case verifies the superiority of the proposed WSF method. The adaptability of the ensemble NWP and the validity of the TFS are also discussed.