Negative correlation learning-based RELM ensemble model integrated with OVMD for multi-step ahead wind speed forecasting

Abstract

Accurate and reliable wind speed forecasting is vital in power system scheduling and management. Ensemble techniques are widely employed to enhance wind speed forecasting accuracy. This paper proposes a negative correlation learning-based regularized extreme learning machine ensemble model (NCL-RELM) integrated with optimal variational mode decomposition (OVMD) and sample entropy (SampEn) for multi-step ahead wind speed forecasting. For this purpose, the original wind speed time series is firstly decomposed into a few variational modes and a residue using OVMD, and then the decomposed subseries with approximate SampEn values are aggregated into a new subseries to reduce the computational burden. Secondly, a NCL-RELM ensemble model is employed to model each aggregated subseries. The NCL technique is employed to enhance the diversity among multiple sub-RELM models such that the predictability of a single RELM model can be enhanced. Finally, the prediction results of all subseries are added up to obtain an aggregated result for the original wind speed. The simulation results indicate that: (1) the NCL-RELM model performs better than other ensemble approaches including BAGTREE, BOOST and random forest; (2) the proposed OS-NCL-RELM model obtains the best statistical metrics from 1- to 3-step ahead forecasting compared with the other nine benchmark models.