Cluster-based ensemble learning for wind power modeling from meteorological wind data

Abstract

Reliable and efficient power modeling from meteorological wind data is vital for optimal implementation and monitoring of wind energy, and it is important for understanding turbine control, farm operational optimization, and grid load balance. Based on the idea that similar wind conditions lead to similar wind powers; this paper constructs a modeling scheme that orderly integrates three types of ensemble learning algorithms—bagging, boosting, and stacking—and clustering approaches to achieve wind power modeling from multiple wind-based meteorological factors in a wind farm. The paper also investigates the applications of different clustering algorithms and methodologies to determine cluster numbers in the modeling. The results reveal that all ensemble models with clustering exploit the intrinsic information in wind data and thus outperform models without clustering by approximately 15% on average in modeling wind power. The model with the best-performing Farthest First clustering is computationally rapid and with an improvement of around 30% compared with the baselines. Given the diversity introduced by clustering algorithms, the power modeling performance is further boosted by about 5% by introducing stacking that fuses ensembles with varying clusters. The proposed modeling framework thus demonstrates promise by delivering efficient and robust performance on the targeted problem.