Cluster division in wind farm through ensemble modelling

Abstract

Abstract To increase the equivalence accuracy of wind farms and expand their applicability under multiple operating conditions, this paper proposes a method combining blending with extreme gradient boosting (XGBoost) to realize clustering index dimensionality reduction and dynamic time warping (DTW) optimization. Density‐based spatial clustering of applications with noise (DBSCAN) is aimed at realizing clustering and fusing the clustering results. This approach can help process multi‐dimensional time‐series feature operation data of wind turbines to formulate accurate and effective wind farm plans for the division of wind turbine clusters onsite. First, the XGBoost‐Blending approach is used to select the clustering indicators for dimensionality reduction. Second, a clustering method based on DBSCAN‐DTW is established to divide the clusters and perform ensemble clustering. Finally, MATLAB/Simulink is used to build a simulation model. In this manner, a three‐phase short‐circuit is introduced at the grid‐connection point of a wind farm. A case study is performed under a variety of wind speed scenarios. The results verify the accuracy and wide applicability of the equivalence model formulated using the proposed method.