Abnormal Wind Turbine Data Identification Using a Dirichlet Process Gaussian Mixture Model

A large amount of abnormal data will be generated during the actual wind turbine operation, thus the raw data can't be directly applied to the subsequent work such as wind turbine power prediction and generation performance evaluation. This paper proposes an abnormal data identification method based on the Dirichlet Process Gaussian Mixture Model (DPGMM) to preprocess the raw data effectively. Firstly, all data points are allocated into corresponding power bins created in the horizontal power direction with a certain interval in the wind speed-power (V-P) coordinate system. And then, the DPGMM model that can adaptively determine the optimal number of Gaussian components is used to cluster the data points in each power bin. At last, combined with the parameters of each Gaussian component confidence ellipse and data points distribution characteristics in V-P coordinate system, the abnormal Gaussian components and their clustering, abnormal data can be accurately identified. Using actual wind turbine SCADA data, the proposed method is demonstrated to be effective.

Automated summary

This study proposes an abnormal data identification method based on DPGMM to preprocess raw data from wind turbine operation. By allocating data points into power bins and clustering them using DPGMM model, combined with confidence ellipses and data point distribution characteristics, abnormal data can be accurately identified.