Research On Fault Early Warning Of Wind Turbine Based On Working Condition Similarity And Statistical Principle

Most wind farms are located in complex environments such as mountains,seas,and wastelands.Wind turbine components are vulnerable to impact and failure.This not only affects the benefit of the wind farm,but also brings great challenges to the work of the operation and maintenance personnel.However,not all failures occur suddenly,and many are degenerative.If the wind farm operators can initially detect faults,they can better maintain the wind turbine components,making the wind farm stable and economic operation.Therefore,the study of fault early warning can be produced.In the past,the fault early warning of a single wind turbine was studied,but there are a lot of wind turbines in the wind farm.In order to simplify the number of wind turbine models in a wind farm and to find the similarity between wind turbines,a fault early warning method of wind turbine based on the working condition similarity and the statistical principles for modeling and analyzing is proposed.Firstly,the wind speed,power,pitch angle and generator torque which can represent the working condition of the wind turbine are clustered by using the improved NJW spectral clustering algorithm based on dynamic time warping.Dozens of wind turbines in a wind farm are divided into several clusters,and wind turbines in one cluster are similar in operating conditions.Secondly,the normal operation model of Elman neural network based on Bayesian regularization algorithm is established by choosing one wind turbine in each cluster.Then,the interval estimation of the statistical principle is used to set the maximum value of the evaluation index for the normal operation of each cluster.Finally,the pre-fault data of the wind turbine is fed into the model of the same cluster,and the evaluation index is used to judge whether the wind turbine exceeds the set threshold.Based on the SCADA data of a wind farm in Zhangjiakou,Hebei province,the results show that the 22 wind turbines can be divided into three clusters according to the similarity of operating conditions.The normal operation representative models are established in three clusters,and then the fault wind turbines is used to verify the method.The results show that the method can realize fault early warning at least 13 hours in advance.This method can simplify the number of wind turbine models,analyze the operation state of the wind turbine,and effectively realize the fault early warning of the wind turbine.