Study On Pre-processing And Early Warning Technology Of Wind Turbine Operation Data

With the rapid development of wind power industry,it has become a fundamental requirement for the development to ensure the safe and stable operation of wind turbines in this field.As the core component of the power output of wind turbines,the blades have a high failure rate and long maintenance time due to their complex mechanical structure and operation in harsh natural environments.Therefore,the research on wind turbine blade fault early warning technology has significant application value,including promoting the safe operation of the power system and reducing operating costs.The precision of early warning of fan blade cracking fault is mainly affected by two aspects.On the one hand,the operational data provided by Supervisory Control and Data Acquisition(SCADA)contains a lot of abnormal data,and on the other hand,there is coupling between multi-state data.Therefore,the focus of all wind power researchers is the effective use of operating data to realize early warning of wind turbine blade cracking faults.In view of the above problems,this work focuses three aspects,including fan operation data preprocessing,multi-state data feature extraction,blade cracking fault detection and early warning.This paper focused on the problem of a large number of outliers in the fan operation data.Firstly,the fundamental of research is analyzing the causes and distribution characteristics of outliers.Then,a multi-model fusion method based on rules,Polynomial Fitting and DBSCAN is proposed to identify outliers in wind turbine operating data.Finally,to correct the identified outliers,we introduced a combined prediction model based on Back-Propagation Neural Network and Least Squares Support Vector Machine.In result,the method of multi-model fusion is more accurately and suitable than a single model in identifying abnormal data.Compared with each single prediction model,the abnormal data correction method based on which is combined can effectively improve the correction accuracy.The analysis showed that the correlation between the condition parameters of the same components of the wind turbine is obvious,which is similar to the different.This section addresses the issue of the large number of non-fault related features in the turbine operating data.Firstly,a comprehensive correlation index is proposed to analyze the correlation between each state data.Then a Stack Autoencoder model is established on the basis of learning the mutual correlation of the wind turbine original data.It aimed to reduce the dimensionality of the multi-state data.Finally,cross entropy is used as the criterion to further extract correlation features.According to this,it provided more sufficient data support for the subsequent research on wind turbine blade cracking monitoring and early warning.Compared with Principal Component Analysis(PCA)and Independent Component Analysis(ICA),the Stack Autoencoder model has the highest accuracy for wind turbine blade cracking fault detection.Thus,the Stack Autoencoder feature extraction model proposed in this paper can effectively improve the feature quality of the data.This section deals with the problem of difficult fault detection in complex operating conditions of wind turbines.An improved Random Forest method is presented for wind turbine blade cracking fault early warning.To avoid the drawbacks of random forest feature subspace selection,a new method is proposed for the detection of wind turbine blade cracking faults which is called Cross Entropy Stack Autoencoder Random Forests(CESAE-RF).In order to improve the accuracy of wind turbine blade cracking fault warnings,an early warning process is introduced.This early warning process included a threshold for the percentage of faults within a sliding window.The analysis showed model based on CESAE-RF performed better than that on the Support Vector Machine(SVM)and K-neighbourhood(KNN)model.At the same time,the CESAE-RF model scores about 12% higher than other two models on average.For more accurate warning of wind turbine blade cracking,it can be achieved by setting a threshold value for the percentage of faults in the sliding window.