Research On Fault Diagnosis Of Hydroelectric Generating Unit Integrating Multi-source Wide Area Data

Since the hydropower station has unique peak-shaving and valley-filling operating characteristics,it can play an important role in regulating loads and maintaining the safe and stable operation of the power system.It has gradually become an indispensable adjustment tool for our national power system.Due to the frequent switching of complex working conditions of hydroelectric generating units,there are many unstable factors that affect the operation status of the unit during operation,so the safety problems of the unit are increasingly prominent.The fault diagnosis of traditional hydropower units is mainly based on an expert system,which has a high subjective dependence and cannot effectively and objectively analyze the unit status and fault category.In order to overcome the limitations of the traditional fault diagnosis model,this paper takes hydropower units as the object,conducts research on the three types of multi-source data collected by the condition monitoring system and their corresponding fault characterization,data processing and analysis methods,and fault diagnosis model,focusing on solving Multi-source wide-area data brings fault information redundancy and the high time-consuming of traditional fault diagnosis models.A fault diagnosis model of hydro-generator unit incorporating multi-source wide-area data is proposed,which opens up ideas for the realization of hydropower unit fault diagnosis.,Has a certain engineering value.This article first briefly understands several typical failure types of hydropower units,introduces different data sources that can be collected by the condition monitoring system,such as vibration signals,operating condition parameters,air gap parameters,and axis trajectory charts,etc.,and analyzes the different types Failure characterization of data.Secondly,in view of the problem of redundancy of fault information caused by the high sample space dimension after fusing the original data of multi-source wide area data,a feature reduction method based on improved principal local and overall principal component analysis(LGPCA)is proposed.This method combines the advantages of principal component analysis(PCA)and local preserving projection(LPP).The dimensionality reduction process not only retains global feature information,but also covers popular features between local samples;and by introducing cosine similarity The problem that the LGPCA method is too dependent on artificial experience when constructing the near field is improved;finally,the three dimensionality reduction methods are compared by strength analysis,which confirms the effectiveness of the proposed method.Thirdly,the feature extraction method of vibration signal is studied,and the advantages and disadvantages of empirical mode decomposition(EMD)and ensemble empirical mode decomposition(EEMD)signal decomposition are compared.The wide area data such as time domain and frequency required in this paper The characteristics of the domain and time-frequency domain are introduced.Subsequently,random forest theory was selected as the classifier of the diagnosis model,and a fault diagnosis model of hydroelectric generating unit fusion multi-source wide area data was proposedFinally,based on the actual operation data of a hydropower station's tubular unit,an example analysis was carried out.By comparing the fault diagnosis results under the fusion of different data types and different feature reduction methods,the diagnosis of fusion multi-source wide area data proposed in this paper was proved.The method can not only improve the accuracy of fault diagnosis,but also effectively reduce the time required for diagnosis and improve the efficiency of unit diagnosis.