| On-load tap changers(OLTCs)with complex mechanical structure and electrical characteristics,are the key component for on-load voltage regulation of converter transformers.The faults resulting from OLTCs include mechanical and electrical faults,which influence the operation stability of converter transformers.Generally,the mechanical faults account for the largest proportion and the electrical faults result from mechanical faults.Therefore,it is important to monitor and evaluate the mechanical status of OLTCs for improving the stability of converter transformers and even the Ultra High Voltage Direct Current(UHVDC)projects.Based on the OLTC vibration signals and the driving motor current signals,this paper studies the vibration signal preprocessing,fault diagnosis,and anomaly detection algorithms for the mechanical status evaluation of OLTCs.The main work includes:Firstly,the vibration signal preprocessing methods are investigated,including the endpoint detection and noise reduction.For the endpoint detection,the frequency and energy characteristics of the three-stage vibration signals are studied,and an endpoint detection method of OLTC vibration signals based on zero-crossing rate and short-time energy is proposed,which realizes the precise location of the start and end time of stage II in vibration signals.For the noise reduction,the period,frequency and amplitude characteristics of the background noises from the vibration signals are studied,and the source of the background noises are determined.The noise reduction method based on the Second Order Blind Identification(SOBI)algorithm is proposed,and the background noises due to core and winding vibration are separated successfully.Secondly,the OLTC fault diagnosis method based on machine learning and multi-source data fusion is studied.Considering that the OLTC vibration signals are non-stationary,nonperiodic,and have large number of sampling points,and lack labeles,a feature extraction and fault diagnosis method based on Bayesian Optimization-Convolutional Ladder Network(BOConv LN)is proposed,which improves the feature extraction ability and fault diagnosis accuracy for OLTC vibration signals.For the current signals of the driving motor,the two features,namely the OLTC switching duration and root mean square(RMS)of the current are extracted,and a fault diagnosis model based on Support Vector Machine Classifier(SVC)is established.In order to integrate the models based on the three channels of vibration signals and one channel of driving motor current signals,the multi-source data fusion method based on ensemble learning is carried out,and an OLTC integrated fault diagnosis model is established,which has a diagnosis accuracy of 100% for the normal status,transmission shaft jams,poor switch lubrication,and top cover looseness of OLTCs.Thirdly,the OLTC anomaly detection method based on machine learning and multi-source data fusion is studied.For vibration signals,a vibration signal anomaly detection method based on Deep Support Vector Data Description(Deep SVDD)is proposed,and the high-dimensional feature mapping and automatic feature extraction of vibration signals are realized via deep neural networks.For the driving motor current signals,an abnormal detection method based on One Class-Support Vector Machine(OC-SVM)is proposed.In order to integrate the models based on the three channels of vibration signals and one channel of driving motor current signals,the transform method of anomaly scores for the Deep SVDD and OC-SVM algorithms is studied,and an OLTC integrated anomaly detection model is established,which has a detection accuracy of 99.58% for OLTCs.Finally,a status evaluation system of OLTC system with sensing layer,communication layer,and evaluation layer is designed.The sensor selection,data acquisition device construction,data transmission mode design,and development of Web software,database,and algorithm model are carried out,and tested on the system on the experimental platform of VRG-OLTC. |
PDF Full Text Request