Research On Blind Source Separation Of Vibration Signal And State Identification Method Of Converter Transformer

In recent years,China’s power grid has gradually developed in the direction of UHV,long distance and large capacity,and the global interconnection of UHVDC transmission and UHV system has made the operating environment of transformers gradually harsh,as the core equipment of DC transmission system,the safe and stable operation of converter transformers is crucial.At present,scholars at home and abroad have carried out extensive research on the vibration characteristics and condition monitoring of ordinary power transformers based on vibration method,but the research on converter transformers is not perfect.In order to ensure the smooth operation of converter transformers and realize the monitoring and identification of converter transformer operating status,this paper carries out a series of studies on the blind source separation algorithm and load state identification of vibration signals of converter transformers,as follows:A set of inverter vibration test system based on Lab VIEW was built to realize data acquisition,data analysis,data storage,and data retrieval functions,and complete the collection and preprocessing of converter transformer vibration data.The layout of the converter transformer vibration signal test sensor was carried out,and the sensor optimization layout scheme was proposed from the aspects of different positions of the converter transformer box plane,the special structure of the box,and the small dislocation of the same measurement point sensor in multiple experiments.According to the vibration mechanism of converter transformer and the spectral characteristics of vibration signal,the vibration source and vibration transmission path of converter transformer were analyzed.In order to realize the independent analysis of core and winding vibration,a blind source separation algorithm for vibration signal of single-channel converter transformer based on CEEMDAN-KPCA is proposed,and the algorithm verification is realized through separation simulation simulation signal experiments.The vibration signal of the measured converter transformer is separated,and the two groups of vibration signals are successfully separated,and the two vibration signals are determined according to the vibration characteristics of the converter transformer,respectively,the core and the winding.In order to realize the vibration signal feature extraction of converter transformer,a variational modal decomposition(LWOA-VMD)feature extraction method based on Levy flight strategy whale optimization is proposed,which realizes the optimal search of decomposition coefficient and penalty factor,and verifies the feasibility and effectiveness of the algorithm by adaptive decomposition of the vibration signal of the no-load test at rated voltage and the vibration signal of the load test under rated current.A multi-dimensional mixed feature parameter vector of vibration signal is established,and the feature extraction of vibration signal is carried out from multiple dimensions.The research on converter transformer status recognition method is carried out,and the deep confidence network(DBN)is introduced as the basic algorithm for converter transformer operation status recognition.LWOA algorithm is used to optimize the selection of hyperparameter of DBN,forming a classification recognition algorithm based on LWOA-DBN.Blind source separation was performed on the vibration signals of the converter transformer under different load conditions and operating conditions.The obtained vibration signals of the iron core and winding were feature extracted,and the LWOA-DBN classification and recognition algorithm was used for state recognition.The feasibility of the algorithm was verified,and the classification performance of the algorithm was verified with the help of classification accuracy and recognition efficiency.The experiment shows that the proposed method for identifying the status of converter transformers has good engineering practicality and can be further applied to fault identification and operational status monitoring of converter transformers.