Research On The Vibration Characteristics And Fault Diagnosis Of Transformer Core And Winding

Power transformer is the core equipment of power system,and its vibration is closely related to the state of core and winding.Therefore,it is important to study the vibration characteristics and fault diagnosis method of transformer to ensure the reliable power supply of power grid.At present,China is in the period of integration and development of energy and digital revolution,and the power system,as the core hub of energy industry,needs to realize digital transformation urgently.Constructing virtual simulation models of transformers and obtaining vibration simulation data to complement each other with vibration data of physical equipment can improve the intelligent operation and maintenance of transformers.This paper establishes a multi-physical field vibration simulation model of transformer from the vibration mechanism of transformer core and winding,analyzes the vibration characteristics of transformer core and winding and optimizes the location of vibration signal measurement points,establishes a virtual simulation model of transformer based on vibration signals,obtains vibration signals of transformer core and winding under various states,and realizes fault diagnosis of transformer core and winding by artificial intelligence algorithm.The main research contents are as follows.Firstly,according to the vibration mechanism and mathematical model of transformer core and winding,the electric-magnetic-force-acoustic coupling model of transformer core and winding is established,and the vibration deformation of transformer core and winding under electromagnetic field is analyzed by using the finite element method.And considering the attenuation effect of transformer oil on the vibration signal,the simulation obtained the vibration acceleration distribution cloud map of the transformer vibration signal propagated to the surface of the tank through the transformer oil medium,and optimized the measurement point location of the transformer vibration acceleration sensor according to this.After that,a virtual simulation model of transformer based on vibration signals is constructed,and a vibration signal acquisition and transmission platform is built,which can collect and transmit the vibration signals of transformers in actual operation in real time.The simulation model is validated and corrected by the measured vibration signals,and the vibration data of various transformer core and winding fault conditions are simulated and generated based on the virtual simulation model,which solves the problem of lack of fault data in the health monitoring database.Finally,considering the simulated and measured vibration data of the transformer,we construct the vibration signal dataset with virtual-real fusion,and propose a three-channel parallel convolutional computation-based long-short time memory neural network for transformer multi-point vibration data,and apply it to the fault diagnosis of transformer core and winding.The model combines the advantages of convolutional neural network and longshort time memory network,which can effectively extract vibration features from the original multi-measurement point vibration data and explore the implied feature parameters in the transformer multi-measurement point vibration data.Through example analysis,this paper verifies the fault diagnosis effect of the three-channel parallel convolutional long and short term memory neural network model on the multi-point virtual and real fusion vibration data set,and compares it with other methods,the results show that the model has higher accuracy and better feature extraction capability.