Research On High Voltage Circuit Breaker Fault Diagnosis Based On Convolutional Neural Network

In power system,high voltage circuit breaker plays the function of control and protection.Therefore,to ensure the normal operation of high voltage circuit breaker is an extremely critical work.With the development of artificial intelligence technology,machine learning and deep learning algorithms have been gradually applied to the state monitoring and fault diagnosis of high voltage circuit breakers,making the monitoring and maintenance strategies of high voltage circuit breakers develop rapidly.In this paper,the high voltage circuit breaker with spring-operated mechanism is studied.Based on the vibration signal of the operating mechanism of the high-voltage circuit breaker and the current signal of the dividing and closing coil,the paper proposed a fault diagnosis method of the operating mechanism of the high-voltage circuit breaker based on the convolutional neural network and designed a fault monitoring and simulation system for hv circuit breaker.The main work of this paper is as follows:1)Working mechanism and fault analysis of high voltage circuit breaker.Taking the high-voltage circuit breaker with spring operating mechanism as the research object,the working mechanism and typical failure mechanism of the primary and secondary sides of the circuit breaker are analyzed and also analyzed the characteristics of the operating mechanism vibration signal and the closing coil signal of the high-voltage circuit breaker.This provides a theoretical basis for the fault diagnosis algorithm of hv circuit breaker.2)Fault diagnosis method based on vibration signal.The wavelet transform method is used to transform one-dimensional vibration signal into time-frequency graph for Alex Net convolutional neural network training fault diagnosis model.The highlight of this method is that it adopts the method of image recognition.Compared with the RBF(radial basis neural network)fault diagnosis method and SVM(support vector machine)fault diagnosis method constructed with the characteristic value of wavelet band energy,its recognition accuracy is better.3)Fault diagnosis method based on split and close coil signal.The mathematical model of fault simulation is built and the correctness of the simulation model is verified by simulation experiment.The time-current composite characteristic value of the current signal of the dividing and closing coil is extracted by using the improved empirical mode adaptive and time-domain extremum method,which is used for the training of 2d-cnn convolutional neural network and the fault diagnosis model is constructed.Based on 2D-CNN,another fault diagnosis method of1D-CNN convolutional neural network is constructed.Compared with the 2d-cnn fault diagnosis method,its highlight is that the 1d-cnn does not change the physical meaning of the signal itself,which makes the fault recognition rate more accurate.4)High voltage circuit breaker fault detection and simulation system is designed based on matlab-gui interface.Based on the preposition theory and fault diagnosis method,a fault detection and simulation system for circuit breakers is designed,which includes a state detection subsystem based on coil current,a simulation subsystem based on coil current,and a state detection subsystem based on vibration signal.