Study On Mechanical Condition Monitoring Method Of High Voltage Circuit Breaker Based On Control Circuit Detection

High voltage circuit breaker is an indispensable equipment in the power system,which is very important to ensure the power supply reliability and safe operation of the power system.Its mechanical state is the main reason affecting the reliability.Aiming at the problems that traditional monitoring methods are difficult to detect on line,extract effective features,and perform complicated experimental steps,this paper proposes a Convolutional neural network(CNN)based online monitoring method for the mechanical state of high-voltage circuit breakers.The convolutional layer in the network is used to transform the characteristics of the vibration signal of the high-voltage circuit breaker,and the fault sensitive features are extracted effectively in combination with the ability of pooling layer to strengthen the important features.In order to visualize the extracted features,t SNE dimensionality reduction method is used to express the features of the nodes in the convolutional network.It can be seen from the feature spatial distribution that the convolutional neural network can effectively summarize the timing information in the vibration signal,which is conducive to the subsequent fault mode discovery.Based on the acceleration sensor,A/D conversion circuit,peripheral circuit(including power circuit,protection circuit)and DPS processor,this paper constructs A hardware system for monitoring the mechanical state of high voltage circuit breakers.The vibration signal of the mechanical state of the high voltage circuit breaker is collected by the acceleration sensor.Amplify the signal amplifying circuit of sensor acquisition,and then transformed using A/D conversion circuit,the analog signals into digital signals,and finally to get the high voltagecircuit breaker vibration signals to the computer,for the high voltage circuit breaker mechanical condition monitoring provides A good foundation,in order to better in the CNN network model designed in this paper to analyze vibration signal processing.The convolutional layer of CNN contains multiple filters.The input signal is convolved with the filter,and the output is combined with the weight of the previous layer of the network as a feature mapping.The convolutional layer adopts the sparse connection method,that is,the neurons are only connected with some neurons in the upper layer.Therefore,the convolutional neural network has the characteristic of weight sharing,which on the one hand reduces the number of neurons,and on the other hand reduces the time complexity of the whole network.In this paper,the mechanical characteristics of zw32-12 fg /630-20 vacuum circuit breaker are studied by means of on-line monitoring and fault diagnosis.Failure experiments of trip closed electromagnetic plugging,spindle plugging and semi-shaft plugging were set up on vacuum circuit breakers.Receiver Operator Characteristic curve(ROC),Precision Recall curve(PRC),Gain Lift curve and lorentz curve were used as criteria for comparative analysis with traditional SVM online monitoring.The results show that CNN online monitoring model can effectively extract fault signal characteristics,so it can judge the type of mechanical fault of hv circuit breaker more quickly and accurately,and has better monitoring effect.