Application Of Time Series Deep Learning In Vibration Signal Analysis Of High Voltage Circuit Breaker

The high-voltage circuit breaker is a vital equipment in the power system,and its main faults come from the operating mechanism.This subject takes the high-voltage circuit breaker operating mechanism as the experimental research object,and comprehensively analyzes the feasibility and superiority of using its vibration signal for fault diagnosis.The article mainly conducts in-depth research on the fault identification and diagnosis of the vibration signal of the circuit breaker,and forms the experimental evidence.The main contents of the thesis are as follows:1.Mainly introduce and summarize related deep learning theories.First,it summarizes the operation-related principles and formula derivation of traditional convolutional neural network CNN;then summarizes the "three gates" related mechanisms and formulas of the long and short-term memory neural network LSTM,and the principle that it can remember the characteristics of longer-term sequences In the end,the relevant principles of the attention mechanism are summarized and the calculation formula of the attention value is given,and the reason why a little extra calculation is exchanged for a better model convergence effect is described.2.It mainly proposes the network structure and logical framework of the new time-series deep learning model CNN＿LSTM built by the long-and short-term memory network LSTM.Use the actual circuit breaker vibration signal to optimize the parameters of the CNN＿LSTM network model,verify the performance of the model by controlling some hyperparameters as a single variable,and compare the CNN＿LSTM deep learning model under the optimal parameters with other diagnostic algorithms,Get the advantages and disadvantages of the new time series deep learning model.3.It mainly introduces the principle structure of the attention mechanism SENet used and the ACNN＿LSTM time-series deep learning model jointly formed by the above CNN＿LSTM.Use the commonly used model performance evaluation indicators such as F1 value,loss value,ROC curve and PR curve to evaluate the newly proposed ACNN＿LSTM time series deep learning model,and compare the effects with multiple neural network diagnosis algorithm models,reflecting ACNN＿LSTM The superiority of neural network model performance and the efficiency of learning and diagnosis.This subject verified the influence of the network structure parameters in the CNN＿LSTM time series deep learning model on the performance of the model through actual experiments,and formed the optimal network model for the fault diagnosis of the vibration signal of the circuit breaker.In actual comparison with other deep learning models,the CNN＿LSTM time series deep learning model uses its time series memory ability in the data feature extraction stage to achieve better model performance.Aiming at the lower learning efficiency caused by the complex network structure of the above-mentioned CNN＿LSTM model,this topic proposes an ACNN＿LSTM time-series deep learning model based on the attention mechanism.Using the attention mechanism,the new model exchanges a small amount of additional calculations for more efficient and faster learning capabilities in the iterative learning process,which is also confirmed by the actual fault category diagnosis and performance comparison with other deep learning algorithms.