Research on Compensation Capacitor Fault Diagnosis Method Based on WPD-CNN
In order to further mine the characteristics of compensating capacitance contained in dynamic detection data,a fault diagnosis method of compensating capacitor based on WPD-CNN is proposed for ZPW-2000A track circuit,combined with wavelet packet decomposition and convolution neural network.The frequency band range of trend term and compensation capacitance in the detection curve is found out by using the method of power spectrum analysis.Then the original signal is decomposed by wavelet packet decomposition method,and the wavelet packet coefficients in the characteristic frequency band are extracted to construct the compensation capacitance characteristic matrix.The training set and test set are constructed by using dynamic detection data,and the characteristic matrices of different fault types are input into the convolution neural network for training and learning,and verified on the test set.The experimental results show that the WPD-CNN method extracts the features of a single signal with only 5.9ms,and the overall fault identification accuracy is 98.4%.It can effectively identify the faults of compensation capacitors in different positions and provide a basis for fault diagnosis of compensation capacitors.
万方数据
