Compensation capacitor fault diagnosis method based on time-frequency enhanced residual network
To address the issue of low fault diagnosis accuracy in existing compensation capacitor fault diagnosis methods for track circuits under high noise interference in complex environments,an intelli-gent fault diagnosis algorithm based on transfer learning,Continuous Wavelet Transform(CWT),and Time-Frequency Enhanced Residual Network(TFEResNet)is proposed.First,CWT is employed to integrate the time-domain and frequency-domain information of the original induced voltage signal,generating a wavelet time-frequency map.This map effectively enhances the model's ability to cap-ture fault characteristics by mapping compensation capacitor fault features to local positions at different times and scales.The wavelet time-frequency map is then input into the constructed TFEResNet model for transfer learning training,which is used for feature extraction and fault classification.TFER-esNet can extract complex time-frequency features from the map,mitigating the adverse effects of re-dundant and irrelevant noise in the signal,thereby improving diagnosis accuracy and generalization ca-pability of the model.Experimental results show that,in high-noise environments,the proposed algo-rithm outperforms other methods in compensation capacitor fault diagnosis,achieving an accuracy of 99.28%.Additionally,it shows superior performance in precision,recall,and F1-score,demonstrat-ing the effectiveness of the method and providing a novel approach for data-driven compensation ca-pacitor fault diagnosis in track circuits.
万方数据
维普
