High-Voltage Circuit Breaker Fault Diagnosis Using a Hybrid Feature Transformation Approach Based on Random Forest and Stacked Autoencoder

• Published in: IEEE transactions on industrial electronics (1982) Vol. 66; no. 12; pp. 9777 - 9788
• Main Authors: Ma, Suliang, Chen, Mingxuan, Wu, Jianwen, Wang, Yuhao, Jia, Bowen, Jiang, Yuan
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Binary codes; Circuit breakers; Circuit faults; Fault diagnosis; Feature extraction; feature transformation; High voltages; high-voltage circuit breaker (HVCB); Machine learning; Mapping; Neural networks; Optimization; Radio frequency; random forest (RF); stacked autoencoder (SAE); Support vector machines; Transformations; Vibrations; Wavelet analysis; wavelet packet decomposition; Wavelet transforms
• ISSN: 0278-0046; 1557-9948
• DOI: 10.1109/TIE.2018.2879308

In recent years, machine learning techniques have been applied to test the fault type in high-voltage circuit breakers (HVCBs). Most related research involves in improving the classification method for higher precision. Nevertheless, as an important part of the diagnosis, the feature information description of the vibration signal of an HVCB has been neglected; in particular, its diversity and significance are rarely considered in many real-world fault-diagnosis applications. Therefore, in this paper, a hybrid feature transformation is proposed to optimize the diagnosis performance for HVCB faults. First, we introduce a nonlinear feature mapping in the wavelet package time-frequency energy rate feature space based on random forest binary coding to extend the feature width. Then, a stacked autoencoder neural network is used for compressing the feature depth. Finally, five typical classifiers are applied in the hybrid feature space based on the experimental dataset. The superiority of the proposed feature optimal approach is verified by comparing the results in the three abovementioned feature spaces.