Hybrid Model of Multiple Echo State Network Integrated by Evidence Fusion for Fault Diagnosis of a High-Voltage Circuit Breaker

Clearance joints can reduce motion accuracy and shorten service life for high-voltage circuit breakers, and they are a significant source of mechanical faults. For clearance joint fault diagnosis, the anti-interference ability is important. However, robustness is rarely analyzed in current studies of fault diagnosis of high-voltage circuit breakers. Most conventional diagnosis models are based on individual classifiers with poor robustness, and their performance highly depends on expert prior information. In this paper, a novel Dempster-Shafer multiple echo state network (DS-MESN) is proposed to enhance the model accuracy and robustness of fault diagnosis. First, energy distributions in intrinsic mode functions by variational mode decomposition are considered as the inputting vector for diagnostic model training and testing. Then, multiple echo state network modules with different network parameters are adopted as subclassifiers, and their diagnostic outputs are fused by DS evidence theory. The DS-MESN can compromise the performance of multiple echo state network modules, making its performance more stable and objective. The superiority of the reported methodology is demonstrated by experimental data from a real ZN12 high-voltage circuit breaker.