Circuit Breaker Fault Analysis Based on Wavelet Packet Time - Frequency Entropy and LM Algorithm to Optimize BP Neural Network

Abstract

High-voltage circuit breakers are the most important control and protection measures in power systems, and their reliable operation is critical to the safety and stability of power systems. However, the high-voltage circuit breaker machinery often fails, the vibration signal of the high-voltage circuit breaker hides the rich fault information. The change of the vibration signal reflects the mechanical state of the circuit breaker, and the vibration vector feature vector extraction and classification is the most important problem in fault diagnosis. In this paper, a BP neural network and wavelet packet time - frequency entropy method based on LM optimization algorithm are proposed. The vibration signal of the circuit breaker is extracted and faulted. The vibration signal of the high voltage circuit breaker is decomposed by wavelet packet, and then the time-frequency entropy of the vibration signal is obtained as the eigenvector. The feature vector is input to the BP neural network optimized by LM to determine the working state and fault type of the circuit breaker. Experiments show that the BP neural network optimized by LM algorithm and wavelet packet time-frequency entropy can be used to judge the mechanical failure of high voltage side circuit breaker more efficiently.