Mechanical Fault Identification of High-Voltage Circuit Breaker Based on Neural Network and Lifelong Learning

The mechanical state identification of the high-voltage circuit breaker (HVCB) is crucial for ensuring the stable operation of the power grid. The establishment of an intelligent fault diagnosis model is an important step in the digitalization of power equipment. Traditional machine learning has provided diverse methods for the classification of mechanical states of HVCB. However, the mechanical state of HVCB dynamically changes with the variation of operating conditions and the increase in operation cycles. Lifelong learning offers insights for the diagnosis model of HVCB to continuously adapt to the latest state. Therefore, a diagnosis model based on feedforward neural network (FNN) and elastic weight consolidation (EWC) is presented in this paper. FNN is utilized for feature learning of new tasks, while EWC preserves knowledge of old tasks. EWC balances the importance of new and old tasks in the training process and reduces the loss of old knowledge. Relevant numerical experiments are carried out to evaluate the performance of EWC, which uses features of opening travel curves as the fault database. The experimental results show that FNN-EWC effectively enhances the classification ability for the mechanical state of HVCB in the presence of data distribution differences between old and new tasks, as EWC mitigates the forgetting of prior knowledge during the acquisition of new tasks by the diagnostic model.