Life prediction of residual current circuit breakers using Bayesian optimization-based bidirectional LSTM
Guojin Liu, Xujing Wang, Jianhua Miao, Xu Liu, Chenxiao Wang
Abstract
Residual current circuit breakers (RCCBs) play a critical role in preventing electrical leakage accidents; however, accurately predicting their service life remains a challenge in power system safety management. This study proposes a Bayesian optimization-based Bidirectional Long Short-Term Memory (BO-BiLSTM) neural network to address this issue. Accelerated aging tests were conducted with temperature as the stress variable to analyze the degradation path of residual operating current. To optimize hyperparameter selection in Bidirectional Long Short-Term Memory (BiLSTM), a Bayesian optimization-based BO-BiLSTM model was introduced, significantly enhancing the accuracy of degradation trajectory predictions. The model successfully estimated the pseudo-failure lifetime of RCCBs at 55 °C and 65 °C and extrapolated predictions for 77 °C, with validation against accelerated aging test data confirming its reliability. Furthermore, by integrating the Arrhenius equation with accelerated life test results, the RCCB’s service life under room-temperature operating conditions (25 °C) was projected, demonstrating its applicability for real-world deployment in power distribution systems.