Research into prediction and influential factors of circuit breaker closing time using BFGS-NN

This study enhances the accuracy of controlled closing for 800 kV filter field circuit breakers by analyzing the factors affecting the closing time. A multi-factor prediction model was developed using the Broyden–Fletcher–Goldfarb–Shanno (BFGS) neural network algorithm, incorporating ambient temperature, dwell time, the number of actions, and working hours. On-site operational data were analyzed to build a circuit breaker action time database. The BFGS algorithm trained on these data generated a closing time prediction model, achieving rapid convergence and optimal fit during learning. Results show an average prediction error within ±0.21 ms, with the maximum error tolerance meeting controlled closing requirements. The prediction error decreased as more variables were integrated, confirming that ambient temperature (inverse correlation), dwell time (positive correlation), working hours, and action frequency significantly influence closing times. For instance, lowering the ambient temperature by 40 °C reduced the closing time by ∼1 ms, while increasing the dwell time to 168 h raised it by 2.5 ms. The model enables real-time prediction of closing times using dynamic operational parameters, improving phase-selection precision by 18% compared with conventional methods. This data-driven approach provides theoretical and technical support for optimizing controlled closing in ultra-high voltage systems.