When transformers or shunt capacitors are energized on no-load, unknown residual flux or residual voltage can provoke severe inrush current. The resulting high peaks may cause relay mal-operation and endanger equipment. Although controlled closing can effectively reduce inrush current, it demands extremely precise control of the closing angle. Owing to mechanical scatter and environmental influences, an SF₆ breaker’s closing time is inherently dispersed. This paper proposes a neural-network predictor trained with historical data and ambient variables; an Augmented Consistency Regularization Neural Network (ACR-NN) is proposed to cope with the limited data set. Tests show a prediction error below 0.5 ms, fully satisfying the requirement for inrush suppression and verifying the algorithm’s practicality and effectiveness.

inrush-current suppression,neural network,prediction algorithm,ACR-NN,SF6 circuit breaker