To address the issues of low computational accuracy and high randomness in loop-closing current calculations

this study proposes an interval prediction method for loop-closing current that incorporates uncertainty. First

to overcome the complexity of data caused by numerous influencing factors in loop-closing current prediction

an adaptive LASSO regression-based variable selection method is introduced to identify key factors and construct a multivariate dataset capable of effective loop-closing current prediction. Second

a feature extraction module is developed using temporal convolutional networks and graph convolutional networks to fully exploit global feature information from the multivariate data. Subsequently

a bidirectional gated recurrent unit neural network is employed to capture long-term dependencies among multivariate data for the time-period prediction of loop-closing current. The prediction errors are subjected to probability density estimation

and the final interval prediction results are obtained by superimposing the time-interval predictions. Finally

case study simulations validate the superiority and feasibility of the proposed method.