Fault identification and localization in transmission lines using an IVY-optimized generalized additive model

Abstract

Accurate fault-type identification and fault-distance estimation are essential for the reliable operation of transmission lines, yet their simultaneous realization remains challenging under nonlinear transient responses and limited labeled data. This study proposes an IVY-optimized generalized additive model (IVY-GAM) framework that combines physically guided feature construction, interpretable nonlinear modeling, and adaptive hyperparameter optimization. A 200 km two-terminal transmission-line model is established in MATLAB/Simulink to generate 385 fault samples covering 11 fault categories. Twelve features are constructed from phase-mode transformation, db4 wavelet decomposition, transient modulus maxima, three-phase voltage statistics, and double-ended wavefront-arrival indices. IVY is then used to optimize the key GAM parameters for fault identification and localization. Under a fixed train–validation–test partition, IVY-GAM achieves an identification accuracy of 99.987% and a localization $$R^2$$ of 0.99674, with an MAE of 3.16 km, an RMSE of 3.91 km, and a maximum error of 11.84 km. Stratified five-fold out-of-fold evaluation further yields an identification accuracy of 99.22% and a macro-averaged F1-score of 0.992, while the localization task obtains a mean MAE of 7.08 km and a mean $$R^2$$ of 0.9520. Feature-group ablation, permutation importance, and GAM main-effect analyses show that three-phase voltage means primarily support fault-type discrimination, whereas double-ended arrival indices dominate distance estimation and modal maxima provide complementary information. The proposed framework therefore offers a balanced solution for accurate, interpretable, and computationally efficient transmission-line fault diagnosis.