Incipient fault identification method for 10 kV power cables based on sheath current and DVAE-SAO-CatBoost

Abstract

In regard to the identification of incipient faults in 10 kV power cables, the existing researches based on data-driven methods are inadequate due to the oversimplification of the incipient fault model, which should be regarded as complex and multi-class faults. To address this issue, a comprehensive analysis of the evolution of incipient faults in 10 kV power cables is presented, and a novel method based on sheath current and denoising variational autoencoder-snow ablation optimizer-categorical boosting (DVAE-SAO-CatBoost) algorithm is proposed in this paper. Firstly, a novel dataset of complex and multi-class incipient faults for 10 kV power cables is developed based on sheath current via PSCAD/EMTDC software. Secondly, a DVAE network is employed for the purpose of feature extraction, thereby generating a feature dataset. Finally, a CatBoost network based on SAO algorithm for hyperparametric optimization is proposed as a classifier, with the feature dataset as input and the final output being the identification of the incipient faults. The results of the experiments demonstrated that the proposed method achieved near 100% identification accuracy in the environment with minimal noise. Even in a high noise environment with a signal-to-noise ratio of 5 dB and in a high data loss condition with a 25% data missing rate, the proposed method still demonstrated identification accuracies of 91.6% and 90.8%, respectively. Compared with the comparison methods, the proposed method consistently exhibited superior identification accuracy and robustness.