Hybrid Autoencoder-LSTM Augmentation for Improved Arc Fault Classification Under Data Scarcity

Le résumé fourni par la source

Arc faults pose serious risks in modern electrical systems, resulting in electrical failures and safety hazards. Although recent advances in artificial intelligence have boosted detection accuracy, obtaining diverse and sufficient fault data remains challenging. In this paper, we propose a hybrid Autoencoder-Long Short-Term Memory (LSTM) data augmentation algorithm to mitigate the scarcity of arc fault samples. We then integrate these augmented datasets into a lightweight classification framework, Efficient-ArcNet, to assess its multi-class detection performance. Our experimental results show a marked accuracy improvement for a specific fault class with limited original dataset, rising from 93.18% to 97.73%, while retaining high accuracy in other fault categories. By employing the proposed method, we effectively address the issue of limited fault data, thereby enhancing arc fault detection and improving overall system reliability.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.