| Title |
AI-Based Time-Series Analysis of Earth Leakage Current for Risk Prediction in Electrical Installations |
| Authors |
정인주(Inju Jung) ; 송진우(Jinwoo Song) ; 양주란(Jooran Yang) ; 김형표(Hyungpyo Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.3.644 |
| Keywords |
Earth leakage current; LSTM; Predictive maintenance; Time-series forecasting; Smart power infrastructure |
| Abstract |
This study aims to analyze the temporal variation of earth leakage current under moisture-exposed conditions and to evaluate the predictive performance of an LSTM-based time-series model. Unlike experiments intended to induce insulation degradation, the leakage environment was artificially simulated through water-contact and immersion to validate the model’s ability to detect pattern changes. Long-term leakage current data were collected from lighting-load conditions and applied to an LSTM network for future-current prediction. The model demonstrated high similarity to the measured data and accurately forecasted increases in leakage current up to the 7~9mA range. These results indicate that gradual leakage-current increases can occur under moisture exposure and that such patterns can be effectively learned by the LSTM model, although the findings do not constitute direct evidence of insulation degradation. The proposed model employs a lightweight architecture suitable for MCU-based embedded systems, enabling real-time processing in low-resource environments. This highlights its applicability to IoT-based monitoring platforms and its potential contribution to predictive maintenance frameworks for electrical installations. |