| Title |
Development of Meta Learning Model for Anomaly Detection of Time-series Data |
| Authors |
김민수(Minsu Kim) ; 김성섭(Seongseop Kim) ; 이승우(Seungwoo Lee) ; 권영민(Youngmin Kwon) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.2.47 |
| Keywords |
Meta-learning; Few-shot learning; Time-series anomaly detection; Multi-time scale feature extraction; Prototypical Network |
| Abstract |
As energy infrastructure expands, the importance of time-series anomaly detection technology is increasing. However, existing deep learning-based models require large amounts of labeled data, and significant time and cost are needed for data collection and retraining when new anomaly types emerge. This paper proposes a meta-learning-based anomaly detection model that can rapidly adapt to new anomaly types with minimal example data. The proposed model compresses time-series data into low-dimensional representations through multi-time-scale feature extraction and applies Prototypical Network-based few-shot learning to classify anomaly types with only three samples per class. Experimental results demonstrate that the proposed model achieves higher classification accuracy than conventional models across various anomaly types, including leaks, nighttime usage surges, and lifestyle pattern changes. |