상세정보
학술대회
Fast Anomaly Detection Model based on Long Sequences for Energy Storage Systems
- 발행일
- 202110
- 출처
- International Conference on Information and Communication Technology Convergence (ICTC) 2021, pp.1091-1094
- 협약과제
- 21PR4100, 친환경에너지 공급자원 제어시스템 개발, 신철호
- 초록
- This paper presents a fast anomaly detection model to detect anomalies immediately for energy storage systems (ESSs). ESS anomalies may cause performance degradation and even more serious failures like thermal runaway. However, to immediately detect anomalies in multivariate time series, shorter time steps are required resulting in longer sequences of time series data. Long short-term memory (LSTM) networks with long sequences of input variables may introduce gradient varnishing and exploding problems. In this paper, we exploit convolutional neural network (CNN) related models, such as CNN, CNN-LSTM and ConvLSTM, to alleviate the problems. The experimental results show that ConvLSTM outperforms other models, which can detect anomalies for long sequences of more than 576 time steps.
- KSP 제안 키워드
- Convolution neural network(CNN), Detection model, Long-short term memory(LSTM), Multivariate time series, Thermal runaway, Time series data, anomaly detection, energy storage system, input variables, long sequence, model-based