ETRI-Knowledge Sharing Plaform



논문 검색
구분 SCI
연도 ~ 키워드


학술대회 Comparative study in anomaly diagnosis technique for time series data
Cited 0 time in scopus Download 1 time Share share facebook twitter linkedin kakaostory
김낙우, 이현용, 이준기, 이병탁
International Conference on Consumer Electronics (ICCE) 2022 : Asia, pp.401-403
22ZK1100, 호남권 지역산업 기반 ICT 융합기술 고도화 지원사업, 강현서
In this paper, by comparing the values of the anomaly index in various anomaly diagnosis techniques on the same dataset, it is intended to help in selecting an anomaly diagnosis technique suitable for the desired task. We compare the models from the commonly used CNN or LSTM-based autoencoder anomaly diagnosis technique to its variant model DAGMM (Deep Autoencoding Gaussian Mixture model), deep learning-based SVDD (Support Vector Data Description), and recently USAD (Unsupervised Anomaly Detection) with each other. Since the anomaly index value at the time of occurrence of anomalies varies depending on the model, it is possible to consider the model selection according to the data distribution through these index values.
KSP 제안 키워드
Anomaly index, Data Distribution, Gaussian mixture Model(GMM), Index value, Learning-based, Support Vector Data Description, Time series data, comparative study, deep learning(DL), model selection, unsupervised anomaly detection