{"title":"Multivariate Time Series Anomaly Detection via Temporal Encoder with Normalizing Flow","authors":"Jiwon Moon, S. Song, Jun-Geol Baek","doi":"10.1109/ICAIIC57133.2023.10067087","DOIUrl":null,"url":null,"abstract":"In the recent manufacturing process, as the introduction of smart factories spreads, high-dimensional data are being collected in real-time from various sensors of production facilities. However, existing anomaly detection models often do not reflect temporal factors, and even if they do, models that reflect temporal information are separately trained, resulting in a problem of falling into local optima. Therefore, it is very difficult to detect process anomalies in real-time by reflecting both correlations between high-dimensional variables and temporary dependency. This study proposes Temporal Encoder with Normalizing Flow (TENF), which can reflect both the correlation between variables and the time dependency in real-time using a relatively simple structure model. TENF consists of a Temporal Encoder for reflecting temporal dependencies and a NF Module for learning the distribution of high-dimensional data and is learned in an end-to-end manner. Experiments on multivariate time series data with similar characteristics to those generated in the manufacturing process demonstrate experimentally superior anomaly detection performance compared to existing models.","PeriodicalId":105769,"journal":{"name":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2023-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2023 International Conference on Artificial Intelligence in Information and Communication (ICAIIC)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICAIIC57133.2023.10067087","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
In the recent manufacturing process, as the introduction of smart factories spreads, high-dimensional data are being collected in real-time from various sensors of production facilities. However, existing anomaly detection models often do not reflect temporal factors, and even if they do, models that reflect temporal information are separately trained, resulting in a problem of falling into local optima. Therefore, it is very difficult to detect process anomalies in real-time by reflecting both correlations between high-dimensional variables and temporary dependency. This study proposes Temporal Encoder with Normalizing Flow (TENF), which can reflect both the correlation between variables and the time dependency in real-time using a relatively simple structure model. TENF consists of a Temporal Encoder for reflecting temporal dependencies and a NF Module for learning the distribution of high-dimensional data and is learned in an end-to-end manner. Experiments on multivariate time series data with similar characteristics to those generated in the manufacturing process demonstrate experimentally superior anomaly detection performance compared to existing models.