Anomaly Detection Method for Time Series Data Based on Transformer Reconstruction
Open Access
- 17 February 2023
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 58-63
Abstract
Multiple temporal anomaly detection algorithms have important research significance in many application fields, such as system state estimation, fault prediction and diagnosis, network behavior anomaly detection and so on. Aiming at the problems of abnormal noise, high dimensionality, lack of labeling, and difficulty in learning abnormal features of various temporal data, an anomaly detection model TRAD based on Transformer reconstruction was proposed, which used self-conditioning to extract robust multi-modal features to obtain the stability of training. At the same time, the adversarial training process is used to amplify the reconstruction error. Experiments on three public datasets show that the proposed model not only has excellent detection performance, but also has strong applicability and generalization ability for unknown heterogeneous time series data.Keywords
This publication has 9 references indexed in Scilit:
- Architectural patterns for the design of federated learning systemsJournal of Systems and Software, 2022
- Network Intrusion Detection Model Based on CNN and GRUApplied Sciences, 2022
- Accurate real-time monitoring of fine dust using a densely connected convolutional networks with measured plasma emissionsChemosphere, 2022
- Graph Neural Network-Based Anomaly Detection in Multivariate Time SeriesProceedings of the AAAI Conference on Artificial Intelligence, 2021
- HitAnomaly: Hierarchical Transformers for Anomaly Detection in System LogIEEE Transactions on Network and Service Management, 2020
- USAD2020
- Matrix Profile XII: MPdist: A Novel Time Series Distance Measure to Allow Data Mining in More Challenging Scenarios2018
- WADI2017
- Efficient Anomaly Detection by Isolation Using Nearest Neighbour Ensemble2014