Fast Informer-Based Time Series Detection
Weiguo Shen, Tianhao Xia, Dongwei Xu, Qi Xuan, Yun Lin, Wei Wang · 2024
Time series data are pivotal in modern technology and industrial applications,but they are frequently impacted by anomalies due to system failures,cyberattacks,or unexpected events. Therefore,swift and accurate anomaly detection in time series is crucial. This paper introduces ‘TransIDS’,an intrusion detection method that focuses on the efficient extraction and detection of time series features. TransIDS uses the Informer model to capture temporal features within data frames more effectively,improving detection performance and reducing runtime. We also fine-tune the Informer model and introduce an automatic anomaly threshold selection method,optimizing it for anomaly detection tasks. We validate TransIDS through experiments on various datasets and compare it with three existing methods. The results show that TransIDS outperforms the compared models in detection performance.