ISA-MTAD: Improved structured autoencoder for multivariate time series anomaly detection

doi:10.1016/j.displa.2025.103271

ISA-MTAD: Improved structured autoencoder for multivariate time series anomaly detection

2025-10-30

0

PRE

AI

张东平 (Dongping Zhang) *

DOI：10.1016/j.displa.2025.103271

Original

Original request for help

Save

Abstract

En 中文

• A novel semi-supervised multivariate time series anomaly detection framework (ISA-MTAD) with dual decoder structure is proposed. • A two stage training strategy is employed to learn prior knowledge of normal data and enhance feature extraction of abnormal data. • The proposed method eliminates threshold setting by selecting decoders based on reconstruction error, improving robustness. • Extensive experiments on four datasets (KDD99, SWaT, WADl, SKAB) demonstrate superior performance overbaseline models.

ISA-MTAD: Improved structured autoencoder for multivariate time series anomaly detection

Abstract

Journal

Organization