Anomaly Detection for Multivariate Times Series Data of Aero-engine Based on Deep LSTM Autoencoder
Guoqing Zhu, Lin Huang, Dongliang Li, Li Gong · 2024
The structure of modern industrial equipment is usually complex, which will lead to data explosion and multivariate time series problems. An approach of anomaly detection for multivariate time series based on a deep autoencoder is proposed. In the proposed approach, long short-term memory is employed to optimize the autoencoder model. Data could be mapped between multi-dimensional feature space and low-dimensional latent space while extracting the time information. The bottleneck structure of the autoencoder is used to reconstruct the time series in the latent space. Then, the reconstruction error was calculated and taken as the basis for anomaly detection. In this way, anomaly detection for multivariate time series is realized. Ultimately, the proposed approach is verified by the dataset from an aero-engine simulation released by NASA. It is comparatively analyzed with several typical published approaches that have high detection accuracy to demonstrate its effectiveness.