HIFI: Anomaly Detection for Multivariate Time Series with High-order Feature Interactions

Abstract

Monitoring complex systems results in massive multivariate time series data, and anomaly detection of these data is very important to maintain the normal operation of the systems. Despite the recent emergence of a large number of anomaly detection algorithms for multivariate time series, most of them ignore the correlation modeling among multivariate, which can often lead to poor anomaly detection results. In this work, we propose a novel anomaly detection model for multivariate time series with HIgh-order Feature Interactions (HIFI). More specifically, HIFI builds multivariate feature interaction graph automatically and uses the graph convolutional neural network to achieve high-order feature interactions, in which the long-term temporal dependencies are modeled by attention mechanisms and a variational encoding technique is utilized to improve the model performance and robustness. Extensive experiments on three publicly available datasets demonstrate the superiority of our framework compared with state-of-the-art approaches.

Spatio-Temporal Anomaly Detection with Graph Networks for Data Quality Monitoring of the Hadron Calorimeter

“…Deep learning has become effective for AD modeling because of its capability to capture complex structures, extract end-to-end automatic features, and scale for large data sets [ 1 , 2 ]. Several DL models have been proposed in the literature for diverse data types, such as structural [ 1 ], time series [ 7 , 8 , 9 , 12 , 13 , 16 , 27 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 ], image [ 10 , 26 ], graph network data [ 14 , 15 , 24 , 25 , 39 ], and spatio-temporal [ 10 , 14 , 15 , 17 , 18 , 19 , 20 , 21 , 22 , 24 , 25 , 39 ]. Spatio-temporal (ST) data are commonly collected in diverse domains, such as visual streaming data [ 17 , 18 , 19 , 20 , 21 , 22 , 23 ], transportation traffic flows [ 24 , 25 ], sensor networks [ 14 , 15 , 39 ], geoscience […”

“…The wide range of unsupervised DL AD methods discover anomalies in temporal context using density clustering on latent space [ 9 ], data reconstruction [ 9 , 13 , 30 ], and prediction [ 16 , 27 , 30 , 33 , 34 ]. Variants of recurrent neural networks (RNNs) [ 7 , 8 , 9 , 13 , 21 , 22 , 24 , 27 , 34 , 35 ], convolutional neural networks (CNNs) [ 9 , 18 , 19 , 20 , 21 , 27 , 30 , 33 , 34 ], generative adversarial networks (GANs) [ 12 , 29 , 35 , 36 ], graph neural networks (GNNs) [ 24 , 25 , 30 , 37 , 38 ], and transformers [ 37 ] have been explored and achieved competitive performance for multivariate temporal or ST AD.…”