Hypergraph Learning based Recommender System for Anomaly Detection, Control and Optimization

TL;DR

This work tackles anomaly detection in high-dimensional multisensor time-series data typical of industrial systems. It introduces HgAD, a hypergraph-based framework that jointly learns dynamic hypergraph structure (HgSL), captures hierarchical spatiotemporal sensor representations (HgED), forecasts sensor values (HgF), and detects anomalies with robust deviation analysis (HgD). It further provides root-cause analysis through computation hypergraphs and offers offline prescriptive actions via HgPC using genetic algorithms. Empirically, HgAD achieves state-of-the-art performance across diverse benchmarks (e.g., SWaT, WADI, SMAP, MSL, TE P, HAI) and its ablations confirm the importance of structure learning, attention, and hypergraph pooling for accurate, interpretable anomaly detection.

Abstract

Anomaly detection is fundamental yet, challenging problem with practical applications in industry. The current approaches neglect the higher-order dependencies within the networks of interconnected sensors in the high-dimensional time series(multisensor data) for anomaly detection. To this end, we present a self-adapting anomaly detection framework for joint learning of (a) discrete hypergraph structure and (b) modeling the temporal trends and spatial relations among the interdependent sensors using the hierarchical encoder-decoder architecture to overcome the challenges. The hypergraph representation learning-based framework exploits the relational inductive biases in the hypergraph-structured data to learn the pointwise single-step-ahead forecasts through the self-supervised autoregressive task and predicts the anomalies based on the forecast error. Furthermore, our framework incentivizes learning the anomaly-diagnosis ontology through a differentiable approach. It derives the anomaly information propagation-based computational hypergraphs for root cause analysis and provides recommendations through an offline, optimal predictive control policy to remedy an anomaly. We conduct extensive experiments to evaluate the proposed method on the benchmark datasets for fair and rigorous comparison with the popular baselines. The proposed method outperforms the baseline models and achieves SOTA performance. We report the ablation studies to support the efficacy of the framework.

Figures (10)

• Figure 1: The $\text{HgAD}$ framework operates on the time-varying hypergraph-structured data $\mathcal{G}^{(t)}$ at time step t and predicts it as normal or not-normal.
• Figure 2: The hypernodes represent the multiple IoT sensors(denoted by $v_{i}$). The hyperedges(denoted by $e_{i}$) connect an arbitrary number of hypernodes.
• Figure 3: Overview of the HgAD framework. The differentiable Hypergraph Structure Learning(HgSL) module learns the discrete optimal hypergraph structure to facilitate the downstream anomaly detection. The HgED module learns the higher-order dependencies and hypergraph forecasting module further distills the knowledge for lower forecasting error. The Hypergraph-Deviation(HgD) module flags the anomalous behaviour based on single time-point forecast error.
• Figure 4: The HgCNN operator performs the neural message-passing schemes on the hypergraphs.
• Figure 5: The HgPool operator performs a downsampling operation. For illustration purposes, we rejected the hypernodes($v_2, v_6$) and the corresponding hyperedges($e_2, e_6$) in the hypergraph(left) to obtain the pooled hypergraph(right).