An Improved Reconstruction-Based Multiattribute Contrastive Learning for Digital-Twin-Enabled Industrial System

Digital twin (DT) is a promising technology for responding to Industry 4.0 and realizing comprehensive automation and virtualization. In the Web3.0-powered 5G/6G era, the expansion of the industrial data and closer interaction among cross-industrial entities pose new security challenges for DT industrial systems. As a prevalent computing paradigm, Graph Anomaly Detection provides an effective solution to ensure the security of DT industrial systems. However, the existing unsupervised graph anomaly detection methods tend to treat multiple graph attributes in isolation during the reconstruction process, resulting in insufficient semantics and suboptimal reconstruction performance. To overcome these challenges, we propose a multiattribute contrastive learning framework, which realizes graph anomaly detection by capturing both graph attribute patterns and their hidden relationships. First, we use an improved multiattribute aligned reconstruction approach to represent the anomaly information effectively. Besides that, the positive instance aggregation-based contrastive constraints are proposed, which can reduce the loss generated by mappings between different data dimension in feature representation space. Finally, to verify our proposal, extensive experiments have been conducted on five benchmark datasets, and the results show that our method obtains the state-of-the-art performance.

I. Introduction

The Fourth Industrial Revolution is in bloom, and the emergence of the global Covid-19 pandemic has further accelerated the industrial digital transformation. People’s desire for full automation and virtualization in industry is increasingly intense. Under these conditions, as one of the key technologies, digital twin (DT) has received extensive attention and research [1]. DT provides a digital mirror of the elements and attributes in physical system. Through this way, the supervision and decision making of the real world can be made by the operation in the digital world. In the Web3.0-enabled 5G/6G era, the scale of industrial data is further expanded, and the cooperation among cross-industrial entities is much closer [2]. However, the communication medium becomes a potential weakness when facing data attacks, posing a greater challenge to the security of DT industrial systems [3], [4]. DT industrial systems need a more transparent and trusted way to exchange massive amounts of data. To achieve the above goals, blockchain, with its decentralized, tamper-proof, and traceable nature, ensures a secure exchange among different DT systems [2]. For the decentralized DT industrial systems, the security and reliability of the interaction process among entities are guaranteed through blockchain and various encryption methods. Despite this, when the interaction behavior itself exhibits anomalies, it is hard to address the challenge effectively using only blockchain technology. Meanwhile, as a natural graph structure, each entity in the DT industrial systems with decentralized architecture serves as a node, and the interactions among them form the edges in the graph. In this context, graph anomaly detection methods provide a practical solution to capture the abnormal behavior in the system by modeling the decentralized system as a graph and utilizing graph-based techniques.