摘要详情

ID / 提交时间

262 / 2025-06-15 18:11:39

标题

Anomaly Detection via Heterogeneous Graph Attention Network for Space-based Cloud Service Platform

关键字

Space-based Platform,Graph Anomaly Detection,Heterogeneous Graph Attention Network,Spatial-temporal modeling

主题及专题

Special Sessions > SSL 01 Advanced signal processing enabled intelligent fault diagnosis and measurement

状态

全文待审

作者

Yuangui Yang / Xi'an Jiaotong University

Chenye Hu / Xi'an Jiaotong University

Xiangyuan Min / Xi'an Jiaotong University

Yuan Qiu / Shanghai Key Laboratory of Collaborative Computing in Special Hegerogenous Networks Aerospace Electronic Technology Institute

Hui Wang / Shanghai Aerospace Electronic Technology Institute

摘要

Reliability of space-based information networks relies on accurate cloud service data anomaly detection. However, existing algorithms neglect the high-dimensional, multi-modal, and dynamically coupled properties of those data, leading to low accuracy and poor interpretability. To solve this issue, we propose an anomaly detection method using a multi-scale dual-stream heterogeneous graph attention network (MDHAN). First, a multi-scale feature extraction module is designed to accurately capture the local details and global trends of the data through different convolutional kernels. Second, the spatial-temporal graph attention feature extraction module is constructed to deeply excavate the heterogeneous relationship and dynamic temporal dependence among features. Finally, a joint “reconstruction + prediction” optimization objective is proposed to generate a comprehensive and integrated anomaly score. The experimental results show that MDHAN can significantly reduce the false negative rate and false positive rate. It provides an efficient and interpretable anomaly detection paradigm for the autonomous operation and maintenance of cloud service systems, and promotes the development of reliability assurance technology for space-based information networks.