A method and apparatus for anomaly detection in small-sample cold source systems based on graph neural networks

By constructing a graph neural network-based method for detecting anomalies in cold source systems using small samples, a graph structure is built to extract node embedding features and perform self-supervised perturbation training. This solves the problems of false alarms, missed alarms, and insufficient early warning in existing technologies for detecting anomalies in cold source systems, and achieves highly sensitive anomaly detection for cold source systems.

CN122020261BActive Publication Date: 2026-06-30ZHEJIANG YUANSUAN TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG YUANSUAN TECH CO LTD
Filing Date
2026-04-13
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing methods for detecting anomalies in cold source systems rely on manual rules and static threshold judgments, which are prone to false alarms or missed alarms. Furthermore, supervised machine learning models require a large number of labeled samples, while traditional unsupervised detection algorithms are insufficient in their ability to model multivariate dynamic coupling relationships and time series trends, making it difficult to capture subtle trend anomalies and resulting in insufficient early warning capabilities.

Method used

A small-sample anomaly detection method for cold source systems based on graph neural networks is adopted. By constructing a graph structure to extract node embedding features, combining it with a long short-term memory network for temporal modeling, generating simulated positive and negative samples, performing self-supervised perturbation training, calculating multi-scale coupled anomaly scores, optimizing the latent feature space, obtaining coupled sensitivity vectors, and finally determining the anomaly detection results.

Benefits of technology

It significantly improves the sensitivity of anomaly detection in cold source systems, enabling the capture of multivariate dynamic coupling relationships and time series trend characteristics under small sample conditions, and achieving early detection and interpretable analysis of weak anomalies.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122020261B_ABST
    Figure CN122020261B_ABST
Patent Text Reader

Abstract

This invention provides a method and apparatus for anomaly detection in a small-sample cold source system based on a graph neural network, relating to the technical field of anomaly detection. The method includes: processing multi-source monitoring data to obtain time-series data; constructing a graph structure based on the physical coupling relationships between various monitoring indicators in the time-series data; extracting node embedding features from the graph structure using a graph neural network; performing time-series modeling on the node embedding features using a long short-term memory network to obtain a prediction vector; generating simulated positive and negative samples through self-supervised perturbation training of the graph structure; obtaining a coupling sensitivity vector based on the prediction vector, positive samples, and negative samples; calculating an anomaly sensitivity set based on the coupling sensitivity vector; determining a multi-scale coupling anomaly score based on the anomaly sensitivity set; and determining the target anomaly detection result based on the multi-scale coupling anomaly score. This invention can significantly improve the sensitivity of anomaly detection in cold source systems.
Need to check novelty before this filing date? Find Prior Art