A power communication network intrusion detection method and device

By combining a hybrid deep learning model of CNN and RNN, and utilizing BRGRU and ACWE sub-models, the problems of insufficient accuracy and small sample recognition capability in intrusion detection in power communication networks are solved, and efficient detection of complex attack behaviors is achieved.

CN120956470BActive Publication Date: 2026-06-19INFORMATION & COMM CO OF STATE GRID JILIN ELECTRIC POWER CO LTD +2
2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INFORMATION & COMM CO OF STATE GRID JILIN ELECTRIC POWER CO LTD
Filing Date
2025-08-08
Publication Date
2026-06-19

AI Technical Summary

⚠Technical Problem

Existing intrusion detection models for power communication networks face challenges in the power communication environment, including insufficient accuracy and the inability to identify small-sample attacks due to complex traffic patterns and diverse attack methods.

⚗Method used

A hybrid deep learning model combining CNN and RNN is adopted. CNN extracts local spatial features of the data, BRGRU extracts temporal features, and ACWE sub-model is used to provide adaptive anomaly category judgment weights to construct an intrusion detection method and device for power communication networks.

🎯Benefits of technology

It improves the accuracy and robustness of intrusion detection in power communication networks, enabling better identification of complex attack behaviors, especially rare anomaly types.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

This invention provides a method and apparatus for intrusion detection in power communication networks. It obtains multiple time-series input matrices based on multiple target traffic data from the power communication network and inputs them into a target traffic detection model. The model includes a CNN sub-model containing multiple convolutional layers with different kernel sizes to extract features at different scales and fully preserve the time-series matrix information. A BRGRU sub-model contains multiple GRU network layers to calculate corresponding candidate hidden states and activated hidden states from the input features, which are then input into the next GRU network layer. The candidate hidden states can supplement the problems of insufficient learning by upper layers, thereby solving the gradient vanishing and network degradation problems in GRU and improving the accuracy of feature extraction and state detection. An ACWE sub-model adaptively provides judgment weights for different types of anomalies, avoiding model bias towards broad categories and enhancing the model's accuracy in judging rare anomaly types.
Need to check novelty before this filing date? Find Prior Art