An industrial control protocol identification method based on a fusion BERT and CNN network

By integrating BERT and CNN networks to identify industrial control protocols, the method converts industrial control protocol messages into grayscale images and combines the loss functions of the two models, thus solving the accuracy and applicability problems of existing industrial control protocol identification technologies and achieving higher-precision protocol identification.

CN116389614BActive Publication Date: 2026-06-19GUANGDONG UNIV OF TECH
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
GUANGDONG UNIV OF TECH
Filing Date
2023-04-04
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing industrial control protocol identification methods suffer from high time and space complexity, high false alarm rate, and insufficient applicability in complex network environments, making it difficult to achieve high-precision identification of multiple protocols.

Method used

An industrial control protocol recognition method based on the fusion of BERT and CNN networks is adopted. By converting the industrial control protocol message data into grayscale images and combining BERT and CNN classification models, the loss functions of the two networks are fused to achieve higher accuracy in protocol recognition.

Benefits of technology

It achieves high-precision identification of more complete information in industrial control protocol messages, can identify multiple industrial control protocols, and improves the identification effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116389614B_ABST
    Figure CN116389614B_ABST
Patent Text Reader

Abstract

This invention belongs to the field of Industrial Internet of Things (IIoT) and discloses a method for identifying industrial control protocols based on a fusion of BERT and CNN networks. This method is used to identify complex protocol messages in an industrial control environment. The protocol identification method first extracts n... 2 One byte, and will be truncated to n 2 Each byte of data is converted into an n×n grayscale image through value mapping. The byte data and grayscale image are then input as dual channels into classification model 1 and classification model 2 respectively for training. The two training results are then fused to obtain a trained model. By capturing data packets in a real-time industrial environment, the above model is used for inference, achieving accurate recognition results. This invention can automatically recognize input protocol messages, achieving high-precision recognition results while preserving more complete information in the message. Compared to manually selecting message field features, it reduces information loss.
Need to check novelty before this filing date? Find Prior Art