A hidden data attack detection method based on MFLSTM

By using multi-feature long and short-term memory network (MFLSTM) and random forest scores to calculate feature criticality and connectivity, the problem of inability to effectively detect hidden data attacks in low-resource environments is solved, and higher detection accuracy and stability are achieved.

CN113868642BActive Publication Date: 2025-03-28BEIJING UNIV OF TECH
2 Cites 0 Cited by

Patent Information

Application Number
CN202111059127.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-10
Publication Date
2025-03-28
Estimated Expiration
2041-09-10

AI Technical Summary

Technical Problem

Existing intrusion detection technologies cannot effectively detect hidden data attacks in low-resource environments, and cannot effectively detect using the criticality and connection between data.

Method used

The intrusion detection method based on multi-feature long and short-term memory network (MFLSTM) is used to calculate the feature criticality and connection through random forest scores, and valuable feature data are screened for model training, and attack detection is performed using residual judgment method.

Benefits of technology

In a low-resource environment, the MFLSTM algorithm can more accurately detect hidden data attacks, improve the accuracy and stability of detection, and effectively detect hidden data attacks in both high-resource and low-resource environments.

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

Abstract

The present invention discloses a hidden data attack detection method based on a multi-feature long short-term memory network, including: step 1, feature criticality evaluation and feature connectivity evaluation composed of random forest out-of-bag data and heat map, to prepare for subsequent model training; step 2, using feature data with excellent criticality and connectivity at different times as input, to train the multi-feature long short-term memory network, and at the same time, filtering out data with parameters that do not meet the standards during training, so that the model can learn information between different feature data that is helpful for data prediction; step 3, after successfully performing attack detection, the present invention determines whether to trigger an alarm based on the frequency of the threshold exceeding the residual. The improved model proposed by the present invention is superior to the long short-term memory network model before the improvement in many aspects. The criticality and connectivity between the feature data added under the same sensor side can help the model detect abnormal attack behaviors and make timely and effective responses.
Need to check novelty before this filing date? Find Prior Art