| Lyu He (吕 赫)*,Jia Xueqi**,Yu Xiaolong**,Xiong Kaizhou*.[J].高技术通讯(英文),2026,32(3):298~307 |
|
| A lightweight intrusion detection algorithm for train network control systems based on bi-level multi-objective optimization |
| |
| DOI：10. 3772 / j. issn. 1006-6748. 2026. 03. 009 |
| 中文关键词: |
| 英文关键词: train network control system, intrusion detection, multi-objective optimization,hyperparameter optimization, neural architecture search |
| 基金项目: |
| Author Name | Affiliation | | Lyu He (吕 赫)* | (*China Academy of Railway Sciences Locomotive & Car Research Institute, Beijing 100081, P. R. China)
(**Hangzhou Innovation Institute of Beihang University, Zhejiang Key Laboratory of Industrial Big Data and Robot Intelligent Systems, Hangzhou 311115, P. R. China) | | Jia Xueqi** | | | Yu Xiaolong** | | | Xiong Kaizhou* | |
|
| Hits: 60 |
| Download times: 65 |
| 中文摘要: |
| |
| 英文摘要: |
| Efficient intrusion detection in train network control systems (TNCS) must achieve a balance
between high detection accuracy and the limited computational capacity. To solve this challenge of
balancing, a lightweight intrusion detection algorithm for TNCS based on bi-level multi-objective op-
timization ( BLMOO-ID ) framework is developed. At the upper level of the framework , the non-
dominated sorting genetic algorithm-II ( NSGA-II ) is employed to explore neural architecture config-
urations and hyperparameter settings that balance the detection accuracy against the number of model
parameters. At the lower level, a joint-optimization loss composed of cross-entropy, a cosine - based
center loss, and L2 regularization is utilized to optimize each candidate model. Through iterative in-
teraction between the two levels, Pareto-efficient solutions are progressively produced, providing ar-
chitectures that achieve favorable trade-offs between detection accuracy and the model complexity.
Experiments on distributed denial of service ( DDoS ) attack datasets constructed from three publicly
available benchmark datasets, demonstrate that the proposed method significantly outperforms several
manually designed grid-searched baseline models, achieving 100% detection accuracy with only
23 610 parameters. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
