一种基于多级特征选择的工业互联网入侵检测方法
武汉纺织大学电子与电气工程学院, 湖北武汉 430200
| 摘 要： | 针对当前工业互联网入侵检测中因冗余特征及特征表征能力不足而导致的检测准确率低等问题，本文提出一种基于多级特征选择的工业互联网入侵检测方法。首先，利用皮尔逊相关系数（PCC）计算特征之间的线性相关程度，通过设定相关系数阈值剔除高度相关的冗余特征，完成特征子集的初步降维；随后，引入带自适应交叉、变异概率与精英保留策略的自适应精英遗传算法（AEGA），并基于兼顾分类性能与特征数量的适应度函数，对初步筛选后的特征子集进行二次筛选，得到最优特征集合。实验基于公开的TON-IoT网络数据集进行验证，原38维特征经过两级特征筛选后缩减至26维，特征缩减率达到31.58%，两级筛选协同作用，有效减小了特征冗余。并且在相同机器学习分类算法条件下，与互信息和粒子群优化等特征选择方法相比，本文方法结合随机森林和K近邻分类器均取得更优检测性能。为进一步提高分类准确率，充分挖掘优选特征的表征信息，基于优化后的特征集合构建CNN-GRU融合模型，检测准确率达到了95.61%，相较于单一CNN和GRU网络模型分别提高了2.19%和0.47%，验证了融合模型在入侵检测中的有效性。实验结果表明，本文提出的多级特征选择方法能够有效减少数据冗余、降低模型计算开销，为资源受限工业终端提供轻量化、高精度的入侵检测解决方案。 |
| 关 键 词： | 入侵检测; 多级特征选择; 遗传算法; 深度学习 |
| DOI： | 10.57237/j.cst.2026.04.001 |
View:
Downloads:
收稿日期: 2026-07-26
接受日期: 2026-08-17
在线出版日期: 2026-11-02
An Intrusion Detection Method Based on Multi-level Feature Selection for Industrial Internet
College of Electronic and Electrical Engineering, Wuhan Textile University, Wuhan 430200, China
| Abstract: | To address the limited detection accuracy of industrial Internet intrusion detection systems caused by feature redundancy and inadequate feature representation, this study proposes a multi-stage feature selection-based intrusion detection method for the industrial Internet. First, the Pearson correlation coefficient is employed to quantify the linear correlations among features, and highly correlated redundant features are eliminated according to a predefined correlation threshold, thereby achieving preliminary dimensionality reduction. Subsequently, an adaptive elite genetic algorithm, incorporating adaptive crossover and mutation probabilities together with an elite preservation strategy, is introduced to further optimize the feature subset. A fitness function that jointly considers classification performance and the number of selected features is designed to identify the optimal feature subset. Experiments are conducted on the publicly available TON-IoT network dataset. The proposed two-stage feature selection strategy reduces the original 38 features to 26, achieving a feature reduction rate of 31.58% while effectively alleviating feature redundancy. Under identical classification settings, the proposed method demonstrates superior detection performance with both Random Forest and K-Nearest Neighbors classifiers compared with mutual information- and particle swarm optimization-based feature selection methods. Furthermore, to exploit the discriminative information contained in the selected features and further improve detection accuracy, a CNN-GRU hybrid model is developed based on the optimized feature subset. The proposed CNN-GRU model achieves a detection accuracy of 95.61%, outperforming the standalone CNN and GRU models by 2.19% and 0.47%, respectively. These results demonstrate the effectiveness of the proposed framework in reducing feature redundancy and improving intrusion detection performance, while reducing the dimensionality of the model input and associated computational overhead. The proposed approach provides a potential solution for efficient and accurate intrusion detection in resource-constrained industrial environments. |
| Keywords: | Intrusion Detection; Multi-Level Feature Selection; Genetic Algorithm; Deep Learning |
参考文献
| [1] | Nuaimi M, Fourati L C, Hamed B B. Intelligent approaches toward intrusion detection systems for Industrial Internet of Things: A systematic comprehensive review [J]. Journal of Network and Computer Applications, 2023, 215: 103637. https://doi.org/10.1016/j.jnca.2023.103637 |
| [2] | Venkatesan S. Design an intrusion detection system based on feature selection using ML algorithms [J]. Mathematical Statistician and Engineering Applications, 2023, 72(1): 702-710. https://philstat.org/index.php/MSEA/article/view/2000 |
| [3] | Thakkar A, Lohiya R. A survey on intrusion detection system: feature selection, model, performance measures, application perspective, challenges, and future research directions [J]. Artificial Intelligence Review, 2022, 55(1): 453-563. https://doi.org/10.1007/s10462-021-10037-9 |
| [4] | Turukmane A V, Devendiran R. M-multisvm: An efficient feature selection assisted network intrusion detection system using machine learning [J]. Computers & Security, 2024, 137: 103587. https://doi.org/10.1016/j.cose.2023.103587 |
| [5] | Barhoush M, Abed-alguni B H, Al-qudah N E A. Improved discrete salp swarm algorithm using exploration and exploitation techniques for feature selection in intrusion detection systems [J]. The Journal of Supercomputing, 2023, 79(18): 21265-21309. https://doi.org/10.1007/s11227-023-05444-4 |
| [6] | Singh S. Poly logarithmic naive Bayes intrusion detection system using linear stable PCA feature extraction [J]. Wireless Personal Communications, 2022, 125(4): 3117-3132. https://doi.org/10.1007/s11277-022-09701-2 |
| [7] | Ghasemi H, Babaie S. A new intrusion detection system based on SVM–GWO algorithms for Internet of Things [J]. Wireless Networks, 2024, 30: 2173-2185. https://doi.org/10.1007/s11276-023-03637-6 |
| [8] | Benaddi H, Jouhari M, Elharrouss O. A lightweight hybrid approach for intrusion detection systems using a chi-square feature selection approach in IoT [J]. Internet of Things, 2025, 32 101624-101624. https://doi.org/10.1016/j.iot.2025.101624 |
| [9] | Disha R A, Waheed S. Performance analysis of machine learning models for intrusion detection system using Gini Impurity-based Weighted Random Forest (GIWRF) feature selection technique [J]. Cybersecurity, 2022, 5(1): 1 https://doi.org/10.1186/s42400-021-00103-8 |
| [10] | Alshehri M S, Saidani O, Alrayes F S, et al. A Self-Attention-Based Deep Convolutional Neural Networks for IIoT Networks Intrusion Detection [J]. IEEE Access, 2024. https://doi.org/10.1109/ACCESS.2024.3380816 |
| [11] | Xu B, Sun L, Mao X, et al. IoT intrusion detection system based on machine learning [J]. Electronics, 2023, 12(20): 4289. https://doi.org/10.3390/electronics12204289 |
| [12] | Liu Y, Deng W, Liu Z，et al. Semi-supervised attack detection in industrial control systems with deviation networks and feature selection [J]. The Journal of Supercomputing, 2024, 80(10): 14600-14621. https://doi.org/10.1007/s11227-024-06018-8 |
| [13] | Nandhini U, SVN K S. An improved Harris Hawks optimizer based feature selection technique with effective two-staged classifier for network intrusion detection system [J]. Peer-to-Peer Networking and Applications, 2024, 17 (5): 2944-2978. https://doi.org/10.1007/s12083-024-01727-6 |
| [14] | Mishra A K, Paliwal S. Mitigating cyber threats through integration of feature selection and stacking ensemble learning: the LGBM and random forest intrusion detection perspective [J]. Cluster Computing, 2023, 26(4): 2339-2350. https://doi.org/10.1007/s10586-022-03735-8 |
| [15] | 王相月, 赵利辉. 基于多阶段特征选择和CNN-GRU的网络入侵检测模型 [J]. 中北大学学报 (自然科学版), 2024, 45(01): 6673. https://doi.org/10.3969/j.issn.16733193.2024.01.009 |
