基于改进遗传算法的入侵检测技术的设计与实现  

Design and implementation of intrusion detection technology based on improved genetic algorithm

作　　者：王硕 李成杰 崔丽琪 李聪[3] 乐秀权 戴志坚[4] WANG Shuo;LI Chengjie;CUI Liqi;LI Cong;YUE Xiuquan;DAI Zhijian(School of Computer Science and Engineering,Southwest University for Nationalities,Chengdu Sichuan 610225,China;National Key Laboratory of Wireless Communication,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China;Institute of Telecommunication and Navigation Satellites,China Academy of Space Technology,Beijing 100094,China;Shenzhen Institute for Advanced Study,University of Electronic Science and Technology of China,Shenzhen Guangdong 518110,China)

机构地区：[1]西南民族大学计算机科学与工程学院,四川成都610225 [2]电子科技大学通信抗干扰全国重点实验室,四川成都611731 [3]中国空间技术研究院通信与导航卫星总体部,北京100094 [4]电子科技大学(深圳)高等研究院,广东深圳518110

出　　处：《太赫兹科学与电子信息学报》2024年第3期249-260,共12页Journal of Terahertz Science and Electronic Information Technology

基　　金：中央高校基本科研业务费专项基金优秀学生培养工程资助项目(2023NYXXS034);基础加强资助项目(2020-JCJQ-ZD-119)。

摘　　要：天地一体化网络处在开放的电磁环境中,会时常遭受恶意网络入侵。为解决网络中绕过安全机制的非授权行为对系统进行攻击的问题,提出一种改进的遗传算法。该算法以决策树算法为适应度函数,通过删除数据集中的冗余特征,显著提高了对网络攻击的拦截率。通过机器学习进行异常分类,并利用遗传算法的特征选择功能,增强机器学习方法的分类效率。为验证算法的有效性,选用UNSW_NB15和UGRansome1819数据集进行训练和检测。使用随机森林、人工神经网络、K近邻和支持向量机等4种机器学习分类器进行评估,采用准确性、F1分数、召回率和混淆矩阵等指标评估算法的性能。实验证明,遗传算法作为特征选择工具能够显著提高分类准确性,并在算法性能上取得显著改善。同时,为解决弱分类器的不稳定性,提出一种集成学习优化技术,将弱分类器和强分类器集成进行优化。实验证实了该优化算法在提高弱分类器稳定性方面性能卓越。For addressing the issue of unauthorized actions bypassing security mechanisms to attack systems in the integrated network of heaven and earth in the open electromagnetic environments,an improved Genetic Algorithm(GA)is proposed.It uses the Decision Tree(DT)algorithm as the fitness function,and significantly improves the interception rate of network attacks by deleting redundant features in the dataset.Anomaly classification is performed through machine learning,and the feature selection function of the genetic algorithm is employed to enhance the classification efficiency of machine learning.To verify the effectiveness of the proposed algorithm,the UNSW_NB15 and UGRansome1819 datasets are selected for training and testing.Four machine learning classifiers,namely Random Forest(RF),Artificial Neural Network(ANN),K-Nearest Neighbor(KNN),and Support Vector Machine(SVM),are used for evaluation.The performance of the algorithm is evaluated through indicators such as accuracy,F1 score,recall rate,and confusion matrix.The experiment results prove that the genetic algorithm as a feature selection tool can significantly improve the classification accuracy and achieve significant improvement in algorithm performance.Meanwhile,to tackle with the instability of weak classifiers,this paper further proposes an ensemble learning optimization technique,which integrates weak classifiers and strong classifiers for optimization.The experiment confirms the excellent performance of this optimization algorithm in improving the stability of weak classifiers.

关 键 词：机器学习 遗传算法 决策树 特征选择 

分 类 号：TN927[电子电信—通信与信息系统]