基于深度学习的网络安全入侵检测系统优化算法  

Optimization algorithm for network security intrusion detection system based on deep learning

作　　者：殷玲 YIN Ling(Bozhi Security Technology Co.,Ltd.,Nanjing 210000,China)

机构地区：[1]博智安全科技股份有限公司,南京210000

出　　处：《计算机应用文摘》2025年第6期167-169,共3页

摘　　要：针对现有基于深度学习的网络入侵检测系统在特征处理、模型结构和算法实现等方面存在的问题,文章提出了一种优化算法。该算法包含3个核心创新:设计了基于动态窗口和互信息的特征处理方案,以有效提取网络流量特征;提出了改进的CNN-LSTM混合模型,通过注意力机制和特征融合策略增强模型表达能力;优化了检测过程,引入置信度评估和批量处理机制以提升检测效率。实验结果表明,该算法在NSL-KDD数据集上取得了96.2%的检测准确率,较优化前提升4.9%,检测延迟从85.6ms降至32.3ms,显著提升了检测性能和效率。This paper proposes an optimization algorithm to address the problems in feature processing,model structure,and algorithm implementation of existing deep learning based network intrusion detection systems.This algorithm contains three core innovations:a feature processing scheme based on dynamic windows and mutual information is designed to effectively extract network traffic features.An improved CNN-LSTM hybrid model was proposed,which enhances the model̓s expressive ability through attention mechanism and feature fusion strategy.Optimized the detection process,introduced confidence evaluation and batch processing mechanisms to improve detection efficiency.The experimental results show that the algorithm achieved a detection accuracy of 96.2%on the NSL-KDD dataset,an improvement of 4.9%compared to before optimization.The detection delay decreased from 85.6 ms to 32.3 ms,significantly improving the detection performance and efficiency.

关 键 词：网络安全 入侵检测 深度学习 特征工程 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]