白胤廷, 王志强, 池亚平, 王天宇. 加密流量分类研究进展与展望[J]. 北京电子科技学院学报, 2025, 33(2): 31-48.
    引用本文: 白胤廷, 王志强, 池亚平, 王天宇. 加密流量分类研究进展与展望[J]. 北京电子科技学院学报, 2025, 33(2): 31-48.
    BAI Yinting, WANG Zhiqiang, CHI Yaping, WANG Tianyu. Research Progress and Prospects in Encrypted Traffic Classification[J]. Journal of Beijing Electronic Science and Technology Institute, 2025, 33(2): 31-48.
    Citation: BAI Yinting, WANG Zhiqiang, CHI Yaping, WANG Tianyu. Research Progress and Prospects in Encrypted Traffic Classification[J]. Journal of Beijing Electronic Science and Technology Institute, 2025, 33(2): 31-48.

    加密流量分类研究进展与展望

    Research Progress and Prospects in Encrypted Traffic Classification

    • 摘要: 本文系统综述了基于深度学习的加密流量分类技术的研究进展与未来发展方向。随着网络流量的加密化,传统的基于内容分析的流量分类方法逐渐失效,研究者们转向基于统计特征、行为模式和深度学习的分类方法。文章首先回顾了传统方法、基于机器学习的方法以及基于深度学习的加密流量分类技术,重点介绍了卷积神经网络(CNN)、生成对抗网络(GAN)、图神经网络(GNN)等深度学习模型在加密流量分类中的应用及其优势。接着,文章分析了当前加密流量分类面临的主要挑战,包括数据分布偏移、过拟合、对抗样本攻击以及无效节点过多等问题。最后,文章提出了未来的研究方向,如基于实时学习的自适应模型、对抗训练、联邦学习以及跨领域大数据模型的改进等,以提升加密流量分类的鲁棒性和泛化能力。

       

      Abstract: This paper provides a systematic overview of the research advancements and future perspectives in encrypted traffic classification techniques utilizing deep learning. With the increasing encryption of network traffic, traditional content-based traffic classification methods have gradually become ineffective. Consequently, researchers have shifted their focus to classification approaches based on statistical features, behavioral patterns, and deep learning. The paper starts with a review of traditional methods, machine learning-based techniques, and deep learning-based approaches for encrypted traffic classification, followed by highlighting the application and effectiveness of deep learning models—such as Convolutional Neural Networks(CNN), Generative Adversarial Networks(GAN), and Graph Neural Networks(GNN)—in classifying encrypted traffic. Subsequently, the paper analyzes the key challenges in encrypted traffic classification, including data distribution shifts, overfitting, vulnerability to adversarial sample attacks, and the presence of excessive invalid nodes. Finally, the paper outlines several promising future research directions, including the development of real-time adaptive learning models, adversarial training techniques, federated learning frameworks, and enhancements to cross-domain big data models, all aimed at improving the robustness and generalization performance of encrypted traffic classification systems.

       

    /

    返回文章
    返回