Network traffic classification method and device for heterogeneous environment, terminal and storage medium

Abstract

The invention discloses a network traffic classification method and device for a heterogeneous environment, a terminal and a storage medium. The method comprises the following steps: distributing a global traffic classification module of a target network environment to each client; the sampling value of each client is obtained, and the sampling value of each client is determined based on the model skewness of the client; determining a target client according to the sampling value of each client; and updating the global traffic classification model according to each target client to obtain a target global traffic classification model. The target clients are selected through the model skewness, and the global traffic classification model is updated through the local traffic classification model of each target client, so that the problem that the existing mobile network traffic classification based on federated learning cannot eliminate the influence that clients with large data distribution differences participate in averaging in model aggregation in a heterogeneous environment can be solved, and the traffic classification efficiency is improved. And the flow classification precision is reduced. 本发明公开了异构环境的网络流量分类方法、装置、终端及存储介质，通过将目标网络环境的全局流量分类模分发至各客户端；获取各客户端的采样值，其中，每一客户端的采样值基于该客户端的模型偏度确定；根据各客户端的采样值确定目标客户端；根据各目标客户端对全局流量分类模型进行更新，得到目标全局流量分类模型。本发明通过模型偏度选取目标客户端，再通过各目标客户端的局部流量分类模型更新全局流量分类模型，可以解决现有的基于联邦学习的移动网络流量分类在异构环境下，无法消除模型聚合中数据分布差异大的客户端参与平均的影响，导致流量分类精度下降的问题。