Abstract: In order to solve the problems in current work that rely on the traditional method of traffic classification，such as low accuracy，limited application region，an effective approach for network traffic classification named GA-LM is proposed.This method employs the classification method based on neural network as the classification model of network traffic，and applies L-M algorithm which is an improved BP algorithms and Genetic Algorithm（GA） that optimizes neural network weights，which will speed up the convergence of the neural network and improve the classification performance.The experimental data sets that are collected from actual networks are conducted experiments.The results show that the convergence speed of the method is faster and GA-LM has better feasibility and high accuracy which can be used effectively to the application of network traffic classification.

Key words: traffic classification, statistical features, error back propagation, Levenberg-Marquardt algorithm, genetic algorithm

摘要： 针对传统的流量分类方法准确率低、开销大、应用范围受限等问题，提出一种有效的网络流量分类方法（GA-LM）。该方法将基于神经网络的分类方法作为网络流量的分类模型，采用L-M算法构造分类器，并用遗传算法优化网络初始连接权值，加速了网络收敛过程，提高了分类性能。通过对收集到的实际网络流量数据进行分类，实验结果表明GA-LM比标准BP算法和L-M算法的收敛速度快，具有较好的可行性和高准确性，从而可有效地用于网络流量分类中。

关键词: 流量分类, 统计特征, 误差反向传播, Levenberg-Marquardt算法, 遗传算法