基于自适应神经网络的P2P流量检测研究

针对基于神经网络的P2P流量检测问题面临检测精度与样本训练时间相互矛盾的问题,提出了基于自适应的神经网络检测模型。通过对神经网络训练过程中的误差分析,改进神经网络的学习,动态调整神经网络的学习速率,以满足神经网络对预期结果的快速收敛性。详细阐述了应用于P2P流量检测的自适应神经网络构建过程,给出了自适应学习速率的相关计算公式。通过实验测试,结果表明采用自适应的神经网络具有训练速度快,检测精度高的特点,对7种典型P2P的识别精度基本达到了90%以上。

Study on P2P Stream Detection based on Auto-Adapted Neural Network

For the conflicts of between detection accuracy and sample training time in P2P stream detection based on neural network, this paper proposes an adaptive neural network detection model. Through the error analysis in the training process, the learning of neural network is improved, and the learning speed of neural network dynamically adjusted, thus to satisfy the fast convergence of the expected results. Moreover, this paper also gives the construction process of adaptive neural network applied in P2P stream detection, including the related computing formulas for the learning speed. The experiment show that the adaptive neural network is fast in training speed and high in detection accuracy, and for the seven typical P2P stream detections, its detection accuracy is up to more than 90%.