面向入侵检测的Taylor神经网络构建与分析

Construction and Analysis of Taylor Neural Network for Intrusion Detection

深度学习方法已成为网络入侵检测的重要手段,但现有深度学习模型无法挖掘出网络入侵数据特征值间隐藏的函数映射关系。对此,设计了Taylor神经网络模型(TNN)。利用Taylor公式对多项式函数的逼近能力与神经网络的优化能力对入侵数据特征间的关系进行挖掘与利用。首先,介绍Taylor神经网络的基本结构。为了将Taylor神经网络引入入侵检测领域,设计了Taylor神经网络层(TNL),并将其与传统深度神经网络结合构建Taylor神经网络模型。为优化Taylor公式的展开项数,引入人工蜂群算法,但传统的人工蜂群算法存在开采能力较差,易陷入“早熟”等问题,因此设计了一种基于高斯过程的人工蜂群算法。实验结果表明,基于Taylor神经网络的入侵检测算法在NSL-KDD和UNSW-NB15数据集上的准确率具有明显优势。

Deep learning methods have become an important means of network intrusion detection,but the existing deep learning models cannot dig out the hidden function mapping relationships among the characteristic values of network intrusion data.In this regard,this paper designs a Taylor neural network model(TNN).The Taylor formula is used to mine and utilize the relationship between the polynomial function approximation ability and the neural network optimization ability.Firstly,this paper introduces the basic structure of Taylor neural network.In order to introduce the Taylor neural network into the field of intrusion detection,the Taylor neural network layer(TNL)is designed and combined with the traditional deep neural network to build the Taylor neural network model.In order to optimize the number of expansion terms of Taylor formula,artificial bee colony algorithm is introduced,but the traditional artificial bee colony algorithm has problems such as poor mining ability and easy to fall into“premature”.An artificial bee colony algorithm based on Gaussian process is designed.Experimental results show that the accuracy of intrusion detection algorithm based on Taylor neural network has obvious advantages on NSLKDD and UNSW-NB15 datasets.