基于概率的并行粒子群AKO-RVM入侵检测  

Intrusion detection using automatic kernel width optimization RVM based on probabilistic parallel PSO

作　　者：马文海[1] 胡平[1] 

机构地区：[1]南京工业大学计算机科学与技术学院,江苏南京211816

出　　处：《电子技术应用》2016年第11期119-121,125,共4页Application of Electronic Technique

摘　　要：AKO-RVM算法不仅具有高分类精度的特点,相对于RVM算法其在一定程度上降低了性能对初始参数的依赖性,在入侵检测网络安全的方法研究中优于经典RVM算法。然而AKO-RVM样本训练与分类用时较长,为此提出一种基于概率的主辅式并行粒子群AKO-RVM方法,即将训练样本进行分组,先采用并行主辅式粒子群算法确定AKO-RVM核宽参数并进行优化,进而构造RVM分类模型,继而采用一对一分类方法应用于多类检测中。入侵实验结果表明,所提出方法在具有高精度与性能、低依赖性等特点的同时,较大程度上降低了训练所需迭代次数与检测时间。AKO-RVM algorithm not only has the characteristics of high classification accuracy, but also reduces the dependence of the performance of initial parameters relative to the RVM algorithm to a certain extent. In the study of intrusion detection method of network security, it is better than classical RVM algorithm. However training and classification of AKO-RVM sample are longer, therefore this paper puts forward a kind of based on the probability of advocate complementary type parallel particle swarm AKORVM method, the training samples are grouped, advocate complementary type parallel particle swarm optimization algorithm is adopted to define the first AKO-RVM nuclear parameter and optimize it, and then the RVM classification model is constructed. Then adopting the classification method is applied to the type of testing. Invasion of the experimental results show that the proposed method has high accuracy and performance characteristics as well as low dependence＇, at the same time, it largely reduces the number of iterations needed for training and testing time.

关 键 词：入侵检测 相关向量机 概率并行粒子群优化 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]