摘要
针对网络安全态势精确预测,提出一种基于改进广义回归神经网络的预测方法,以改善网络安全态势预测精度。利用滑动时间窗口方法将各个离散时间监测点的网络安全态势值构造成部分线性相关的多元回归数据序列,以其做为样本集输入到改进广义回归神经网络加以训练,进而得到网络安全态势预测模型。在改进广义回归神经网络训练过程中,利用粒子群算法动态地搜索广义回归神经网络最优训练参数,从而克服了广义回归神经网络训练参数选择困难的缺陷。实验结果表明:与传统方法相比基于改进广义回归神经网络的网络安全态势预测方法拥有更好的性能。
Focused on network security situation forecast（NSSF）,a novel NSSF method based on improved general regression neural network（GRNN-PSO）was proposed,in order to improve forecast accuracy.With sliding time window（STW）,all of the network security situation value（NSSV）of every discrete-time monitoring sites had been constructed into multi regression data sequence,which sequence was part of the linear correlation.In order to obtain NSSF model,the multi regression data sequence was trained as sample set by GRNN-PSO.In GRNN-PSO training process,it can overcome the deficiency of selecting difficultly GRNN＇s training parameters.Particle swarm optimization（PSO）was used to search the best training parameters.Finally,the experiments show that the NSSF method based on GRNN-PSO has better performance compared with the traditions.
出处
《华北电力大学学报（自然科学版）》
CAS
北大核心
2011年第3期91-95,共5页
Journal of North China Electric Power University：Natural Science Edition
关键词
网络安全态势预测
广义回归神经网络
粒子群算法
滑动时间窗口
多元回归分析
network security situation forecast
general regression neural network
particle swarm optimization
sliding time window
multi regression analysis
作者简介
王宇飞（1982-），男，硕士研究生，研究方向为网络信息安全、人工智能。
