采用改进最小闭包球向量机的电力信息网络入侵检测方法  

An Intrusion Detection Method for Electric Power Information Network Based on Improved Minimum Enclosing Ball Vector Machine

作　　者：王宇飞[1] 赵婷[1] 李韶瑜[2] 赵保华[1] 李玉杰[2] 

机构地区：[1]中国电力科学研究院信息通信研究所,北京市海淀区100192 [2]甘肃省电力公司,甘肃省兰州市730050

出　　处：《电网技术》2013年第9期2675-2680,共6页Power System Technology

基　　金：国家863高技术基金项目(2011AA05A116)~~

摘　　要：为降低电力信息网络入侵检测的检测误差和检测耗时,提出一种基于改进最小闭包球向量机(minimum enclosing ball vector machine,MEBVM)的入侵检测方法。该方法将入侵检测抽象成多分类问题,通过改进MEBVM对历史数据样本的训练学习来得到入侵检测模型。改进MEBVM利用最小闭包球降低检测耗时,并在训练过程中利用粒子群优化算法动态搜索MEBVM的最优训练参数以降低入侵检测模型误差。最后基于电力信息网络现场数据的实验证明,该方法与传统方法相比具有更高的检测精度和更少的检测耗时。To reduce the detection error and shorten the detection time during the detection of intrusion into electric power information network, based on the improved minimum enclosing ball vector machine （MEBVM） a method to detect the intrusion is proposed. This method abstracts the intrusion detection into multi-classification problem, b）~ means of improving the training and learning of the improved MEVBM by samples of historical data an intrusion detection model is obtained. The improved MEVBM decreases the detection time-consuming by minimum enclosing ball, and during the training the particle swarm optimization （PSO） algorithm is utilized to dynamically search the optimal training parameters of MEBVM to reduce the error of intrusion detection model. Finally, the results of experiments based on field data of electric power information network show that comparing with traditional intrusion detection methods, the results of intrusion detection by the proposed method possess higher detection accuracy and shorter detection time-consuming.

关 键 词：电力信息网络 入侵检测 最小闭包球向量机 粒子群优化算法 多分类问题 误差分析 检测耗时 

分 类 号：TM73[电气工程—电力系统及自动化] TP393[自动化与计算机技术—计算机应用技术]