工业控制系统(industrial control system,ICS)入侵检测模型近年来愈加复杂,参数优化愈加困难,传统单分类器模型表现出明显的局限性。针对该问题,提出一种基于多分类器集成的ICS入侵检测算法,借鉴“分而治之”的思路将高维复杂入侵检测...工业控制系统(industrial control system,ICS)入侵检测模型近年来愈加复杂,参数优化愈加困难,传统单分类器模型表现出明显的局限性。针对该问题,提出一种基于多分类器集成的ICS入侵检测算法,借鉴“分而治之”的思路将高维复杂入侵检测问题分解为多个简单子问题,使用单分类器模型对每个子问题进行分析并获取最优分类,最后采用改进Bagging完成各个分类器结果的融合。同时针对样本不均衡问题,在预处理阶段提出改进的少数样本合成技术(improved synthetic minority over-sampling technique,ImSMOTE)构建平衡数据集。采用密西西比州立大学(Mississippi State University,MSU)的天然气管道测试平台SCADA系统记录的真实数据开展实验,结果表明所提方法能够获得较高的入侵检测准确率,同时少数类别的误检率明显降低,能够有效提升ICS系统的安全性和可靠性。展开更多
Multiple classifier systems based on the combination of a set of different classifiers are adopted to achievehigh pattern-recognition performances. A multiple classifiers integration method based on adaptive weight ad...Multiple classifier systems based on the combination of a set of different classifiers are adopted to achievehigh pattern-recognition performances. A multiple classifiers integration method based on adaptive weight adjusting ispresented in this paper. The useful neighbors are selected from training set by analyzing the pending pattern' s charac-ter, then each classifier's weight can be determined automatically by analyzing the performance of the classifier on theuseful neighborhood set. The final output of the multiple classifiers systems is the effective integration of each calssifi-er's result. The effectiveness of the method is proved by the text classification experiments of the Reuters-21578 textsets.展开更多
文摘工业控制系统(industrial control system,ICS)入侵检测模型近年来愈加复杂,参数优化愈加困难,传统单分类器模型表现出明显的局限性。针对该问题,提出一种基于多分类器集成的ICS入侵检测算法,借鉴“分而治之”的思路将高维复杂入侵检测问题分解为多个简单子问题,使用单分类器模型对每个子问题进行分析并获取最优分类,最后采用改进Bagging完成各个分类器结果的融合。同时针对样本不均衡问题,在预处理阶段提出改进的少数样本合成技术(improved synthetic minority over-sampling technique,ImSMOTE)构建平衡数据集。采用密西西比州立大学(Mississippi State University,MSU)的天然气管道测试平台SCADA系统记录的真实数据开展实验,结果表明所提方法能够获得较高的入侵检测准确率,同时少数类别的误检率明显降低,能够有效提升ICS系统的安全性和可靠性。
文摘Multiple classifier systems based on the combination of a set of different classifiers are adopted to achievehigh pattern-recognition performances. A multiple classifiers integration method based on adaptive weight adjusting ispresented in this paper. The useful neighbors are selected from training set by analyzing the pending pattern' s charac-ter, then each classifier's weight can be determined automatically by analyzing the performance of the classifier on theuseful neighborhood set. The final output of the multiple classifiers systems is the effective integration of each calssifi-er's result. The effectiveness of the method is proved by the text classification experiments of the Reuters-21578 textsets.
