区间分析法是处理非线性最优最劣法(BWM)存在多解问题的有效方法。本文基于权重区间中心点的大小关系,提出了一种无需计算区间优先度即可获得优先度矩阵的全新便捷算法;然后,在此基础之上,利用“尺度变换”的思想,给出了几类求解权重区间中心点的估计方法,并探讨了这些方法的计算偏差;最后,利用本文所提方法研究了一类电网关键物资的选取问题。本研究简化了已有文献中的区间分析法,规避了求解非线性优化模型的复杂计算过程,推广了非线性BWM的应用范围。Interval analysis is an effective method for dealing with the existence of multiple solutions in nonlinear Best-Worst Methods (BWM). This article proposes a convenient algorithm for obtaining the priority matrix without calculating the interval priority based on the size relationship of the center points of the weight interval. Then, based on this, several methods for estimating the center point of the weight interval were proposed using the idea of “scale transformation”, and the calculation bias of these methods was discussed. Finally, the method proposed in this article was used to study the selection of key materials for a class of power grid enterprises. This study greatly simplifies the interval analysis method in existing literature, and avoids the complex calculation process of solving nonlinear optimization models, and extends the application scope of nonlinear BWM.
网络流量特征分布的动态变化产生概念漂移问题,造成基于机器学习的网络流量分类模型精度下降.定期更新分类模型耗时且无法保证分类模型的泛化能力.基于此,提出一种基于散度的网络流概念漂移分类方法(ensemble classification based on d...
详细信息
网络流量特征分布的动态变化产生概念漂移问题,造成基于机器学习的网络流量分类模型精度下降.定期更新分类模型耗时且无法保证分类模型的泛化能力.基于此,提出一种基于散度的网络流概念漂移分类方法(ensemble classification based on divergence detection,ECDD),采用双层窗口机制,从信息熵的角度出发,根据流量特征分布的JS散度,记为JSD(Jensen-Shannon divergence)来度量滑动窗口内数据分布的差异,从而检测概念漂移.借鉴增量集成学习的思想,检测到漂移时对于新样本重新训练出新的分类器,之后通过分类器权值排序,保留性能较高的分类器,加权集成分类结果对样本进行分类.抓取常见的网络应用流量,根据应用特征分布的不同构建概念漂移数据集,将该方法与常见的概念漂移检测方法进行实验对比,实验结果表明:该方法可以有效地检测概念漂移和更新分类器,表现出较好的分类性能.
暂无评论