摘要
为解决传统业务服务质量(Quality of Services,Qo S)评价算法客观性不足、评价效率低等问题,将业务Qo S评价映射为业务Qo S等级分类,提出了一种基于双超平面(Double Decision Hyperplane,DDH)决策图的电力通信网业务实时Qo S评价方法。该方法提取业务网络层特征来表征业务的服务质量,以Qo S等级已知的现有业务作为分类器的训练样本,由分类器的构造顺序来确定决策图的形成。此外,算法对传统类间耦合度进行了重新定义,并在决策图各层按照最小类间耦合度原则依次构造分类双超平面,实现业务Qo S等级的不完全三分类,避免了传统多分类决策图低层结点"误差崩盘",自适应性低及决策图结构固定等问题。仿真结果证明,DDH决策图在电力业务Qo S评价中比经典DAG-SVM方法具有更短判决时间和更优分类性能,可更好地实现电力业务Qo S实时准确评价。
In order to solve the problems of the insufficient objectivity and low evaluation efficiency in traditional services quality （QoS） evaluation algorithms, a novel double decision hyperplanes （DDH） decision graph method, which mapped the QoS evaluation to the QoS level classification, was proposed to evaluate the real-time quality of services （QoS） in electric power communication networks. It extracted representative features to reveal services running conditions in the network layer and exploited the services whose QoS were known to train the classifiers. The decision graph structure was determined via the class grouping algorithm, which formed the groups of classes to be separated at each internal node. Moreover, a novel coupling degree between classes was presented and the double planes in each decision layer were created based on principle of the minimum coupling degree between classes. Since the node discriminations are implemented via incomplete ternary SVMs, the new method can well avoid problems of ＂ error collapse＂ in low decision layer nodes, low adaptability and the fixed structure in conventional decision graph algorithms. Finally, simulation results show that, compared with other classic methods, the novel method exhibits a number of attractive merits such as enhanced classification accuracy, short decision time, as well as achieves a more real-time QoS evaluation for electric power communication network services.
出处
《华北电力大学学报（自然科学版）》
CAS
北大核心
2016年第1期92-98,共7页
Journal of North China Electric Power University：Natural Science Edition
基金
国家高技术研究发展计划资助项目(2014AA01A701)
北京市自然基金资助项目(4142049)
关键词
电力通信网
业务服务质量
双超平面
类间耦合度
DAG-SVM
electric power communication networks
quality of services
double decision hyperplane
coupling degree between classes
DAG-SVM
