摘要
基于支持向量数据描述算法,快速地检测出动态数据流中的异常数据,为监控对象提供早期预警、状态评估和决策支持。该算法在训练阶段只需要正常样本,无需故障样本信息,最终构建能够描述所有正常样本空间的超球体,决策阶段利用与超球中心的距离进行相似度评判。通过实验表明,该方法能有效发现动态数据流模式中的变化。
A method proposed in this paper based on support vector data description algorithm can fast detect anomalies in dynamic data streams,and provide early warning,condition assessment and decision support for monitoring object.In the training phase,the algorithm only requires positive samples without fault information to build hyper-sphere model to describe all of the normal sample space,And use difference of distance from center of hyper-sphere for evaluation of similarity in the decision-making stage.The results show that the algorithm can effectively detect implicit changes of patterns in dynamic data streams.
出处
《自动化技术与应用》
2015年第6期48-50,共3页
Techniques of Automation and Applications
