【作者】 肖健华；

【Author】 XIAO Jian-hua(Institute of Intelligent Technology & Systems,Wuyi University,Jiangmen,529020)

【机构】 五邑大学智能技术与系统研究所；

【摘要】 针对动态野点数据检测过程中可能存在的优化规模过大的问题,提出了一种基于核方法的实时野点检测方法:KROD。首先介绍了核方法下基于边界的野点检测方法,即支持向量数据描述SVDD。并指出,将SVDD应用于动态野点数据的实时检测中,其涉及的优化规模会随着时间的推移而不断增大,最终必然导致优化过程无法执行。为了克服这一不足,在核映射对应的特征空间中引入相对距离的概念,用于选择待优化的样本集,并使优化操作仅对该集合进行,从而大大减小了优化规模,野点数据的实时检测也就得以实现。最后,将KROD应用于滚动轴承的实时检测中,较大程度上缩短了检测时间,并取得了较为理想的检测效果。更多还原

【Abstract】 Aiming at the problem of large optimization size in dynamic outlier detection, this paper proposes a Kernel-based Real-time Outlier Detection (KROD) method. At first, an approach based on Support Vector Data Description (SVDD) to boundary-based outlier detection is introduced, and then this paper points out that when SVDD is used in real-time outlier detection with the increasing number of samples, the optimization size will exceed the memory space of the computer. Consequently, the algorithm will become hard to be performed. For the purpose of overcoming the above drawback, the function of relative distance in feature space is introduced to choose the sample set for optimizing. Because of the reduced optimization size by KROD, the real-time outlier detection can be implemented. At last, KROD is used in the process of quality detection of rolling bearing to reduce the detection time and guarantee the detection effect.更多还原