摘要
传统电子直线加速器故障检测方法多依赖参数阈值比较和单一分类模型,难以有效融合多源异构数据,对复杂故障模式的识别能力有限,易导致漏检与误检。为此,文章提出一种基于Stacking集成学习的故障检测方法。首先,通过布设多类传感器采集设备运行过程中产生的多源异构数据,构建多元数据集;其次,设计Stacking分层架构,融合多种异质基学习器,提取元特征以识别设备运行状态的特定变化;在此基础上结合故障树模型,依据状态特征变化实现故障的准确检测和定位。
Traditional fault detection methods for linear accelerator rely on multi-parameter threshold comparison and single classification model,which are difficult to effectively integrate multi-source heterogeneous data.Their ability to identify complex fault modes is limited,and it is easy to lead to missed detection and false detection.Therefore,this paper proposes a fault detection method based on Stacking ensemble learning.Firstly,a multivariate data set was constructed by deploying multiple sensors to collect multi-source heterogeneous data in the operation of the equipment.Secondly,a Stacking hierarchical architecture was designed to fuse a variety of heterogeneous base learners and extract meta-features to identify specific changes in the running state of the device.On this basis,combined with the fault tree model,the fault is accurately detected and located according to the change of state characteristics.
作者
于汇洋
史思明
史磊
YU Huiyang;SHI Siming;SHI Lei(Chifeng Cancer Hospital,Chifeng,Inner Mongolia 024000,China)
