首页|轨道几何状态检测异常数据实时智能识别

轨道几何状态检测异常数据实时智能识别

Real Time Intelligent Recognition of Abnormal Data in Track Geometry Inspection

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
受外界干扰、数据传输、传感器信号偏移等因素影响,轨道几何状态检测数据会产生异常峰值超限,影响现场检测人员工作效率.考虑到异常数据样本较少的不利因素,本文基于轨道几何检测系统传感器正常数据,通过消除数据趋势项,提取时序数据多维特征组成训练集,训练并构建了基于单分类支持向量机的异常数据智能识别模型.运用该模型对某地铁轨道几何检测系统单边位移的时序数据进行预处理、特征提取和智能分类,试验验证了其识别效果.结果表明:该方法识别效果好,误报率低,异常数据识别准确率高,且具有轻量化、易部署的特点,可满足轨道几何检测系统实时检测要求.
Due to external interference,data transmission,sensor signal offset,and other factors,the track geometry inspection data may generate abnormal peak values exceeding the limit,which affects the efficiency of on-site inspection personnel.Considering the disadvantage of having fewer abnormal data samples,this paper was based on the normal sensor data of the track geometry inspection system.By eliminating data trend terms and extracting multi-dimensional features from temporal data to form a training set,a single classification support vector machine based intelligent recognition model for abnormal data was trained and constructed.The model was applied to preprocess,extract features,and intelligently classify the temporal data of unilateral displacement in a certain subway track geometry inspection system,and its recognition effect was verified through experiments.The results show that this method has good recognition performance,low false alarm rate,high accuracy in identifying abnormal data,and has the characteristics of lightweight and easy deployment,which can meet the real time inspection requirements of track geometry inspection systems.

track geometry inspectionabnormal identificationfeature extractionintelligent recognition modelsingle classification support vector machineelimination of trend items

程朝阳、王昊、侯智雄、李颖、杨劲松、韩志、郝晋斐

展开 >

中国铁道科学研究院集团有限公司 基础设施检测研究所, 北京 100081

轨道几何状态检测 异常识别 特征提取 智能识别模型 单分类支持向量机 趋势项消除

中国铁道科学研究院集团有限公司项目

2021YJ217

2024

10.3969/j.issn.1003-1995.2024.02.05

铁道建筑
中国铁道科学研究院

铁道建筑

北大核心
影响因子：0.623
ISSN：1003-1995
年,卷(期)：2024.64(2)
  • 11