在生产环境下油管泄漏异常极少发生,并且人为调整泵频率、仪表校准等带来的监测数据曲线突变,即假异常,混淆在真异常中难以被区分,导致传统基于机器学习的泄漏异常识别方法召回率较低,误报率较高。针对该问题,提出一种基于真假异常区分...在生产环境下油管泄漏异常极少发生,并且人为调整泵频率、仪表校准等带来的监测数据曲线突变,即假异常,混淆在真异常中难以被区分,导致传统基于机器学习的泄漏异常识别方法召回率较低,误报率较高。针对该问题,提出一种基于真假异常区分的输油管道泄漏异常识别方法,利用一类支持向量机(One-Class Support Vector Machine,OCSVM)学习输油管道的正常工作模式,并利用该模式筛出管道的疑似异常,即真假异常。通过叠加多源数据的方式增加真假异常曲线形态差异,并利用相似性聚类发现泄漏事件的异常模式。将该方法应用到中国西北某老工业采油井场输油生产环境中进行验证,结果表明:泄漏异常识别召回率为100%,假异常排除率达到83.49%。该方法实现了在复杂生产环境中对输油管道泄漏异常事件的实时、高效监测,并为机器学习方法应用于生产环境提供了实践思路。(图8,表4,参28)展开更多
On July 22,2013,the Minxian-Zhangxian M_S6.6 earthquake occurred on the east segment of Lintan-Dangchang fault. The analysis of digital elevation and remote sensing imaging shows that the east segment of Lintan-Dangch...On July 22,2013,the Minxian-Zhangxian M_S6.6 earthquake occurred on the east segment of Lintan-Dangchang fault. The analysis of digital elevation and remote sensing imaging shows that the east segment of Lintan-Dangchang fault is still active and the main thrust feature of the fault switches to left lateral slip. With the field research of intensity and damage,several abnormal areas of degree Ⅷ spread in the isoseismal line of degree Ⅶ and some abnormal areas of degree Ⅶ spread in the isoseismal line of degree Ⅵ. These abnormal areas are distributed along the hanging wall of the fault in a width of 2km. The analysis based on the remote sensing and digital elevation model shows that the segment of the Lintan-Dangchang fault south of Minxian mainly slips in left literal. The fault movement made the soil soft in the fault zone. The earthquake motion propagated along the fault zone. Therefore the strong earthquake motion caused foundation failure in the soft soil along the fault zone and the abnormal intense areas of disaster formed.展开更多
文摘在生产环境下油管泄漏异常极少发生,并且人为调整泵频率、仪表校准等带来的监测数据曲线突变,即假异常,混淆在真异常中难以被区分,导致传统基于机器学习的泄漏异常识别方法召回率较低,误报率较高。针对该问题,提出一种基于真假异常区分的输油管道泄漏异常识别方法,利用一类支持向量机(One-Class Support Vector Machine,OCSVM)学习输油管道的正常工作模式,并利用该模式筛出管道的疑似异常,即真假异常。通过叠加多源数据的方式增加真假异常曲线形态差异,并利用相似性聚类发现泄漏事件的异常模式。将该方法应用到中国西北某老工业采油井场输油生产环境中进行验证,结果表明:泄漏异常识别召回率为100%,假异常排除率达到83.49%。该方法实现了在复杂生产环境中对输油管道泄漏异常事件的实时、高效监测,并为机器学习方法应用于生产环境提供了实践思路。(图8,表4,参28)
基金supported by basic scientific research operating expenses of Institute of Earthquake Science,China Earthquake Administration(2012IES010202)
文摘On July 22,2013,the Minxian-Zhangxian M_S6.6 earthquake occurred on the east segment of Lintan-Dangchang fault. The analysis of digital elevation and remote sensing imaging shows that the east segment of Lintan-Dangchang fault is still active and the main thrust feature of the fault switches to left lateral slip. With the field research of intensity and damage,several abnormal areas of degree Ⅷ spread in the isoseismal line of degree Ⅶ and some abnormal areas of degree Ⅶ spread in the isoseismal line of degree Ⅵ. These abnormal areas are distributed along the hanging wall of the fault in a width of 2km. The analysis based on the remote sensing and digital elevation model shows that the segment of the Lintan-Dangchang fault south of Minxian mainly slips in left literal. The fault movement made the soil soft in the fault zone. The earthquake motion propagated along the fault zone. Therefore the strong earthquake motion caused foundation failure in the soft soil along the fault zone and the abnormal intense areas of disaster formed.
