大规模降雨监测数据异常识别方法
Anomaly Recognition Method of Large-scale Rainfall Monitoring Data
投稿时间:2021-09-28  修订日期:2022-06-30
DOI:
中文关键词:  Hampel法  格拉布斯准则  周边测站分析法  雷达校验  异常识别
英文关键词:Hampel method  Grubbs criterion  surrounding station analysis method  radar calibration  anomaly recognition
基金项目:国家重点研发计划项目(2019YFC1510605);国家自然科学基金(51909274)
作者单位邮编
田济扬 中国水利水电科学研究院 100038
刘含影 中国水利水电科学研究院 
刘荣华 中国水利水电科学研究院 
丁留谦 中国水利水电科学研究院 
刘宇 中国水利水电科学研究院 
摘要点击次数: 1211
全文下载次数: 0
中文摘要:
      为实现大规模雨量监测站数据的异常识别和快速处理,基于Hampel法、格拉布斯准则、周边测站分析法和雷达辅助校验等方法,建立了递进式异常站点筛查体系,通过K-d tree(K-dimension tree)高级数据结构和并行计算方法提高计算效率,并以福建省5234个雨量监测站2015—2021年雨量数据进行了验证,结果表明福建省雨量监测站数据质量逐年提升;各类测站中,雨量站异常站点占全部异常站点的比例最高,各类异常站点在全省相应类型站点中,雨量站异常站点的占比也最高;雷达辅助校验能够有效解决在雨区与非雨区边界、雨强差异较大的雨区边界的正常站点更容易被错误判定为异常值的问题,校验前异常识别准确率为90%左右,校验后准确率提高为95%左右。通过K-d tree和并行计算,全省测站完成一次异常识别需约5~8min,为大规模降雨监测数据异常识别、充分利用雨量监测站有效信息提供可靠的方法。
英文摘要:
      In order to realize the abnormal identification and rapid processing of large-scale rainfall monitoring station data, and to provide accurate and reliable basis for rainstorm flood risk warning and early warning, this article has established a progressive screening system for abnormal sites, it is based on Hampel method, Grubbs criterion, surrounding station analysis method and radar auxiliary verification methods, improved computing efficiency through the K d tree (K-dimension tree) advanced data structure and parallel computing methods, and verified it with the rainfall data of 5234 rainfall monitoring stations in Fujian Province from 2015 to 2021 , The results show that the data quality of the rainfall monitoring stations in Fujian Province is improving year by year; among the various stations, the abnormal stations of rainfall stations account for the highest proportion of all abnormal stations, and the percentage of abnormal stations of rainfall stations among the corresponding types of stations in the province it is also the highest; radar-assisted calibration can effectively solve the problem that normal stations at the boundary of rain areas and non-rain areas and rain areas with large differences in rain intensity are more likely to be wrongly judged as abnormal values. The accuracy of abnormal recognition before calibration is about 90%, the accuracy rate after calibration is improved to about 95%. Through K-d tree and parallel computing, it takes about 5 to 8 minutes to complete an anomaly identification by the stations in the province, providing a reliable method for the identification of large-scale rainfall monitoring data anomalies and making full use of the effective information of rainfall monitoring stations.
  查看/发表评论  下载PDF阅读器
关闭