基于LSSVM交互预测的供水管网爆管检测  

Water Supply Network Burst Detection Based on Least Squares Support Vector Machine Interactive Prediction

作　　者：吴以文 杜坤[1] 吴汉清 罗雄武 宋志刚[1] 黄乐烽 WU Yi-wen;DU Kun;WU Han-qing;LUO Xiong-wu;SONG Zhi-gang;HUANG Le-feng(College of Civil Engineering and Architecture,Kunming University of Science and Technology,Kunming 650500,China;Urban Planning&Design Institute of Yunnan Province,Kunming 650228,China;Kunming Guanfang Architecture&Design Inc.,Kunming 650051,China)

机构地区：[1]昆明理工大学建筑工程学院,云南昆明650500 [2]云南省城乡规划设计研究院,云南昆明650228 [3]昆明官房建筑设计有限公司,云南昆明650051

出　　处：《中国给水排水》2022年第9期58-63,共6页China Water & Wastewater

基　　金：云南省重点领域科技计划项目(202003AC100001)。

摘　　要：随着我国越来越多的供水管网安装SCADA系统,基于数据预测的爆管检测方法越来越受到重视。传统基于数据预测的方法大多根据单点流量、压力的历史监测数据预测当前时刻监测值,当预测值与监测值的差值超过阈值时判定为爆管。然而,实践经验表明,监测数据的丢失与错误会严重影响单点预测结果,进而引起频繁误报与漏报。考虑到实际用水及监测数据的空间关联性(例如水压监测点布置距离越近其监测数据相关性越大),开展了基于最小二乘支持向量机(LSSVM)交互预测的供水管网爆管检测研究。对管网中不同位置监测数据构建多输入单输出LSSVM交互预测模型,选取1倍标准差为阈值进行爆管检测,并与传统的卡尔曼滤波爆管检测结果相对比。结果表明,LSSVM交互预测模型能降低数据丢失、数据错误对预测结果的影响,且对较小的爆管响应更加灵敏,进而有效地提高了基于数据预测的爆管检测性能。With an increasing number of SCADA system installed in the water supply network,the method of pipe burst detection based on data prediction has been paid more and more attention.Most of the traditional methods based on data prediction predict the current monitoring value according to the historical monitoring data of single point flow and pressure.When the difference between the predicted value and the monitoring value exceeds the threshold,the pipe burst was identified.However,practical experience shows that the loss and error of monitoring data will seriously affect the single point prediction results,and then cause frequent false positives and missing reports.Considering the spatial correlation between actual water consumption and monitoring data(for example,the closer the distance of water pressure monitoring points is,the greater the correlation of monitoring data is),water supply network pipe burst was detected based on least squares support vector machine(LSSVM)interactive prediction.A multi input and single output LSSVM interactive prediction model was constructed based on the monitoring data of different positions in the pipe network.A double standard deviation was selected as the threshold value for pipe burst detection,and the detection results were compared with those of traditional Kalman filtering.LSSVM interactive prediction model could reduce the influence of data loss and data error on the prediction results,and was more sensitive to small pipe burst,thus effectively improving the performance of pipe burst detection based on data prediction.

关 键 词：供水管网 爆管 最小二乘支持向量机(LSSVM) 交互预测 

分 类 号：TU991[建筑科学—市政工程]