PDF下载

[1]彭森,程蕊,吴卿,等.基于极限学习机算法的供水管网爆管识别研究[J].中国给水排水,2022,38(7):56-62.

　PENGSen,CHENGRui,WUQing,et al.Identification of Water Supply Network Pipe Burst Based on Extreme Learning Machine Algorithm[J].China Water & Wastewater,2022,38(7):56-62.点击复制

基于极限学习机算法的供水管网爆管识别研究

中国给水排水[ISSN:1000-4062/CN:12-1073/TU] 卷: 第38卷 期数: 2022年第7期 页码: 56-62 栏目: 出版日期: 2022-04-01

Title:Identification of Water Supply Network Pipe Burst Based on Extreme Learning Machine Algorithm

作者:彭森1， 程蕊1， 吴卿1， 程景1， 孟涛2

Author(s):PENG　Sen1， CHENG　Rui1， WU　Qing1， CHENG　Jing1， MENG　Tao2（1. School of Environmental Science and Engineering， Tianjin University， Tianjin 300350， China； 2. North China Municipal Engineering Design & Research Institute Co. Ltd.， Tianjin 300381， China）

关键词:（1.天津大学 环境科学与工程学院; 天津 300350; 2.中国市政工程华北设计研究总院有限公司; 天津 300381）

Keywords:water supply network;  identification of pipe burst area;  ELM algorithm;  K-means clustering algorithm

摘要:供水管网爆管具有定位难、影响范围广的特点，长期困扰着供水企业。针对供水管网爆管区域识别问题，综合考虑多种影响因素下的爆管工况，利用爆管特征值矩阵构建爆管样本数据集，采用极限学习机算法（ELM）建立爆管区域识别模型；应用K-means聚类算法分析节点水力变化特征的相似性，并在此基础上对管网进行监测区域划分与监测点布设，形成多种监测方案；综合爆管识别率等参数，分析ELM在不同监测方案以及在噪声影响下的识别性能。采用实际管网算例进行了爆管区域识别分析，结果表明：该模型可以进行有效的爆管区域识别，同时结合不同分区方案可以提高爆管识别率；监测点的增加可以减小压力监测数据的噪声影响。

Abstract:Water supply network pipe burst has the characteristics of difficult location and wide range of influence, which has troubled water supply enterprises for a long time. To solve the problem of identifying the pipe burst area of water supply networks, the burst conditions were comprehensively considered under various influencing factors, the pipe burst sample data set was constructed by using eigenvalue matrix, and the model for identifying the pipe burst area was established by extreme learning machine (ELM) algorithm. The similarity of node hydraulic change characteristics was analyzed by using K-means clustering algorithm. On this basis, the monitoring area of the pipe network was divided and the monitoring points were arranged to form a variety of monitoring schemes. The identification performance of ELM under different monitoring schemes and noise impact was analyzed by combining the identification rate of pipe burst and other parameters. The burst area was identified and analyzed in a practical pipe network. It was found that the model could effectively identify the pipe burst area. At the same time，it could effectively improve the identification rate of pipe burst by combining different zoning schemes. The addition of monitoring points could reduce the noise impact from pressure monitoring data.

更新日期/Last Update: 2022-04-01