[1]陈京钰,肖诗云,冯新.一种供水管网泄漏区域定位的机器学习方法[J].中国给水排水,2021,37(7):58-65.
CHEN Jing-yu,XIAO Shi-yun,FENG Xin.A Machine Learning Method for Leakage Localization of Water Distribution Network[J].China Water & Wastewater,2021,37(7):58-65.
点击复制
CHEN Jing-yu,XIAO Shi-yun,FENG Xin.A Machine Learning Method for Leakage Localization of Water Distribution Network[J].China Water & Wastewater,2021,37(7):58-65.
一种供水管网泄漏区域定位的机器学习方法
中国给水排水[ISSN:1000-4062/CN:12-1073/TU]
卷:
第37卷
期数:
2021年第7期
页码:
58-65
栏目:
出版日期:
2021-04-01
- Title:
- A Machine Learning Method for Leakage Localization of Water Distribution Network
- Keywords:
- leakage zone identification; binary iteration method; k-means; random forest classifier; feature selection; mean decrease accuracy (MDA)
- 摘要:
- 当采用模式识别方法识别供水管网的泄漏时，如果将每一个节点作为分类器的一个标签，由于供水管网节点泄漏特征的相似性，模型训练的准确率会比较低，因此可以通过聚类泄漏特征相似的节点形成区域，以每个区域作为分类器的标签从而提高模型训练的准确率。提出了一种基于随机森林分类器的二分迭代法识别泄漏区域，根据上一级分类器识别的泄漏区域的节点泄漏变化矩阵，采用k-means聚类将上一级迭代识别的泄漏区域聚类为两类（包含泄漏节点的区域和不包含泄漏节点的区域），从而识别包含泄漏节点的区域。随着候选泄漏区域的缩小，对识别泄漏区域有帮助的测点数量也逐渐减少，因此采用平均准确率减少（MDA）进行分类器特征（所需测点）的选择，在保证识别准确率不变的情况下减少分类器训练时所需的特征。与直接进行区域分块的识别方法相比，二分迭代法降低了选择区域分块数目时的盲目性，对于泄漏区域的识别更有目的性，提高了泄漏区域识别的准确率和效率。
- Abstract:
- When pattern recognition is used to identify the leakage of water distribution network, due to the similarity of leakage characteristics, the accuracy of model training will be relatively low if each node is used as the category label of the classifier. Therefore, nodes with similar leakage characteristics can be clustered to form regions, and each region can be used as a category label of the classifier to improve the accuracy of model training.A method called binary iteration based on random forest classifier was proposed to identify the leakage zone. According to the leakage matrix of the leakage zone identified by the last iteration, k-means clustering was used to cluster the nodes into two kinds(zones that contained leaks and zones that did not contain leaks), so as to identify the region containing the leaks. As the candidate leakage zone decreased, the number of sensors useful to identify the leakage also decreased, so mean decrease accuracy (MDA) was used to select the features required by the classifier, so as to reduce the features required by the classifier training under the condition that the identification accuracy was constant. Compared with the method that the leakage zone was identified directly, binary iteration method could reduce the blindness of selecting the division number, and it was more specific for the identification of leakage zone,which improved the accuracy and efficiency of the identification of the leakage zone.
更新日期/Last Update:
2021-04-01