Prediction of leakage risk in water distribution network using artificial neural networks

• Main Authors: Liang, Po-Jui, 梁博瑞
• Other Authors: Huang, Chih-Pin
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/64pvc8

碩士 === 國立交通大學 === 環境工程系所 === 108 === Leakage in pipeline is not only a waste of resource, but a part of non-revenue water to water company. To overcome this problem, district meter area (DMA), pressure management and other leak founding methods have been widely practiced over decades. However, a lot of human resources and time often spent on locating the leaks. Therefore, it’s preferable to quickly narrow down the range of the leak location to reduce the cost of searching the leak point. The purpose of this study is to predict the leak risk of pipeline, so that it can minify the checked-out range of the leak. In this study, Zhunan and Zhubei were selected as study areas and their historical pipeline and leak point data were picked up from Taiwan Water Company’s geographic information system (TWC-GIS). Two different types of model were created to fit these data, one is PVCP model which only contains the PVCP type of material in data. Another one is a general model which includes all types of material. These models were created by Kears and were evaluated for its accuracy.   The result of fitting and evaluate the model shows a good performance, especially Zhubei’s PVCP model, its mean square error (MSE) between predicted and observed data was 0.246 and R2 reached 0.9157, which exhibited high correlations. Besides, the trend of predicted and observed during evaluated over the year matched approximately. It shows that neural network can predict the risk of leakage effectively, also presents the high-risk area in a more visual way.