Identification of leakage area in large water distribution systems (Case study: Mahan city in Kerman province)

Today, leak detection is an important issue in water distribution networks, because leakage increases costs and causes a waste of water resources. In this study, a new method was used to identify the leak area in the water distribution network at part of Mahan city. First, the network was divided into a number of areas by the K-means clustering algorithm. Then, training samples related to each area were made using random leakage simulation in the network hydraulic model. The number of each area was used as the classification label of the multi-class support vector machine and along with the training samples, the classification model was taught. Finally, the trained model was used as a leak area identification model and was applied to determine possible leak areas with the observed field samples. The results show that out of 10 structures used to build the model, only three structures include; "Three areas and radial basis kernel function", "Three areas and linear kernel function" and "Five areas and linear kernel function", respectively, make an acceptable rate of 99.27%, 99.18% and 98.99% for the "Accuracy" evaluation index. On the other hand, among these three structures, dividing the network into five leak areas and using the linear kernel function divides the network into more areas and makes leak detection easier in restricted areas. As a result, this structure was used to build the final model.