Leakage Zone Identification in Large-Scale Water Distribution Systems Using Multiclass Support Vector Machines

This paper presents a new method for identifying leakage zones of water distribution systems. A large water network is first divided into a number of zones. The zone number is used as the category label of the multiclass support vector machine (M-SVM), which is trained with the data set generated by simulation of the possible leakages using a hydraulic model. The trained M-SVM is used as the leakage zone identification model and applied to determine the likely leakage zones with the observed field data. Two case studies are presented in this paper to demonstrate the effectiveness of the method. The results indicate that this method has many unique advantages in solving the nonlinear and high-dimensional pattern recognition problem with a small sample data set. Together with the method of pressure-dependent leakage detection (PDLD), the proposed approach enables engineers to improve the effectiveness and efficiency of leakage detection for large water distribution systems.