A comparison between performance of support vector regression and artificial neural network in prediction of pipe burst rate in water distribution networks

One of the main reasons for rehabilitation of Water Distribution Networks (WDNs) is pipe failure. To evaluate the mechanical reliability of a water distribution system, it is imperative that a relationship between pipe bursts and the effective parameters of the system is established. In recent years artificial intelligence techniques have been introduced as an effective method for prediction of pipe bursts. This paper compares the performance of Artificial Neural Network (ANN) and Support Vector Regression (SVR) in predicting the Pipe Burst Rate (PBR) in water distribution networks. In addition, the impact of hydraulic pressure on accuracy of the data-driven pipe burst prediction model is studied, where average and maximum hydraulic pressure values are considered as input variables of the model. The data used for the analyses are from two real case studies in Iran. From the obtained results in both case studies it can be concluded that ANN is better (universal) predictor than SVR, but cannot be easily used for generalization purposes since it is not consistent with the physical behavior observed. Thus, for water distribution system management purposes SVR can be preferred over ANN. Also it can be said that use of average hydraulic pressure as an input variable results in improving the performance of the model and the accuracy of predictions.