Pipe Failure Rate Prediction In Water Distribution Networks Using RCNN-SVR and FCMR Methods

It is essential to optimize repair, reconstruction and rehabilitation programs of urban water distribution networks for the correct use of limited water resources. The aim of the present study is to implement and compare two artificial intelligence methods to predict the burst rate of water pipes.

In this article, dataset of the pipe bursts in Joopar water network from 2012 to 2017 has been collected. The parameters for predicting of pipe failure includes: material, age, diameter, water pressure and installation depth, and the correlation coefficient of these parameters with the failure rate has been investigated. In order to predict the failure rate, convolution neural network with support vector machine (RCNN-SVR) and fuzzy regression based on c mean clustering (FCMR) have been used. To compare the performance of these two methods, the criteria such as Root-Mean Squared Error (RMSE), Index Of Agreement (IOA), Mean Absolute Percentage Error (MAPE), and coefficient of determination (R2) have been used.

According to the evaluations, RCNN-SVR method compared to FCMR method shows excellent results. Also, the correlation between age and failure rate in asbestos pipes is high and in polyethylene pipes this value is positive but low. The correlation coefficient between pressure and failure rate is also positive for both pipe types.