Management and Nonlinear Analysis of Disinfection System of Water Distribution Networks Using Data Driven Methods

Abstract

Chlorination unit is widely used to supply safe drinking water and removal of pathogens from water distribution networks. Data-driven approach is one appropriate method for analyzing performance of chlorine in water supply network. In this study, multi-layer perceptron neural network (MLP) with three training algorithms (gradient descent, conjugate gradient and BFGS) and support vector machine (SVM) with RBF kernel function were used to predict the concentration of residual chlorine in water supply networks of Ahmadabad Dafeh and Ahruiyeh villages in Kerman Province. Daily data including discharge (flow), chlorine consumption and residual chlorine were employed from the beginning of 1391 Hijri until the end of 1393 Hijri (for 3 years). To assess the performance of studied models, the criteria such as Nash-Sutcliffe efficiency (NS), root mean square error (RMSE), mean absolute percentage error (MAPE) and correlation coefficient (CORR) were used that in best modeling situation were 0.9484, 0.0255, 1.081, and 0.974 respectively which resulted from BFGS algorithm. The criteria indicated that MLP model with BFGS and conjugate gradient algorithms were better than all other models in 90 and 10 percent of cases respectively; while the MLP model based on gradient descent algorithm and the SVM model were better in none of the cases. According to the results of this study, proper management of chlorine concentration can be implemented by predicted values of residual chlorine in water supply network. Thus, decreased performance of perceptron network and support vector machine in water supply network of Ahruiyeh in comparison to Ahmadabad Dafeh can be inferred from improper management of chlorination.

Keywords

• Disinfection
• Data-driven Methods
• Chlorine Residual
• Perceptron Network
• support
• Vector Machine