Toward Sustainable Water: Prediction of non-revenue water via Artificial Neural Network and Multiple Linear Regression modelling approach in Egypt

Background: This research paper studied the parameters affecting the non-revenue water in Egypt. The neural model was developed to forecast the NRW ratio before and after network rehabilitation using main parameters and statistical data for various measured district metered areas (DMAs) in Egypt. The non-revenue water (NRW) ratio is a parameter of significant concern for the water distribution system's performance evaluation. Therefore, NRW ratio parameter evaluation requires analysis of variables that influence the NRW ratio. Aim/objective: This study aims to offer findings that will lay the basis for sustainable water resources management in Egypt. Consequently, a comprehensive evaluation of water distribution systems and parameters affecting NRW was conducted. Methodology: The predictive models were developed using the historical data of the water company for various measured district metered areas (DMAs) in Egypt. The automated meta-models were developed at steady-state and over a prolonged period using a series of principles, rules, and constraints. It was based on multi hidden units of feed-forward neural networks trained by the back-propagation algorithm. Thus, the iterative methods of weight adjustment and activation function of Levenberg-Marquardt was used. Two scenarios were assumed, one that has complete data of the DMA investigated while the other assumed total lack of knowledge. Therefore, four models were built to forecast multiple parameters affecting the percentage of non-revenue water. However, three models were combined to reach the final NRW ratio with no historical data. An optimal number of neurons was determined to be 10. Each model result was tested against numeral statistical indicators showing a different level of accuracy. Results: The model results showed high accuracy in NRW final estimation. The performance indicators (i.e., RMSE, MAE, and correlation) also showed that the machine-learning algorithm better identifies complex relationships between different parameters. The models developed in this research can be applied to other DMAs in Egypt. Overall, these findings indicate that the machine-learning model may be adequate for water companies seeking immediate, cost-effective and long-term improvement of their water distribution systems.