Prediction of non-revenue water ratio in water distribution systems

Abstract

In the evaluations of water distribution systems (WDSs) in terms of water loss and performance, the Non-Revenue Water ratio (NRW) stands out as one of the most important parameters. Within the scope of this study, in order to predict the NRW ratio, a large number of models at different variable combinations were generated using the Adaptive Neuro-Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN) methods. The performance of the models formed has been evaluated by taking R-2 , RMSE, MAE, SI, and Bias criteria as references. According to the study results, the model performances increase with the number of inputs in general, and the ANN models are more successful than ANFIS. Considering the modeling, the best-performing combination through the ANN method is WSQ-NJ-NL-NF, this one is the WSQ-NJ-NL-MPD combination in the ANFIS method which has three variables common. As a result, using variables common is significant for NRW predictions. On the other hand, NRW prediction performances need to improve by taking different variable combinations and methodological approaches into account, according to the ANFIS model results.