Performance evaluation of ANN and ANFIS models for estimating velocity and pressure in water distribution networks

Abstract

In this study, two artificial intelligence techniques: (1) artificial neural networks (ANNs) using different algorithms such as Lavenberg–Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) and (2) Adaptive Neuro-Fuzzy Inference System (ANFIS) are used to predict velocity and pressure for Gadhra (DMA-5) real water distribution network (WDN), East Singhbhum district of Jharkhand, India. In case 1, flow rate and diameter are used as independent variables to predict velocity. In case 2, elevation and demand are used as independent variables to predict pressure. 80% of the data are used to train, test, and validate the ANN and ANFIS prediction models, while 20% of the data are used to evaluate data-driven models. Sensitivity analysis is performed in ANN-LM to understand the relationship between the independent and dependent variables. The performance indices of RMSE, MAE, and R2 are evaluated for ANN and ANFIS for different combinations. The ANN-LM, with 2-16-1 architecture, is found as a superior to predict velocity and ANN-LM with architecture 2-17-1 is found as a superior to predict pressure. ANN-LM had the best prediction in estimating velocity (RMSE = 0.0189, MAE = 0.0122, R2 = 0.9568) and pressure (RMSE = 0.3244, MAE = 0.2176, R2 = 0.9773).