Artificial neural network-based water distribution scheme in real-time in long-distance water supply systems

Abstract

Hydraulic models of long-distance water supply systems are usually used to regulate valves and pumps to realize the expected water distribution. Establishing and calibrating the hydraulic model is time-consuming and requires many engineering parameters, which are usually uncertain. This paper proposes a metamodel based on artificial neural networks (ANNs) to replace the computationally costly hydraulic model. The metamodel is designed to bypass the modeling and calibration processes of the hydraulic model and directly estimate the target state of valves and pumps to realize real-time water distribution. The proposed approach uses the water levels of reservoirs and the flow demands of water plants as input data to the ANN. The metamodel's output prescribes the opening of regulating valves and the speed of pumps. A realistic case study is presented to validate the accuracy and efficiency of the approach. The results show that ANN is feasible as a state predictor to realize real-time water distribution in practical water supply projects.