Optimal architecture for artificial neural networks as pressure estimator

Abstract

ABSTRACT The knowledge of hydraulic parameters in water distribution networks can indicate problems in real time, such as pipe bursts, small leakages, increase in pipe roughness and illegal connections. However, an accurate indication relies on the quantity and quality of the data acquired, i.e., the number of sensors used to monitor the network and their location. It is not economic feasible have a great number of sensors, thus, the use of artificial intelligence, such as Artificial Neural Networks (ANNs) can reduce the lack of information necessary to identify problems, estimating hydraulic parameter through the few information collected. The reliability of ANNs depends on its architecture, so, in this paper, different conditions are tested for ANN training to identify which are the most relevant parameters to be adjusted when the ANN is used for pressure estimation.