On the use of SINDy for WDN

[EN] With the growing interest of water utilities on digitalization, running multiple scenarios can become cumbersome with limited budget and short data collections. The total number of hydraulic simulations required (usually using commercial software), becomes a burden for near real-time operation. In order to circumvent the computational burden (limitation), since a couple of decades, several Machine Learning techniques have been used to create a meta-model or surrogates of a Water Distribution Networks (WDN) based on a subset of data available through SCADA. Among the many possible surrogates a Sparse Identification of Non-linear Dynamics (SINDy) method is presented. The method is applied to two datasets: i) to obtain a surrogate of a benchmark network and ii) real data of water consumption of different District Metered Areas (DMA) of a real water utility. The method is: i) computationally inexpensive, ii) less data demanding for calibration than other modern methods, iii) parsimonious, and iv) could be used to infer physical relations among data.