Using complex network analysis for water quality assessment in large water distribution systems

Using symbolic machine learning to assess and model substance transport and decay in water distribution networks

Drinking water infrastructures are systems of pipes which are generally networked. They play a crucial role in transporting and delivering clean water to people. The water quality analysis refers to the evaluation of the advective diffusion of any substance in drinking water infrastructures from source nodes. Such substances could be a contamination for the system or planned for the disinfection, e.g., chlorine. The water quality analysis is performed by integrating the differential equation in the pipes network domain using the kinetics of the substance decay and the Lagrangian scheme. The kinetics can be formulated using a specific reaction order depending on the substance characteristics. The basis for the integration is the pipes velocity field calculated by means of hydraulic analysis. The aim of the present work is to discover the intrinsic mechanism of the substance transport in drinking water infrastructures, i.e., their pipes network domain, using the symbolic machine learning, named Evolutionary Polynomial Regression, which provides "synthetic" models (symbolic formulas) from data. We demonstrated, using one real network and two test networks, that the concentration at each node of the network can be predicted using the travel time along the shortest path(s) between the source and each node. Additionally, the formula models provided by symbolic machine learning allowed discovering that a unique formula based on kinetic reaction model structure allows predicting the residual substance concentration at each node, given the source node concentration, surrogating with a good accuracy the integration of the differential equations.

Shall we always use hydraulic models? A graph neural network metamodel for water system calibration and uncertainty assessment

Representing reality in a numerical model is complex. Conventionally, hydraulic models of water distribution networks are a tool for replicating water supply system behaviour through simulation by means of approximation of physical equations. A calibration process is mandatory to achieve plausible simulation results. However, calibration is affected by a set of intrinsic uncertainty sources, mainly related to the lack of system knowledge. This paper proposes a breakthrough approach for calibrating hydraulic models through a graph machine learning approach. The main idea is to create a graph neural network metamodel to estimate the network behaviour based on a limited number of monitoring sensors. Once the flows and pressures of the entire network have been estimated, a calibration is carried out to obtain the set of hydraulic parameters that best approximates the metamodel. Through this process, it is possible to estimate the uncertainty that is transferred from the few available measurements to the final hydraulic model. The paper sparks a discussion to assess under what circumstances a graph-based metamodel might be a solution for water network analysis.

Dual graph characteristics of water distribution networks-how optimal are design solutions?

Urban water infrastructures are an essential part of urban areas. For their construction and maintenance, major investments are required to ensure an efficient and reliable function. Vital parts of the urban water infrastructures are water distribution networks (WDNs), which transport water from the production (sources) to the spatially distributed consumers (sinks). To minimize the costs and at the same time maximize the resilience of such a system, multi-objective optimization procedures (e.g., meta-heuristic searches) are performed. Assessing the hydraulic behavior of WDNs in such an optimization procedure is no trivial task and is computationally demanding. Further, deciding how close to optimal design solutions the current solutions are, is difficult to assess and often results in an unnecessary extent of experiment. To tackle these challenges, an answer to the questions is sought: when is an optimization stage achieved from which no further improvements can be expected, and how can that be assessed? It was found that graph characteristics based on complex network theory (number of dual graph elements) converge towards a certain threshold with increasing number of generations. Furthermore, a novel method based on network topology and the demand distribution in WDNs, specifically based on changes in 'demand edge betweenness centrality', for identifying that threshold is developed and successfully tested. With the proposed novel approach, it is feasible, prior to the optimization, to determine characteristics that optimal design solutions should fulfill, and thereafter, test them during the optimization process. Therewith, numerous simulation runs of meta-heuristic search engines can be avoided.

Maintaining the long-term accuracy of water distribution models with data assimilation methods: A comparative study

The upgrading of water supply services is calling for more accurate and adaptive numerical models to give insight into actual water distribution systems (WDSs), which underlines the importance of carefully calibrating model parameters. Due to unavoidable uncertainties in the calibration process such as measurement errors, errors in model parameters assumed to be known, and local-optimum of calibration algorithms, calibrated parameters could still contain non-negligible latent errors, and the calibrated model may not able to maintain its long-term accuracy when operating conditions change. To solve this problem, there is growing interest in adopting data assimilation (DA) methods to utilize more comprehensive information in long-term measurements to reduce the impact of uncertainties and maintain the accuracy and stability of calibrated models. In this study, two traditional calibration methods and four DA methods were tested and compared in two WDSs with different structures, which aims to form a general understanding of the behavior and applicability of different methods. The calibration results show that DA methods perform better than traditional methods and are more robust to different types of uncertainties, which provide an effective way to maintain the long-term accuracy of WDS models to enable better management of WDSs. Ensemble-based DA methods such as Particle Filter (PF) and Inferential-Measurement Kalman Filter (IMKF) performed well in the real-life system. They avoid linear approximation and can better estimate the impact of uncertainties to assimilate accurate correction information of the parameters. Gradient-based DA methods such as Extended Kalman Filter (EKF) and Variational Bayesian Adaptive Kalman Filter (VBAKF) have lower computational demand, but they are found to be less robust when dealing with large system uncertainties and nonlinearities.