ABSTRACT
Water distribution networks with dynamically adaptive connectivity offer greater operational flexibility. While this strategy has demonstrated improvements in pressure management and network resiliency, further research is needed to better understand its impact on water quality dynamics. This paper investigates the short-term variability of disinfectant residuals in a real-world distribution network operated with dynamic connectivity. We simulate water quality dynamics under two control configurations with pressure control and automatic flushing valve operations. Our simulation results inform the development of flow variability metrics to reveal the relationship between changing hydraulic conditions and increased water quality dynamics. These metrics can (i) improve observability by supporting the placement of additional water quality monitoring locations and (ii) enhance controllability by enabling the formulation of optimization problems that incorporate hydraulic surrogates for modelling water quality. Furthermore, we validate the identified regions of increased water quality dynamics using continuous disinfectant data from a large-scale experimental programme. Our findings emphasize the need for further analytical and experimental research to manage water quality in distribution networks with dynamically adaptive connectivity and hydraulic control.
HIGHLIGHTS
Investigated the impact of dynamically adaptive connectivity on water quality in a real-world distribution network.
Proposed flow variability metrics to identify areas of the network with increased disinfectant variability.
Validated model simulations using time series analysis of continuous water quality data from a real-world experiment.
Highlighted the need for further research to improve the observability and controllability of water quality in dynamically adaptive networks.
INTRODUCTION
District metered areas (DMAs) have proven highly effective for managing leakage and pressure in water distribution networks. However, traditional single-feed DMAs can present significant operational challenges such as reduced network resilience and water quality deterioration. In response to these challenges, Wright et al. (2014, 2015) developed a novel concept of adaptive DMAs, which can dynamically adapt network connectivity (topology) and hydraulic control to accommodate a range of operational conditions and objectives. Although this operational strategy has been shown to improve pressure management and network resiliency (Wright et al. 2015; Giudicianni et al. 2020; Bui et al. 2022; Ulusoy et al. 2023), its impact on water quality dynamics has not been studied. Further research is therefore needed to investigate the impacts of advanced hydraulic control on the temporal variability of disinfectant residuals.
Traditional single-feed DMAs often impose artificial dead ends at their boundaries, resulting in hydraulic conditions characterized by low-flow velocities and long residence times. Consequently, dead-end pipes are generally associated with poor water quality conditions (US Environmental Protection Agency 2002; Machell & Boxall 2014; Blokker et al. 2016; Dias et al. 2017; Armand et al. 2018). Excessive disinfectant decay at dead ends can be attributed to long residence times and the influence of dispersive transport mechanisms (Abokifa et al. 2016; Abokifa et al. 2019). Moreover, the presence of low-flow conditions is associated with greater sediment accumulation, biofilm growth, and iron uptake from corrosion scaling (in metallic pipes), increasing the risk of discolouration incidents (Armand et al. 2017, 2018). While adaptive DMAs are designed to reduce dead ends, the increased changes in hydraulic conditions, specifically the variability in flow paths, could potentially introduce new water quality challenges. The main water quality challenge in adaptive DMAs is the management of source water mixing from multiple inlets, particularly when inlets have markedly different source water characteristics (e.g. disinfectant concentrations and total organic carbon). Although source water mixing has been modelled at the network level to assess the effects of new/different water supplies (Montiel et al. 2002; Ostfeld & Salomons 2004; Mala-Jetmarova et al. 2015; Szuster–Janiaczyk & Bylka 2019), its magnitude and frequency caused by the operation of adaptive DMAs at the distribution level remains unexplored. Additionally, flow path variability can create flow reversals and/or a wider range of pipe flow velocities, potentially leading to higher water age in looped sections of the network. We note that, while not explicitly considered in this work, changes in flow velocities are also linked to discolouration risk in distribution networks (Horn et al. 2003; Ahn et al. 2011; Derlon et al. 2008; Furnass et al. 2013; Boxall et al. 2023), which can in turn affect disinfectant decay behaviour.
In this paper, we investigate the relationship between changing hydraulic conditions and short-term temporal disinfectant variability within dynamically adaptive DMAs. We focus on hydraulic conditions resulting from two control configurations: (i) minimizing average zone pressure (AZP) (Wright et al. 2015) and (ii) integrating AZP with a self-cleaning capacity (SCC) control mode (Jenks et al. 2023a, 2023b). The AZP objective aims to reduce background leakage through pressure reduction, a common operational practice in UK water networks. On the other hand, the SCC objective is a relatively new strategy which aims to proactively decrease the risk of discolouuration in distribution networks. These control configurations are evaluated using a multi-feed operational network located in Bristol, UK, where ongoing experimental and analytical research is carried out by Imperial College London, Bristol Water, and Badger Meter/Analytical Technology (ATi). For each configuration, we identify areas of increased hydraulic dynamics using a set of proposed flow variability metrics computed over a 24-h simulation. These metrics aim to assess the degree of source water mixing between DMA inlets and identify other hydrodynamic conditions known to influence water quality deterioration. The flow variability metrics can also serve as hydraulic surrogates for modelling water quality dynamics in optimization problems formulated for control. Additionally, we simulate water quality dynamics in the operational network with varying inlet disinfectant concentrations to study the potential effects of source water mixing. We then validate our simulation results for the AZP control configuration, which is currently implemented in the field, by performing a time series analysis of data collected from an ongoing water quality sensing programme. The time series analysis focuses on highlighting short-term temporal (i.e. hourly) fluctuations in disinfectant data across the network.
This study represents a significant advancement in the field as it comprises an experimental programme with continuous water quality sensing at high spatial and temporal resolution. Previous water quality field studies have mainly consisted of pilot-scale programmes relying on manual collection techniques (Vasconcelos et al. 1997; Liu et al. 2015; Maleki et al. 2023) or short-term deployment of continuous reagent-based chlorine sensors (US Environmental Protection Agency 2008; Lee et al. 2010). In this work, we analyse continuous disinfectant data collected in a real-world operational network to validate our simulation results, namely the modelling of source mixing (i.e. flow paths). Our simulations and exploratory analysis of sensor data inform the design of subsequent experiments to enhance the observability and controllability of water quality dynamics in adaptive DMAs.
The remainder of this paper is structured as follows. Section 2 introduces the operational case study network, which we use to evaluate different control configurations as well as collect continuous water quality data. Section 3 outlines the paper's methodology, including model development, the proposed flow variability metrics, and time series data processing procedures. In Section 4, we present and discuss our simulation and time series analysis results. Section 5 concludes with suggestions for increasing the observability and controllability of water quality in adaptive DMAs.
CASE STUDY NETWORK
Overview
The Bristol Water Field Lab (Field Lab) is an operational distribution network supplying approximately 8,000 customer connections in Bristol, UK. The network was originally comprised of two single-feed DMAs hydraulically isolated via permanently closed isolation valves (IVs). Wright et al. (2014) introduced the Field Lab as an experimental programme for dynamically aggregating the isolated DMAs through the installation of dynamic boundary valves (DBVs), which are equipped with bidirectional control capabilities. Flow modulation profiles were also derived for advanced pressure control at DMA inlets. The Field Lab has since been a case network for numerous water distribution analysis and optimization studies and a testbed for advances in pressure and water quality sensing technologies. The subsequent sections describe the different control configurations evaluated in the Field Lab (Section 2.2) and the current water quality sensing programme conducted by our research group (Section 2.3).
Control configurations
The original single-feed DMA topology is illustrated in Figure 1(a). In this setup, several IVs sectorize different areas of the network, three of which separate DMAs 1 and 2, and pressure-reducing valves (PRVs) reduce DMA inlet pressures. The adaptive AZP control configuration in Figure 1(b) replaces two of the IVs with remote-control DBVs. These DBVs permit bidirectional control between DMAs, enabling the network to dynamically adapt between an isolated topology at night and an open topology during the day (Wright et al. 2014, 2015). Additionally, existing PRVs are retrofitted with flow modulated control to minimize the network's AZP (and thereby leakage) (Wright et al. 2015). This adaptive AZP configuration is currently implemented in the Field Lab.
We also consider a new water quality control strategy aimed at increasing the number of pipes in the Field Lab experiencing self-cleaning velocities (Jenks et al. 2023a, 2023b). The SCC of the network, defined as its ability to periodically resuspend and flush accumulated material, is improved by jointly optimizing the location and settings of PRV and automatic flushing valve (AFV) actuators, the latter of which discard water from the system in a controlled manner (Figure 1(c)) (Jenks et al. 2023a). In this configuration, PRV and AFV controls switch between the existing AZP mode and a 1-h self-cleaning mode where flushing is activated to mobilize material which has potentially accumulated in pipes. Such self-cleaning controls have been shown to significantly alter flow paths in the Field Lab (Jenks et al. 2023b), particularly due to the removal of existing IVs and DBVs. This operational mode creates a more looped network configuration, potentially causing large shifts in the hydraulic balance point. In this work, the self-cleaning configuration represents advanced control enabled by adaptive DMAs to reduce the risk of discolouration. We simulate the resulting water quality dynamics using a mathematical model of the Field Lab. Planned experimental work, which leverages the analytical methods and findings of this study, aims to deploy the adaptive AZP-SCC control configuration in the Field Lab.
Water quality sensing
The time series data used in this study were downsampled to 15-min intervals to extend battery life and ensure the longevity of data collection from the devices, which collected data from April to July 2023.
METHODOLOGY
Simulation
Model development
The hydraulic model of the Field Lab comprises junction nodes, variable head nodes (e.g. DMA inlets), links, and discrete 15-min time steps. Temporal patterns for variable head and demand nodes were acquired from pressure and flow sensors deployed as part of the Field Lab's ongoing experimental programme. For each time step , we solve the steady-state flow () and hydraulic head () conditions using state-of-the-art modelling software EPANET (Rossman et al. 2020), accessed via the WNTR Python package (Klise et al. 2017). As discussed in Section 2.2, we introduce control through the placement and operation of PRV, DBV, and AFV actuators. For each time step , we set (i) pressure setpoints at the downstream node of PRV links, (ii) dynamic local loss coefficients at DBV links, and (iii) flushing demands at AFV nodes. Such controls are derived from previous work which focused on optimizing hydraulic control within adaptive WDNs for AZP (Wright et al. 2015; Jenks et al. 2023b) and SCC (Jenks et al. 2023a, 2023b) objectives. Details on PRV, DBV, and AFV actuators as well as other model inputs are available in the project's GitHub repository (Jenks et al. 2024).
We use EPANET to simulate the transport and fate of water quality constituents (e.g. disinfectant and source trace). For modelling disinfectant residuals, we assume a first-order reaction relationship with bulk decay coefficient and wall decay coefficients ranging from to . These decay coefficients were calibrated to water quality conditions in the Field Lab. We set initial conditions (at ) across all links and simulated water quality dynamics over a 7-day period to ensure periodic behaviour is reached across all nodes. At each DMA inlet , we assign source concentration . Here, corresponds to the mean or worst-case (10th and 90th percentiles) of disinfectant variability conditions observed at inlet s over the period April to July 2023 (Table 1). We assume a normal distribution with standard deviation mg L−1 for modelling inlet variability from upstream mixing/storage dynamics. Since this work focuses on short-term disinfectant variability resulting from source water mixing between DMAs, we do not explicitly consider any seasonal (climatic) variability in the assigned inlet concentration patterns. Additionally, no clear seasonal trend could be discerned in the time series data collected at DMA inlets mainly due to the relatively short sensing period. We note, however, that seasonal variability is certainly an important consideration in managing disinfectant residuals, and that future work needs to consider data collected over longer time periods.
Inlet . | Mean scenario (mg L−1) . | Worst-case scenario (mg L−1) . |
---|---|---|
DMA 1 | 0.75 | 0.95 |
DMA 2 | 0.65 | 0.40 |
Inlet . | Mean scenario (mg L−1) . | Worst-case scenario (mg L−1) . |
---|---|---|
DMA 1 | 0.75 | 0.95 |
DMA 2 | 0.65 | 0.40 |
For source trace simulations, the contribution of source water is modelled as a conserved constituent (i.e. no reaction kinetics).
Flow variability
We propose three flow variability metrics to identify areas in the network with increased changes in hydraulic conditions: (i) count of pipe flow reversals; (ii) coefficient of variation (CV) of diurnal pipe flow velocities; and (iii) percentage of source water mixing. These metrics are computed using the steady-state hydraulic model of the Field Lab over a 24-h simulation.
Disinfectant variability
Sensor data
RESULTS AND DISCUSSION
Simulation
Flow variability
The single-feed DMA configuration yielded minimal hydraulic dynamics due to the isolation of DMAs 1 and 2, characterized by the closed IVs, a single inlet connection, and predominantly branched topology. Apart from a few looped sections of the network (e.g. near Sensor 7), flow paths in the Field Lab remained unchanged throughout the simulation period. In the adaptive AZP configuration, however, we observed an increase in flow reversals (Figure 5(b)) and flow velocity CV (Figure 5(e)) at links in the vicinity of Sensors 2 and 5. This is a direct consequence of dynamic DBV operations switching between partially opened (daytime) and closed (nighttime) states. Moreover, the dynamically adaptive network connectivity resulted in a shift of the hydraulic balance point from the original DMA boundary to an area just upstream in DMA 2. Consequently, this part of the network conveyed a much larger range in flow and experienced several flow reversals. Such dynamics are corroborated by the mean source trace results in Figure 5(h).
Flow variability was exacerbated in the adaptive AZP-SCC configuration. Most notably, activated PRV and AFV settings during the 1-h self-cleaning period shifted the hydraulic balance point even further towards the inlet of DMA 2. This shift is best depicted by the mean source trace results in Figure 5(i). Additionally, increased flow reversals and pipe flow velocity CV in Figure 5(c) and 5(f), respectively, were observed in looped or dead-end links connecting AFV actuators, highlighting the elevated flow velocities generated by flushing demands. The increase in changing hydraulic conditions from the self-cleaning mode emphasize the potential impact of advanced hydraulic control on water quality dynamics.
Water quality dynamics
The single-feed DMA configuration exhibited minimal disinfectant variability during the 24-h simulation, which was expected due to the relatively static hydraulic conditions. In contrast, the open network configurations corresponding to the adaptive AZP and AZP-SCC control objectives presented moderately higher disinfectant variability. Specifically, Figure 6(b) and 6(c) displays areas with higher standard deviation within the areas of increased hydraulic variability depicted in Figure 5. On the basis of these results, the proposed flow variability metrics can serve as a good surrogate measure for identifying increased disinfectant variability in the absence of a calibrated water quality model.
The observed disinfectant variability is mainly caused by the mixing of source water with different inlet concentrations. A larger disparity in inlet concentrations exacerbates this variability, as illustrated by the worst-case scenario in Figure 6(e) (AZP control) and Figure 6(f) (AZP-SCC control). Furthermore, the worst-case scenario exhibits a clear increase in the spatial extent of nodes with higher disinfectant variability. These simulations demonstrate the importance of controlling inlet concentrations for managing water quality in adaptive DMAs.
Time series analysis
In Figure 7, the relatively high variability observed at Sensors 2 and 5 correlates with increased hydraulic and water quality dynamics from our model simulations, supporting our hypothesis that increases in flow variability from networks operated with dynamic connectivity can lead to greater disinfectant variability. However, as discussed in Section 4.1.2, we emphasize that this correlation depends on the disparity in inlet concentrations and the decay of disinfectant materializing between the inlet and node where mixing occurs; in this case, near the DBVs. Consequently, the relatively modest disinfectant variability observed at Sensors 2 and 5 is likely due to comparable inlet concentrations recorded at Sensors 1 (DMA 2) and 4 (DMA 1). Furthermore, changes in flow velocity for the AZP configuration were relatively minor compared to the AZP-SCC controls, which have a 1-h flushing period. This suggests that additional disinfectant decay materializing from hydraulic dynamics is limited in the Field Lab. Subsequent experiments are planned to observe disinfectant decay behaviour for the more dynamic AZP-SCC control configuration. At Sensor 7, our simulations revealed several pipes with flow reversals and higher flow velocity variability in this region (see Figure 5(b) and 5(e)). The hydraulic balance point in this section of looped pipes appears to be continuously shifting due to changing demands. As a result, flow comprising parcels of water with much higher age is likely contributing to the observed disinfectant variability at Sensor 7.
CONCLUSIONS
The impacts of dynamically adaptive connectivity and hydraulic control on water quality dynamics were investigated in a real-world water distribution network. We simulated the variability of disinfectant residuals resulting from two adaptive control configurations: (i) minimization of AZP and (ii) integration of AZP with a SCC mode. Model simulations were supported by continuous disinfectant data from a large-scale water quality experimental programme. The main conclusions of this study are as follows:
Flow variability metrics. The proposed metrics effectively identified areas of the network with greater short-term disinfectant variability due to hydraulic control. These metrics can enhance water quality observability by informing additional monitoring locations and can be used to model water quality dynamics in optimization problems formulated for control.
Validation of model simulations. Continuous disinfectant data collected from April to July 2023 validated our simulation results for the adaptive AZP control configuration, revealing greater disinfectant variability near the hydraulic balance point of the aggregated (opened) DMA configuration. This variability was more pronounced during the period when DBVs transitioned from their closed to open position, which caused mixing between DMAs.
Influence of inlet concentrations. The extent of disinfectant variability caused by DMAs with dynamic connectivity is influenced by the disparity in disinfectant concentrations at the inlets. Controlling inlet concentrations is therefore critical for managing water quality in dynamically adaptive networks.
These conclusions support further analytical and experimental research needed to enhance water quality management in real-world systems. Future work will focus on the controllability of disinfectant residuals in adaptive networks by incorporating the proposed flow variability metrics into the formulation of optimal design and control problems. These control problems should also explicitly model water quality dynamics as constraints and consider the impacts of uncertainties in disinfectant decay parameters. Concurrently, ongoing experimental research is being conducted to observe water quality conditions over a broader range of hydraulic conditions, including the self-cleaning control configuration presented in this study. This experimental work also aims to further characterize water quality dynamics using field data collected across different seasonal, short-term, and long-term conditions, which is critical for improving the accuracy of water quality models used in control applications.
FUNDING
Bradley Jenks is supported by Imperial College London, Bristol Water Plc, Analytical Technology, and the Natural Sciences and Engineering Research Council of Canada (PGSD-577767-2023). Angeliki Aisopou is a Daphne Jackson fellow, supported by the Royal Society and the Engineering and Physical Sciences Research Council (EPSRC). Ivan Stoianov is an Anglian Water Services and CLA-VAL UK/Royal Academy of Engineering Senior Research Fellow in Dynamically Adaptive Water Supply Networks (RCSRF2324-17-41).
DATA AVAILABILITY STATEMENT
All relevant data are available at: https://github.com/bradleywjenks/wq_dynamics.
CONFLICT OF INTEREST
The authors declare there is no conflict.