Real-time Monitoring and Operational Control of Drinking-Water Systems

Water distribution network models are used by water companies in a wide range of applications. The availability of a good model allows the users to test any methodology that, otherwise, could not be applied directly to the real network. A model can be used to predict future states of the network, to analyse the effect of manipulating the real network before doing it, or to simulate faulty states to locate leakages, among others. This presents the basis of water distribution network modelling. The hydraulic equations are presented in their matrix form, which will be used in the following. Demands can be calculated, using the matrix model, if all heads or flows are known. However, this information is not available in a real case. Consequently, a hydraulic solver is needed to simulate the network, computing heads and flows from a predefined set of demands and boundary conditions. Results are obtained using the extended period simulations of the steady-state models. Transients are not considered due to their low importance in large networks.

Process control and supervision are based mainly on the use of models. These models have to be as accurate as possible to generate reliable results. Complex systems, like water distribution networks, need such models in order to comprehend them. Models presented in Chap. 3 are used in simulation, optimization, supervision, leak detection, etc. When the model is generated, large errors are introduced. These errors discourage the technicians unless they are corrected in a first calibration effort: macrocalibration. This is an ad hoc process that is done manually. The methodology, carried out by the experts, can be partially addressed using artificial intelligence (AI) algorithms. Once the major errors are solved, the parameter tuning, microcalibration, is posed as an optimization problem. Before these procedures are applied, the problem and the information available have to be analysed in order to assure the reliability of the resulting model. Given a number of parameters to be estimated, the measurements required for guaranteeing the identifiability and the well-posedness of the problem may be too exigent. Thus, the sampling design is often associated with a redefinition of the parameters to be estimated. In this chapter, both the parameterization and the sampling design are presented proposing a methodology that has given promising results with real water distribution networks.

Demands are unknown inputs that must be defined to solve the network’s hydraulic equations. They are not physical elements of the network, but are the driving force behind the hydraulic dynamics. A good calibration of demands is mandatory to obtain accurate results when simulating the hydraulic model. The simulation of the models, presented in previous chapters, is useful in a wide range of applications, as leakage detection and isolation, control, quality supervision and other issues presented in following chapters. This simulation is also useful, together with real measurements, for the estimation of demands. This chapter presents an overview of the existing calibration methods and a detailed description of an on-line calibration procedure. The demand calibration requires the parameterization and the sampling design. The three processes are carried out using the information provided by the SVD of the network sensitivity matrix. An academic network is used as case study for a better understanding of the whole methodology and a real application on a DMA is also presented. The methodology applied to demands can be adapted to other parameters in the network such as roughness and emitter coefficients.

This chapter deals with forecasting of hourly water demand data of different sectors of a WTN using the data obtained by their flowmeters. Several methods to forecast the hourly water demand are studied and compared with the aim of being applied for the operational control of any water transport network. The short-term forecast of the intraday series has a main feature: the double periodicity (daily and hourly). To address this issue, several extensions of the classical time series forecasting methods are proposed: seasonal ARIMA, structural models and the exponential methods without external information. This chapter focuses on the daily and hourly forecasts applied to the Barcelona transport water network. In the hourly forecast, the exponential smoothing method is the most accurate. On the other hand, the seasonal ARIMA and the exponential smoothing are similar in the daily timescale.

The efficient use of water resources is a subject of major concern for water utilities and authorities. One of the main challenges in improving the efficiency of drinking-water networks is to minimize water loss in pipes due to leakage. Water leaks in water distribution networks are unavoidable. They can cause significant economic losses in fluid transportation and an increase in reparation costs that finally generate an extra cost for the final consumer due to the waste of energy and chemicals in water treatment plants. Besides, leaks may also damage infrastructure and cause third-party damage and health risks. The loss of about 15% of treated water in developed countries and 35% in developing countries is a rather critical issue in a world struggling to satisfy water demands of a growing population [1]. This chapter presents the state of the art of the leak management including the real-time monitoring that allows the leak detection and localization techniques for repair. Special attention will be devoted to model-based approaches. The deep description of a leak location method based on the sensitivity matrix and the correlation analysis will be provided in this chapter with the application to a real case study.

To ensure the safe supply of drinking-water, the quality must be monitored online. The consequence of inadequate monitoring can result in substantial health and economic risks. Modelling water quality in water distribution networks is necessary in order to ensure the delivery of high-quality drinking-water. A water quality model is a reliable tool only if it predicts how the real water quality behaves. This chapter focuses on the use of models for monitoring abnormal water quality situations in water distribution networks. This chapter presents a methodology that enables, in a least square sense with a normalized quadratic cost function, to efficiently calibrate a water quality model such that the field-observed water quality values match with the ones estimated using the model. Since this function involves a non-explicit expression of the model, a genetic algorithm is applied to optimize the model parameters by minimizing the difference between the model-predicted values and the field-observed ones. Determining and estimating the best parameters associated with the models are of crucial importance for decision-making. Another important issue is the water quality events that can be manifested by added variability and lower chlorine concentrations at sensor locations in the network. The water distribution network can be perceived as a complex chemical reactor in which various processes occur simultaneously. The second methodology proposed in this chapter is a water quality event detection and location method based on chlorine measurements and chlorine sensitivity analysis of the nodes in a network. Simulations of the water quality of the network carried out with realistic bulk decay and with an abnormal one provide an approximation of this sensitivity.

The goal of this chapter was to review recent developments on sensor placement for monitoring in water networks focusing on leak detection and location. Solving a sensor placement problem for fault diagnosis entails the evaluation of the diagnosis performance of a monitoring system for a certain sensor configuration. In model-based fault diagnosis, the monitoring system is built upon a set of consistency indicators, which are designed based on a model of the water network and evaluated on the measurements provided by the sensors. The leak monitoring performance is usually evaluated based on the sensitivity matrix, which states the leak sensitivity of the consistency indicators. Two approaches, based on different leak sensitivity matrix nature, for sensor placement will be reviewed. On the one hand, structural analysis leads to a binary leak sensitivity matrix, which can be efficiently handled by using optimization algorithms and graph theory. On the other hand, the use of an analytical model leads to a more informative non-binary leak sensitivity matrix. In both approaches, only suboptimal solutions can be attained. This is due to the fact that in the first approach, a structural model is used, which is an abstraction of the analytical model, whereas in the second approach the sensor placement problem must be solved applying clustering techniques. These two approaches will be recalled and compared when applied to a DMA in the Barcelona DWN to decide the best location of pressure sensors for leak monitoring.

In a real water network, a telecontrol system must periodically acquire, store and validate sensor data to achieve accurate monitoring of the whole network in real time. For each sensor measurement, data are usually represented by one-dimensional time series. These values, known as raw data, need to be validated before further use to assure the reliability of the results obtained when using them. In real operation, problems affecting the communication system, lack of reliability of sensors or other inherent errors often arise, generating missing or false data during certain periods of time. These data must be detected and replaced by estimated data. Thus, it is important to provide the data system with procedures that can detect such problems and assist the user in monitoring and processing the incoming data. Data validation is an essential step to improve data reliability. The validated data represent measurements of the variables in the required form where unnecessary information from raw data has been removed. In this chapter, a methodology for data validation and reconstruction of sensor data collected from a water network is developed, taking into account not only spatial models, but also temporal models (time series of each sensor) and internal models of the several components in the local units (e.g., pumps, valves, flows, levels). The methodology is illustrated by means of its application to flow and level meters of the Catalonia Regional Water Network.

A critical aspect of water transport networks (WTN) is the vulnerability of the system. Water system security depends, among other factors, on the capability to detect as soon as possible accidental or intentional contamination, sensor and actuator malfunctions (faults) or incorrect operations. Fault diagnosis (FD) aims at carefully identifying which fault (including hardware or software faults and external perturbations) can be guessed to be the cause of monitored events. In general, when addressing the FD problem, two strategies can be found in the literature: hardware redundancy based on the use of redundancies (adding extra sensors and actuators), and software (or analytical) redundancy based on the use of software/intelligent sensors (or a model) combining the information provided by the sensor measurements and the actuator commands. In large-scale systems, the use of hardware redundancy is quite expensive and increases the number of maintenance and calibration operations. This is the reason why, in WTN applications, FD systems that combine both hardware and analytical redundancy are usually developed. In this chapter, a review of the state of the art in fault diagnosis applied to WTN will be provided. Next, model-based FD procedures will be reviewed and mathematically formalized. This framework is based on checking the consistency between the observed and the normal system behaviour using a set of analytical redundancy relations (ARRs). ARRs compare the values of measured variables against the estimated values provided by a normal operation (fault-free) model of the monitored system. Finally, the Barcelona water transport network will be used as the case study to exemplify the FD methodologies.

Water transport networks (WTNs) are generally used to convey water from production plants or sources to storage tanks close to the consumptions areas. Tanks are usually built with enough elevation to guarantee the service pressure required for their associated consumption area. WTNs contain large water mains and control elements, such as pumping stations and valves, linking the sources to consumption areas. Their operational control involves planning the control actions at pumping stations and valves ahead in time for periods of 24 to 48 h, according to demand prediction. Then, the control problem is a resource allocation problem, with costs associated with water acquisition and treatment (production) and with electricity costs of pumping operations. Model predictive control (MPC) techniques are very suitable to perform the real-time operational control of water transport networks, as they can compute, ahead of time, the best admissible control strategies for valves, pumps or other control elements in a network to meet demands and achieve an operational goal. Typical goals in the management of water transport networks are minimum energy consumption, cost minimization, service safety, smoothness of control actions, pressure regulation and others. This chapter will show the fundamentals of control-oriented modelling in water transport networks and it will be shown with real case studies that MPC can provide an efficient solution to predictive water resource allocation, which outperforms traditional operational management, improving the above-mentioned operational goals.

This chapter proposes a nonlinear model predictive control (NMPC) strategy for WDNs including both flow and pressure constraints. A WDN might be regarded as a nonlinear system described by differential-algebraic equations (DAEs), when flow and hydraulic head equations are considered in the model. The main operational goal of WDNs is the minimization of the economic costs associated with pumping. In addition to the minimization of costs, the optimal operation of WDNs should guarantee water supply with required flows and pressures at all the control/demand nodes in the network. Other operational goals related to safety and reliability are usually sought. From a control point of view, NMPC is a suitable control strategy for WDNs, since the optimal operation of the network cannot be established a priori by fixing reference volumes in the tanks. Alternatively, the NMPC strategy should determine the optimal filling/emptying sequence of the tanks taking into account that electricity price varies between day and night and that the demand also follows a 24-h repetitive pattern. On the other hand, as a result of the ON/OFF operation of pumps in pumping stations, a two-layer control scheme has been utilized: the NMPC strategy at the hourly sampling timescale is chosen in the upper layer while the pump scheduling approach at the minutely sampling timescale dealing with pumps in the ON/OFF manner is proposed in the lower layer. Finally, results of applying the proposed control strategy to a portion of the Barcelona WTN are provided in simulation.

Two formulations of the stochastic model predictive control (SMPC) problem for the control of large-scale drinking-water networks are presented in this chapter. The first approach, named chance-constrained MPC, makes use of the assumption that the uncertain future water demands follow some known continuous probability distribution while at the same time, certain risk (probability) for the state constraints to be violated is allocated. The second approach, named tree-based MPC, does not require any assumptions on the probability distribution of the demand estimates, but brings about a complexity that is harder to handle by conventional computational tools and calls for more elaborate algorithms and the possible utilization of sophisticated devices.

This chapter proposes a reliable fault-tolerant model predictive control applied to drinking-water transport networks. After a fault has occurred, the predictive controller should be redesigned to cope with the fault effect. Before starting to apply the fault-tolerant control strategy, it should be evaluated whether the predictive controller will be able to continue operating after the fault appearance. This is done by means of a structural analysis to determine loss of controllability after the fault complemented with feasibility analysis of the optimization problem related to the predictive controller design, so as to consider the fault effect in actuator constraints. Moreover, by evaluating the admissibility of the different actuator-fault configurations, critical actuators regarding fault tolerance can be identified considering structural, feasibility, performance and reliability analyses. On the other hand, the proposed approach allows a degradation analysis of the system to be performed. As a result of these analyses, the predictive controller design can be modified by adapting constraints such that the best achievable performance with some pre-established level of reliability will be achieved. The proposed approach is tested on the Barcelona WTN.

Large-scale systems (LSS), such as WTNs, present control theory with new challenges due to the large size of the plant and of its model. In order to apply decentralized or distributed control approaches to LSS, there is a prior problem to be solved: the system decomposition into subsystems. The importance of this issue has already been reported in the general literature of decentralized control of LSS. The decomposition of the system into subsystems could be carried out during the modelling of the process by identifying subsystems as parts of the system on the basis of physical insight, intuition or experience. But, when a large-scale complex system with many states, inputs and outputs is considered, it may be difficult, even impossible, to obtain partitions by physical reasoning. A more appealing alternative is to develop systematic methods, which can be used to decompose a given system by extracting information from its structure and representing it as a graph. Then, this structural information can be analysed by using methods coming from graph theory. This chapter discusses partitioning approaches towards the development of subsystem decomposition methods for LSS by reviewing automatic decomposition algorithms and techniques based on graph partitioning. The general aim of the discussed methods is to provide decompositions consisting of sets of non-overlapping subgraphs whose number of vertices is as similar as possible and the number of interconnecting edges between them is minimal. The real case study based on the Barcelona WTN in Chap. 2 is used to exemplify the discussed decomposition methodologies.

This chapter discusses the application of non-centralized MPC (NCMPC) approaches to DWTNs. The aim of DMPC is to reduce the computational burden and to increase the scalability and modularity with respect to the centralized counterpart, but still maintaining a convenient level of suboptimality with respect to the desired control objectives. Moreover, the advantage of NCMPC approach is the simplicity of its implementation given the absence of negotiations among controllers, which allows for a simple implementation.

This work addresses the design of a control strategy for drinking-water transport networks (DWTNs) based on evolutionary-game theory (EGT). This theory allows to model the evolution of a population composed by a large and finite number of rational agents, which are able to make decisions. As an analogy with a multi-variable control system for DWTN, the whole population represents the total available water resource in the system, and each agent represents a small portion of the resource. In the population evolution, each agent makes the decision to select one of the system valves and/or pumps in order to establish its corresponding value of resource. Agents make these decisions pursuing an improvement of their benefits described by a fitness function, which is associated to the control objective, i.e., agents receive more benefits as the control objective is achieved. This global objective in the DWTN is established by the company in charge of the management of the network, e.g., maintain safety volumes within the tanks, minimize the water costs, minimize the costs of the energy to operate the actuators. The aforementioned evolution process, in which agents make decisions, is used to solve an optimization problem that is described in terms of current measurements of the DWTN tank volumes and subject to constraints over the decision variables in the system (physical limits of flows through valves and pumps). Furthermore, since the control problem is given in terms of instant measurements, this control strategy might be implemented without the need of an explicit model of the DWTN. In this work, two different data-driven population-games-based control designs for DWTNs are presented, and both the necessary assumptions and conditions to implement the proposed methodologies are clearly stated. Finally, the effectiveness of the proposed control approach through the system performance improvement is shown by using the considered DWTN case study.

The coordinated management of regional and urban networks is a challenging real-time control problem because of the reduced water resources, intensive energy requirements and increased attention towards the environmental impact of water use. Optimal coordinated management of such water networks involves more difficulties because of different dynamics and control requirements of each network. This chapter proposes a multi-layer model predictive control (MPC) with temporal multi-level coordination for regional and urban networks. Inside each network, an MPC-based controller is used. Between the regional and urban networks, a temporal multi-level coordination mechanism is used to generate control strategies that consider the objectives and timescales in the different networks. According to real practices, a regional network works in a daily scale in order to achieve the global management policies for the different reservoirs, while the urban network works in a hourly scale and is in charge of manipulating the actuators (pumps and valves) set points to satisfy local objectives. Real-life pilot demonstration in Catalonia Regional Network (Spain) will be used to prove the general applicability of the proposed solution and its effectiveness for improving the efficiency of water use, energy consumption and the reduction in computing load.

The volume of data collected by a water utility is constantly growing. In this new era, data are important because guarantees the success of decisions based on the relevant values and underlying information extracted from noisy data. For instance, automatic meter reading (AMR) systems offer households and businesses the chance to understand and reduce their energy and water usage in much greater detail than previously possible, when meter readings were taken once a quarter, or even annually. Moreover, AMR could help utility firms to improve the accuracy of billing and cut visits to properties to read meters. However, with AMR, there is an exponential growth of data: an AMR produces 17500 readings per year, with a single reading every half hour. These data should be first processed in a real-time streaming, in order to be validated before being stored and translated into a metadata model which may be usable in multiple further applications. Thus, utilities have found scaling smart meter management systems difficult to handle. This motivates the use of Big Data technologies in this application domain. On the other hand, applying data analytics and knowledge discovery tools to AMR data combined with other streams of information (data coming from the billing system, call centre service and meteorological information) could help with fraud detection, maintenance requirements prediction, water/energy user consumption patterns determination and response generation to variations in the demand. This chapter presents novel algorithms and methodologies to carry out real-time streaming data processing, data analytics, data quality assessment and improvement, as well as prediction and visualization tasks, at extremely large scale and with diverse structured and unstructured data from multiple sources such as water, power, telecommunication and other utilities, as well as from social media. The algorithms and methodologies will be illustrated using real data coming from several water utilities.