Lost in optimisation of water distribution systems? A literature review of system operation
Summary (3 min read)
1 Introduction
- Water distribution systems (WDSs) represent a vast infrastructure worldwide, which is critical for contemporary human existence from all social, industrial and environmental aspects.
- A level of flexibility exists in the WDSs, which enables the supply of required water under different operational schedules, more or less economically.
- Since the 1970s, substantial research has addressed the operational optimisation of WDSs (Ormsbee and Lansey 1994) with two main areas of focus.
- Development in the use of various methods to optimise operation of WDSs is not only an interesting subject for research, but is also very complex.
2 Aim, scope and structure of the paper
- The aim of this paper is to provide a comprehensive and systematic review of publications for operational optimisation of WDSs since the end of the 1980s to nowadays to contribute to the existing review literature (Lansey 2006; Ormsbee and Lansey 1994; Walski 1985 ).
- Publications included in this review are relevant to optimal pump operation, valve control and optimal system operation for water quality purposes of both urban drinking and regional multiquality WDSs.
- The purpose of this part of the paper is to provide the current status, analysis and synthesis of the current literature, and to suggest future research directions.
- The table forms a significant part of the paper referring to over a hundred publications and is structured chronologically.
3.1 Pump operation
- Typically, electricity consumption is one of the largest marginal costs for water utilities.
- The latter two methods reduce the number of variables hence decrease the size of the search space.
- A multi-objective approach has been increasingly applied to pump optimisation problems to include considerations other than costs.
- Most recently, water quality has been traded off against pump operating costs (Arai et al.
3.1.1 Real-time control
- Time is an important factor for industrial applications.
- Evidence from the literature suggests that computational efficiency of metaheuristic algorithms in conjunction with the network simulator, such as EPANET, for large WDSs is not sufficient, however.
- Several authors have investigated how to decrease computational effort of the network simulator and/or an optimisation algorithm to provide an optimal solution in real-time.
- ANNs, which are applied most frequently, were used to determine real-time, near optimal control of WDSs by integrating with a GA incorporating demand forecasting (based on seasonal, weekly and daily periodic components) and operating continually based on SCADA data and demand forecast updates (Martinez et al.
- An open question is how to control the error of the surrogate model to ensure that the solution found is still optimal when the full network simulator is employed to validate it.
3.2.1 Urban drinking water distribution systems
- There does not seem to be a unique optimisation model for the operation of drinking WDSs.
- The third optimisation model minimises disinfectant concentration deviations at customer demand nodes from desired values (Goldman et al.
- The explanation provided was that the objective function implemented in the third model (i.e. concentration deviations) does not force the algorithm to reduce pump operating time/costs further after all of the constraints are satisfied.
- Ostfeld and Salomons (2006) discovered that pumping costs are significantly reduced if water quality is absent from the optimisation model and conversely, that the best water quality outcome corresponds to the highest pump operating costs.
3.2.2 Regional multiquality water distribution systems
- Multiquality WDSs are "systems in which waters of different qualities are taken from sources, possibly treated, conveyed and supplied to the consumers" (Ostfeld and Salomons 2004) .
- These costs were combined into one objective, with water quality requirements at customer demand nodes included as constraints.
- Optimisation problems in the above papers were solved as single-objective.
- Interestingly, when two water quality objectives (each representing a separate water quality parameter) are incorporated together with a pumping cost optimisation into a model, the relationship between water quality and pumping costs is not necessarily conflicting (Mala-Jetmarova et al. 2015) .
3.3 Valve control
- Valve controls were used in conjunction with both optimal pump operation and optimal system operation for water quality purposes.
- Additionally, percentages/degrees of valve closures (Kang and Lansey 2009; Kang and Lansey 2010) or openings (Ostfeld and Salomons 2006) were used to optimise chlorine levels across a network.
- In general, the pumping flow is often the main decision variable used in operational optimisation of WDSs.
4.1 Application area
- As described in Section 3, there are three application areas: pump operation (Section 3.1), water quality management (Section 3.2) and valve control (Section 3.3).
- The largest portion of papers (41%) is concerned with optimisation of pump operation only.
- Optimisation of water quality exclusive of any other operational controls (i.e. pumps and/or valves) is addressed in 15% of papers. .
- The introduction of water quality criteria, with or without valve control for pressure management (e.g. for leakage control) or water quality manipulation, appeared much later in the literature.
- Lately, more emphasis was put on holistic assessment of WDS operation, and thanks to more sophisticated simulation and optimisation methods having been introduced.
4.2 Optimisation model
- Regarding optimisation models, each is mathematically defined by three types of components: objectives, constraints and decision variables.
- Therefore, their assessment is limited to their interpretation of the provided information in the publications, where explicit formulation was partially presented or missing altogether.
- The number of types of a decision (i.e. control) variable included in optimisation models ranges from one to seven.
- A majority of optimisation models, 41% and 33%, uses one or two types of a decision variable, respectively.
- Use of more than two types of a decision variable is less frequent and the number of such models tends to decrease with the increasing number of decision variables used.
4.3 Solution methodology
- Optimisation methods have developed significantly since the 1970s.
- Further efforts to improve the computational efficiency of various optimisers led to the development and integration of surrogate models within optimisation algorithms.
4.4 Test network
- A large variety of test networks has been used in operational optimisation of WDSs.
- Large networks are being simplified for the purpose of optimisation (Cembrano et al.
- There are two test networks, which have been used 10 or more times.
5 Future research
- In contrast, it is important to develop understanding of the impact of assumptions while using simplified simulation models or surrogate models (for example in real-time control) and to control the error of the surrogate model to ensure that the solution found is still optimal.
- Concerning the methods for search space reduction, an open question is how to perform it without compromising the fidelity of the optimisation problem and undue simplification of the real system.
- While using metaheuristic algorithms, methodologies for algorithm parameter selection such as in Gibbs et al. (2010b) and Zheng et al. (2015) need to be developed.
- In multi-objective optimisation approach, methods need to be developed for selecting the best solution(s) from the Pareto set, which is representative and sufficiently small to be tractable.
6 Summary and conclusion
- This paper presented a literature review of optimisation of WDS operation since the end of 1980s to nowadays.
- The papers reviewed are relevant to optimal pump operation inclusive of real-time control, valve control and optimisation for water quality purposes for urban drinking as well as regional multiquality WDSs.
- The value of the paper is that it brings together the majority of journal publications for operational optimisation of WDSs, over two hundred in total, which have been published over the past three decades.
- Uniquely, it also contains extensive information for over one hundred publications in a tabular form, listing optimisation models inclusive of objectives, constraints, decision variables, solution methodologies used and other details.
- The lack of understanding and accepted means for incorporating uncertainties in demand forecasting and network behaviour prediction models (both quantity and quality) are, among others, the factors limiting wider implementation of those models.
Did you find this useful? Give us your feedback
Citations
257 citations
Cites background from "Lost in optimisation of water distr..."
...In the past decade, multiobjective optimal design and rehabilitation of a WDN has attracted an increasing attention [40]....
[...]
185 citations
Cites background from "Lost in optimisation of water distr..."
...such as vehicle routing [1], robot gripper optimization [2], and water distribution system design [3]....
[...]
169 citations
112 citations
111 citations
Cites background or methods or result from "Lost in optimisation of water distr..."
...However, WDS simulations may still be computationally prohibitive even with more efficient deterministic or hybrid optimisation methods, especially as the fidelity of the model and the number of decision variables increase [22]....
[...]
...practical and rep sentative ubset of the non-dominated et that i sufficiently small to be tractable [22]....
[...]
...The research question resulting from this challenge is how to select the best solution(s) from the Pareto set, which may involve providing the decision makers with a practical and representative subset of the non-dominated set that is sufficiently small to be tractable [22]....
[...]
...In order to be consistent with the previous review [22], network size is expressed by the number of nodes within a network....
[...]
...it was not until the 1990s when these methods became more popular [20] due to their ability to solve complex, real-world problems for which deterministic methods incured difficulty or failed to tackle them at all [21,22], and to also control multiple objectives....
[...]
References
14,477 citations
Additional excerpts
...Since PSO considers unconstrained problems, a penalty function is used to handle constraints....
[...]
...Wegley et al. (2000) SO Optimal pump operation considering variable speed pumps using PSO....
[...]
...Optimisation method: PSO (Eberhart and Kennedy 1995)....
[...]
...ACO = ant colony optimisation ADP = approximate dynamic programming AMALGAM = a multialgorithm genetically adaptive method ANN = artificial neural network ARIMA = autoregressive integrated moving average ASA = adaptive search algorithm ASib = ant system iteration best (algorithm) CCPP = calcium carbonate precipitation potential CNSGA = controlled elitist nondominated sorting genetic algorithm COPA = changing operation in pollutant affectation (module) CPU = central processing unit CWQ = consistent water quality (sources) D = design DAN2-H = hybrid dynamic neural network DBP = disinfection by-products DCA = direct calculation algorithm DP = dynamic programming DPG = decomposed projected gradient DRAGA = dynamic real-time adaptive genetic algorithm EA = evolutionary algorithm EF = emission factor ENCOMS = energy cost minimisation system EPS = extended period simulation fmGA = fast messy genetic algorithm FMS = full mixing step FP = full parameterisation (approach) GA = genetic algorithm GAPS = genetic algorithm for pump scheduling GHG = greenhouse gas (emissions) H-W = Hazen-Williams (head-loss equation) HSA = harmony search algorithm ILDS = improved limited discrepancy search IP = integer programming ISM = interpretive structural modelling ISS = in-station scheduling (approach) IWQ = inconsistent water quality (sources) LDS = limited discrepancy search LLS = linear least square LP = linear programming LPG = linear programming combined with a greedy algorithm LRO = linear robust optimal (policy) MILP = mixed integer linear programming MINLP = mixed integer nonlinear programming MIP = mixed integer programming MIQP = mixed integer quadratic programming MO = multi-objective MOGA = multiple objective genetic algorithm NLP = nonlinear programming NPGA = niched Pareto genetic algorithm NPV = net present value NSGA = nondominated sorting genetic algorithm NSGA-II = nondominated sorting genetic algorithm II OI = operational intervention OP = operation OPTIMOGA = optimised multi-objective genetic algorithm PBA = particle backtracking algorithm PMS = partial mixing step POWADIMA = potable water distribution management (a research project) PP = partial parameterisation (approach) PRV = pressure reducing valve PSO = particle swarm optimisation Q-C = flow-quality (model) Q-H = flow-head (model) Q-C-H = flow-quality-head (model) QP = quadratic programming RM = reduced model (i.e. skeletonised model of a WDS) RR = replacing reservoir (approach) SA = simulated annealing SARIMA = seasonal autoregressive integrated moving average SCADA = supervisory control and data acquisition SDW = safe drinking water SLO = series of the local optima SO = single-objective SPEA = strength Pareto evolutionary algorithm SPEA2 = strength Pareto evolutionary algorithm 2 SQP = sequential quadratic programming TDS = total dissolved solids TOC = total organic carbon WDS = water distribution system WTP = water treatment plant...
[...]
...PSO derives solutions from both local and global searches by using a value of the inertial weight....
[...]
5,062 citations
Additional excerpts
..."SPEA2: Improving the Strength Pareto Evolutionary Algorithm."...
[...]
...Lopez-Ibanez et al. (2005) MO Optimal pump operation using SPEA2....
[...]
...ACO = ant colony optimisation ADP = approximate dynamic programming AMALGAM = a multialgorithm genetically adaptive method ANN = artificial neural network ARIMA = autoregressive integrated moving average ASA = adaptive search algorithm ASib = ant system iteration best (algorithm) CCPP = calcium carbonate precipitation potential CNSGA = controlled elitist nondominated sorting genetic algorithm COPA = changing operation in pollutant affectation (module) CPU = central processing unit CWQ = consistent water quality (sources) D = design DAN2-H = hybrid dynamic neural network DBP = disinfection by-products DCA = direct calculation algorithm DP = dynamic programming DPG = decomposed projected gradient DRAGA = dynamic real-time adaptive genetic algorithm EA = evolutionary algorithm EF = emission factor ENCOMS = energy cost minimisation system EPS = extended period simulation fmGA = fast messy genetic algorithm FMS = full mixing step FP = full parameterisation (approach) GA = genetic algorithm GAPS = genetic algorithm for pump scheduling GHG = greenhouse gas (emissions) H-W = Hazen-Williams (head-loss equation) HSA = harmony search algorithm ILDS = improved limited discrepancy search IP = integer programming ISM = interpretive structural modelling ISS = in-station scheduling (approach) IWQ = inconsistent water quality (sources) LDS = limited discrepancy search LLS = linear least square LP = linear programming LPG = linear programming combined with a greedy algorithm LRO = linear robust optimal (policy) MILP = mixed integer linear programming MINLP = mixed integer nonlinear programming MIP = mixed integer programming MIQP = mixed integer quadratic programming MO = multi-objective MOGA = multiple objective genetic algorithm NLP = nonlinear programming NPGA = niched Pareto genetic algorithm NPV = net present value NSGA = nondominated sorting genetic algorithm NSGA-II = nondominated sorting genetic algorithm II OI = operational intervention OP = operation OPTIMOGA = optimised multi-objective genetic algorithm PBA = particle backtracking algorithm PMS = partial mixing step POWADIMA = potable water distribution management (a research project) PP = partial parameterisation (approach) PRV = pressure reducing valve PSO = particle swarm optimisation Q-C = flow-quality (model) Q-H = flow-head (model) Q-C-H = flow-quality-head (model) QP = quadratic programming RM = reduced model (i.e. skeletonised model of a WDS) RR = replacing reservoir (approach) SA = simulated annealing SARIMA = seasonal autoregressive integrated moving average SCADA = supervisory control and data acquisition SDW = safe drinking water SLO = series of the local optima SO = single-objective SPEA = strength Pareto evolutionary algorithm SPEA2 = strength Pareto evolutionary algorithm 2 SQP = sequential quadratic programming TDS = total dissolved solids TOC = total organic carbon WDS = water distribution system WTP = water treatment plant...
[...]
...Other metaheuristic algorithms included particle swarm optimisation (PSO) (Wegley et al. 2000), ant colony optimisation (ACO) (Hashemi et al. 2014; Lopez-Ibanez et al. 2008; Ostfeld and Tubaltzev 2008), nondominated sorting genetic algorithm II (NSGA-II) (Prasad et al. 2004), strength Pareto evolutionary algorithm 2 (SPEA2) (Kurek and Ostfeld 2013), harmony search algorithm (HSA) (Kougias and Theodossiou 2013), limited discrepancy search (LDS) (Ghaddar et al. 2014) and other multi-objective algorithms (Baran et al. 2005)....
[...]
...Two optimisation problems are solved, each includes a different water quality measure, the first chlorine concentrations and the second water age. tanks using SPEA2. monitoring nodes (including tanks)....
[...]
4,976 citations
4,867 citations
"Lost in optimisation of water distr..." refers methods in this paper
...The algorithms are evaluated using standard multi-objective test functions (Zitzler et al. 2000)....
[...]
...Gaps is written in C++ and was applied to several test cases by Poloni and Pediroda (2000); Van Veldhuizen and Lamont (1998); Zitzler et al. (2000) involving both continuous and discrete variables....
[...]
[...]
3,645 citations
Related Papers (5)
Frequently Asked Questions (15)
Q2. What are the future works mentioned in the paper "Lost in optimisation of water distribution systems? a literature review of system operation" ?
Future research challenges for operational optimisation of WDSs are listed in Figure 6 and grouped according to steps involved in optimisation: ( i ) simulation model, ( ii ) optimisation model, ( iii ) optimisation method, and ( iv ) solution postprocessing. Regarding optimisation problems with water quality aspects, future research may consider the development of an optimisation model with an inbuilt flexibility for a general WDS, which could be customised for a specific WDS. A further research challenge is to analyse relationships between pumping costs and water quality using a set of realistic case studies to ascertain whether they are conflicting objectives or they can be somehow integrated, leading to reduced optimisation problem complexity. A methodology for an objective comparison of optimisation methods should be developed, so the best optimisation method for a particular case can be selected.
Q3. Why have deterministic methods started to appear in recent years?
In recent years however, deterministic methods have started toreappear, because they are more computationally efficient, thus more suitable for real-time control, as well asother applications (Creaco and Pezzinga 2015).
Q4. What are the constraints included in the optimisation models?
Please note that hydraulic constraints (such as conservation of mass of flow, conservation of energy, andconservation of mass of constituent) were not included in these statistics as they are normally included asimplicit constraints and forced to be satisfied by WDS modelling tool, such as EPANET.
Q5. What is the common type of surrogate model?
which are by far the most commonly used surrogate models, are based upon real neurologicalstructures and can be represented as directed graphs.
Q6. What is the proportion of objectives included in optimisation models?
The number of objectives included in optimisation models ranges from one to four, with a vast majorityof models (84%) being single-objective.
Q7. What is the importance of water distribution systems?
Water distribution systems (WDSs) represent a vast infrastructure worldwide, which is critical forcontemporary human existence from all social, industrial and environmental aspects.
Q8. What is the popular example of a hypothetical WDS?
Regarding optimisation problems with water quality aspects,future research may consider the development of an optimisation model with an inbuilt flexibility for ageneral WDS, which could be customised for a specific WDS.
Q9. What are the main types of surrogate models used to replace and approximate network simulations?
Surrogate models are efficienttools used to replace and approximate network simulations which can be very computationally expensiveand/or may become an obstacle in real-time implementations.
Q10. What is the level of flexibility in the WDSs?
A level of flexibility exists in the WDSs, which enables the supply of required water underdifferent operational schedules, more or less economically.
Q11. What are the main criteria for the classification of water quality optimisation for WDSs?
Based on the selected literature analysis, the following are the four main criteria for the classification ofoperational optimisation for WDSs: (i) application area, (ii) optimisation model, (iii) solution methodologyand (iv) test network.
Q12. What are the main application areas for optimisation of water quality?
As described in Section 3, there are three application areas: pump operation (Section 3.1), water qualitymanagement (Section 3.2) and valve control (Section 3.3).
Q13. What were the first optimisation models for multiquality WDSs?
The first optimisation models formultiquality WDSs considered pump operating costs only (Mehrez et al. 1992; Percia et al. 1997).
Q14. What is the common type of optimisation of water quality?
Optimisation of water quality exclusive of any other operational controls (i.e. pumps and/or valves) isaddressed in 15% of papers.
Q15. What is the promising way for selecting the solution from the Pareto set?
This mismatch leads to the research question of what is the most promising way for selecting the bestsolution from the Pareto set, which may involve providing the decision makers with a globally representativesubset of the non-dominated set that is sufficiently small to be tractable.