Optimization of Water Network Topology and Pipe Sizing to Aid Water Utilities in Deciding on a Design Philosophy: A Real Case Study in Belgium

Round 1

Reviewer 1 Report

Although the manuscript has the potential to advance our understanding of pipe network topology and water supply distribution, the explanation and figure require extensive revision. Please address the comments to improve the quality of your article.

1. First, all the figures are absolutely horrible. The figures require extensive revision.

2. Again figures 3 are horrible! I would make quality figures using Python, MATLAB, or whatever other professional software.

3. What is new about this research? Numerous optimization-based studies have been conducted on natural and artificial water distribution networks. I would suggest adding two to three lines to the introduction regarding natural and artificial water distribution network research using the following references. 

For natural, Gao et al. (2022), Analyzing the critical locations in response of constructed and planned dams on the Mekong River Basin for environmental integrity, Environmental Research Communications, https://iopscience.iop.org/article/10.1088/2515-7620/ac9459. And for artificial, Sarker (2021), Pipe Network Design and Analysis: An Example with WaterCAD, engrXiv. https://doi.org/10.31224/osf.io/c3aky 

Author Response

Although the manuscript has the potential to advance our understanding of pipe network topology and water supply distribution, the explanation and figure require extensive revision. Please address the comments to improve the quality of your article. 

Author’s response: 

We want to thank Reviewer 1 for the revision of our manuscript and the feedback given. We’ve taken all concerns raised into consideration. Replies to each one of the concerns are below.  

We also want to make clear that we had to anonymize the city used in the case study, as this was demanded by the water utility in question at a later stage. This means that we replaced “Ypres” by “City X” or in general ‘a city’ throughout the text. 

The author's list has also been adapted, using the form required by the journal.    

Comment 1: 

“First, all the figures are absolutely horrible. The figures require extensive revision.” 

Authors’ response: 

We have reconstructed the images from scratch in Python with as uniform a style as possible and with a 600-dpi quality.  

Comment 2: 

Again figures 3 are horrible! I would make quality figures using Python, MATLAB, or whatever other professional software. 

Authors’ response: 

We have reconstructed the images from scratch in Python with as uniform a style as possible and with a 600-dpi quality. 

Comment 3: 

What is new about this research? Numerous optimization-based studies have been conducted on natural and artificial water distribution networks. I would suggest adding two to three lines to the introduction regarding natural and artificial water distribution network research using the following references. 

For natural, Gao et al. (2022), Analyzing the critical locations in response of constructed and planned dams on the Mekong River Basin for environmental integrity, Environmental Research Communications, https://iopscience.iop.org/article/10.1088/2515-7620/ac9459. And for artificial, Sarker (2021), Pipe Network Design and Analysis: An Example with WaterCAD, engrXiv. https://doi.org/10.31224/osf.io/c3aky 

Authors’ response: 

To clarify the novelty of the proposed approach the section “1.2. Aim and novelties of the contribution” (section 1.3. in the revised manuscript) has been revised, this is now in lines 172-188 (see bold italic text): 

“In this contribution, we demonstrate how numerical optimization can be applied to the design of real-life WDN and what the added value of doing so is. To this end, we address a real-life study aimed at redesigning the (organically grown over many decades) pipe network providing water to a city in Belgium. Key to this redesign was that the utility De Watergroep not only wanted to reinforce its distribution network, but to also structurally modify the network’s topology to enhance the quality of water delivered in the future.  

Interestingly, the formulated mathematical problem represents a type of optimization problem – Simultaneous Layout and Pipe Sizing Optimization (SLaPSO),– that has not been conducted for a real-life sized distribution network. As discussed in the previous subsection, the studies that addressed this problem usually preserved the reliability of a network by penalizing design solutions whose all nodes are not double fed. However, such formulation is not appropriate for real-world networks. Firstly, such formulation would lead to unjustifiably expensive designs, and secondly, such designs are not even desirable in practice because of larger accumulation of sediments and water age resulting from lower pipe velocities. The key novelty of the presented paper lies in a novel approach to secure a network’s reliability using a constraint that penalizes the designs whose branched sections exceed a set threshold for the number of connections.” 

We appreciate the suggestions regarding the literature, but do not feel they contribute to our approach. We therefore chose not to include these in the literature review.  

Reviewer 2 Report

This work describes the procedure for designing a “masterplan” for a real water distribution network in the Netherlands, i.e., how the network could be redesigned from scratch to optimize specific performance objectives. The paper is very well written/structured and reads easily. Moreover, the problem of simultaneous topology and pipe sizing addressed in this work is very interesting, as it is not what it is traditionally considered in water distribution network design literature. The authors make a good job describing the challenges when solving this problem for a water utility. Indeed the most difficult part to design is to gather the data and requirements from the utility and tune the optimization algorithm accordingly. I have some suggestions for improving the paper and some questions for clarification:

1.       Even though it is clear that the procedure was tailored to the specific network and utility needs, what I’m missing is a more general mathematical description of the optimization problem, objectives, and constraints.

2.       Was cost considered in the design of the masterplan?

3.       The optimized network was designed for the current network and needs. If this is a plan to be implemented over the next decade, what changes in the network should be considered? Is the utility just going to redesign the masterplan? How would that translate in terms of capital costs?

4.       What are the peculiarities in the current network that may make this approach not applicable to other networks? e.g., a flat topology versus a network with large variations in elevation.

5.       The authors say in Section 2.1 that Gondwana was used for the optimization. Was this used as it is? Is it solving the SLaPSO problem? Which approach from those described in section 1.3 does it use?

6.       Section 2.3: Is it permitted for the algorithm to also add new pipes? This should be made clear. Also, if it is not allowed, please explain why.

7.       It is not clear how water quality is introduced into the objective or constraints. Is it due to larger water speeds?

 

8.       At least the algorithm used for this work could be made available online since this was funded by an EU grant.

Author Response

Review #2 

This work describes the procedure for designing a “masterplan” for a real water distribution network in the Netherlands, i.e., how the network could be redesigned from scratch to optimize specific performance objectives. The paper is very well written/structured and reads easily. Moreover, the problem of simultaneous topology and pipe sizing addressed in this work is very interesting, as it is not what it is traditionally considered in water distribution network design literature. The authors make a good job describing the challenges when solving this problem for a water utility. Indeed the most difficult part to design is to gather the data and requirements from the utility and tune the optimization algorithm accordingly. I have some suggestions for improving the paper and some questions for clarification: 

Author’s response: 

We want to thank Reviewer 2 for the constructive revision of our manuscript, the given feedback and interesting questions raised. We believe the questions and suggestions raised helped us improve the paper. We’ve addressed each one of the remarks, please see details hereafter.   

We also want to make clear that we had to anonymize the city used in the case study, as this was demanded by the water utility in question at a later stage. This means that we replaced “Ypres” by “City X” or in general ‘a city’ throughout the text. 

The author's list has also been adapted, using the form required by the journal.    

 

Comment 1: 

Even though it is clear that the procedure was tailored to the specific network and utility needs, what I’m missing is a more general mathematical description of the optimization problem, objectives, and constraints. 

Authors’ response: 

To address this comment the subsection “2.3. Optimization problem formulation” of the revised manuscript has been revised to include a more general mathematical description of the optimization problem. Please see the details in the text. 

Comment 2: 

Was cost considered in the design of the masterplan? 

Authors’ response: 

The cost is considered implicitly by minimizing the product length*diameter as a surrogate for the cost. However, only the cost (i.e., its surrogate) of the pipes that are topologically part of the looped section (double fed) of the network is included in the objective. The cost of the pipes forming the branched parts has deliberately been left out of the objective function, because these parts of the network are designed according to deterministic dimensioning rules in the Dutch code of practice (this design will be done by the utility in a consequent stage).  

This is explained in more detail in the revised version of the manuscript between, which is now in lines 277-287. 

Comment 3: 

The optimized network was designed for the current network and needs. If this is a plan to be implemented over the next decade, what changes in the network should be considered? Is the utility just going to redesign the masterplan? How would that translate in terms of capital costs? 

Authors’ response: 

This is a very interesting question, that we pose ourselves and the water utilities often (and that we love to discuss!). We often make a design that fits the information we have ‘today’, but in practice it will take years (decades) to implement the design, and it is expected that a lot will change during this time horizon. This is a challenge when implementing these approaches in practice. In different projects, we are looking into both the design and rehabilitation plan combined, and in how flexible results need to be – these are full research lines, which only highlights how important (and complex) this remark from the reviewer is. It is out of the scope of this project though.  

More specifically for this case, we do not know the time horizon that will be considered to implement this specific plan, this is, the rehabilitation of the current network. The design presented in this paper will serve as a guidance, but as future changes, we advise the water utility to revise the plan and include new relevant information in the problem formulation.  

Regarding uncertainty, the water demand used in the optimization process is that of the peak hour on the maximum day the past ten years, increased by 10% to account for potentially higher demand in the future. Hence, the optimal network design is made to account for future needs, although simplified. Alternative approaches who explicitly consider uncertainty are (again) very interesting but fall behind the scope of this problem.  

We address these concerns in lines 506-513.  

Comment 4: 

What are the peculiarities in the current network that may make this approach not applicable to other networks? e.g., a flat topology versus a network with large variations in elevation. 

Authors’ response: 

The approach considered is sufficiently generic to be applied to other areas. It must be noted that the specifics of objective functions and constraints might have to be fine-tuned for other areas, or other needs/aims of different water utilities, for instance, changing the penalty coefficients or values for the minimum pressure or maximum number of connections, or even adding additional constraints. This is, the method stands, but the details might need to be adapted. We address this concern in lines 502-505.  

 

Comment 5: 

The authors say in Section 2.1 that Gondwana was used for the optimization. Was this used as it is? Is it solving the SLaPSO problem? Which approach from those described in section 1.3 does it use? 

Authors’ response: 

To address this comment an additional paragraph is added to the revised manuscript between lines 213-230: 

“Thus far, the platform has been capable of solving different single and multi-objective optimization problems related WDN design such as: the classical design problem finding the optimal pipe diameters to minimize construction costs, dividing network in District Metered Areas and finding optimal locations for placing water quality or quantity sensors, among several others. To solve the mathematical problem formulated in this study, which fits to the area of scientific literature known as SLaPSO, new Python modules were added to Gondwana to expand its functionality. As described in subsection 1.3., the previous studies that addressed the problem of SLaPSO, the reliability of the supply is usually secured by penalizing design solutions in which nodes are not double fed. However, such formulation is less appropriate for real-world networks, as making designs where each node is double fed would be too expensive. Moreover, such designs are not even desirable in practice because of larger accumulation of sediments and water age resulting from lower pipe velocities. Consequently, a novel approach to secure reliability is proposed in this study suitable for real-world networks. Namely, the novel methodology implemented in Gondwana secures the reliability of the looped part of the network implicitly by introducing a constraint that penalizes the designs whose branched sections exceed a set threshold for the number of connections. The new formulation is described in detail in subsection 2.3.” 

Comment 6: 

Section 2.3: Is it permitted for the algorithm to also add new pipes? This should be made clear. Also, if it is not allowed, please explain why. 

Authors’ response: 

The algorithm does not permit the addition of new pipes. An additional sentence is added to the revised manuscript to clarify this and explains why (lines 275-276): 

“As De Watergroep aims to reduce the number of loops and increase the branched sections, the algorithm does not permit the addition of new pipes.” 

Comment 7: 

It is not clear how water quality is introduced into the objective or constraints. Is it due to larger water speeds? 

Authors’ response: 

The Reviewer is correct. The water quality has not been explicitly included in the optimization problem formulation, instead it was assumed that the reduction of the number of loops and pipe diameters would lead to higher pipe velocities and lower residence times. This is confirmed with the results of the residence time of the original and optimized networks. A part of the results section has been revised to clarify this point (lines 383-388): 

“Although, water quality has not been explicitly included in the optimization formulation, as expected, the optimized design with fewer loops and reduced pipe diameters led to higher velocities and consequently reduction of residence time. Namely, it has been calculated by De Watergroep's hydraulic experts that the average residence time decreases from 7.5 to 3.6 hours. A comparison of the new design with the current network is summarized in Table 3.” 

Comment 8: 

At least the algorithm used for this work could be made available online since this was funded by an EU grant. 

Authors’ response: 

The algorithm has been developed at our own investment costs, prior to the EU grant. We have clarified that the EU-grant was only partially used in line 520. 

Round 2

Reviewer 1 Report

Thanks for the revision. Please change figure 3 in a log-log scale. 

Author Response

We have added a log scale to the vertical axis of figure 3 to strengthen the visualization of the plateau reached by the convergence curve. We kept the horizontal axis linear because we feel a log axis would overly emphasize the first few generations only. If a logarithmic log axis is preferred, we ask the reviewer to clarify.