Gondwana

Perspective on optimally designed distribution networks

The drinking water distribution network that we have in the ground today is the result of an evolution over time of the materials used, and of the design philosophy and objectives. The design often incorporates experience-based knowledge and extra safety margins. Now that many water utilities are facing a mounting replacement challenge, and moreover are considering making their networks ‘smart’, it makes sense to think about the kind of replacement network structures that need to be installed, and how they should be equipped with sensors and appendages. It is valuable, in this context, to conduct the design process in a more holistic manner, taking into account the various aspects, such as supply assurance, water quality, energy and, of course, costs. Although experts currently apply good strategies to these ends, numerical optimisation methods applied by experts could result in better designs than the classical approaches.

KWR has developed the Gondwana optimisation platform to bridge the gap between academic research and practical implementability.

Gondwana’s goal is to successfully meet the needs of water utilities with regard to flexibility and connection with different application possibilities. Examples of Gondwana’s application possibilities include:

• designing network blueprints and transitions from current configurations to network blueprints, with low costs and good performance as objectives. Performance can be determined, among others, by the level of reliability of supply, desired pressure, water quality, energy consumption and number of failures;
• determining optimal DMAs (District Metered Areas), with the aim of maximising the detectability of leakages and other anomalies using as few flow meters as possible;
• determining optimal valve configurations, with the aim of achieving low customer minutes lost (CML) and good section flushability;
• determining optimal locations for water-quality sensors or sampling, with the aim of minimising detection time, maximising detection probability or coverage level, facilitating source determination, etc.

Different scenarios, such as variations in water demand, failures and contaminations, can be taken into account in all applications.

Optimisation is possible as ‘implicit multiobjective optimisation’, which results in a single[JA1]  design that performs the best regarding the selected combination and weighting of objectives.

‘Explicit multiobjective optimisation’ is also possible; this results in a collection of designs, all of which, in varying proportions, perform optimally regarding the selected objectives. This collection is referred to as a ‘Pareto front’. An example is shown in Figure 1, in which the two axes show the performance on two objectives. Minimisation is desired for both objectives. Each point represents one design, but only the brown points shown left bottom-edge (the Pareto front) perform optimally. The user can compare the different designs on the Pareto front and decide to what extent an improvement regarding one objective justifies a concession on the other.

Figure 1: Trade-offs between two objectives (performances on the horizontal axis against those on the vertical axis), with the optimal solutions on the Pareto front shown in brown.

Gondwana optimisation tool: powerful, flexible and user-friendly

Gondwana is built around EPANET* (for the hydraulic simulation of pipe networks) and the Inspyred library** for the meta-heuristic optimisation methods. By means of a graphic interface and a series of tab pages (Figure 2), the user can set up an optimisation problem in a structured manner: load existing networks, define the requirements the renewed network needs to satisfy, and determine in what way it needs to be improved (objectives).

Figure 2: Gondwana’s main screen with tab pages for the definition of component aspects, performance of calculations and analysis of results. The example network concerns the New York tunnels benchmark problem.

Once an optimisation problem is set up, Gondwana can get to work: by randomly making changes to the decision variables, while structurally detecting and further incorporating the several resulting performance improvements, in a process known as an ‘evolutionary algorithm’, the calculations progress, step by step, towards a better performing network. This is a little like ‘natural evolution’, allowing the conduct of a rapid and efficient search through a (large) solutions space. Once the optimisation process is completed, the user can look at the results, compare them with the current design, and write them to the desired format. The tool is available for the performance of problem calculations by KWR.

[1] Rossman, L. A. (2000). EPANET 2 Users Manual, U.S. Environtmental Protection Agency.
[2] A. L. Garrett, Inspyred 1.0 Documentation, Jan. 2015.

Gondwana in practice

Over the last few years, Gondwana has been used for the implementation of DMAs and the design of, and transition to, network blueprints. The results obtained are both convincing and implementable:

• The division of the supply areas into DMAs offers insight into the trade-off between the number of flow meters required and the sensitivity of the DMAs for the detection of anomalies. Based on this information, water utilities can take well-founded decisions about the DMA division that best meets their requirements and expectations.
• The division of a supply area into DMAs can also be done transitionally, whereby the area is divided, step-by-step, into steadily smaller optimal DMAs. By following this sequence, the water utilities can make optimal use of previously acquired flow metres.
• In the design of network blueprints, it is apparent that the currently existing pipe networks can be significantly reduced in size. This means that cost-savings can be made, while the performance, in terms of supply assurance and pressure, is improved.
• The transition from the current structures to the optimal network blueprints offers insight into the relationship between the number of network locations subjected to annual work and the decrease in the number of failures or the improvement of the network’s hydraulic performance. This information enables the water utilities to make better-founded choices.

Looking to the future

At present, supplementary developments in Gondwana are being concluded and/or tested:

• determining optimal valve positions in order to maximise a network’s resilience in crisis situations (such as a pipe failure);
• designing robust network blueprints in which water-demand uncertainty is explicitly taken into account in the design process;
• optimising the locations of pressure sensors in order to enable the best possible detection and localisation of leakages – in collaboration with Brabant Water;
• reducing the number of flow meters required for the implementation of DMAs by closing a few boundary pipes; a procedure which can however render the network vulnerable to disruptions. This development is being undertaken jointly with Dunea.

Additional developments are foreseen in the future.

Gondwana involves the unlocking of powerful optimisation techniques for water utilities for application in various aspects of their networks. The resulting better-founded and optimal design decisions maximise the effectiveness of the resources invested.