Improving parameter selection for water infrastructure optimization

What do Chinese Feng Shui and water infrastructure optimization have in common?

Water and wastewater networks are one of the major infrastructure assets providing essential service to the society. The design of these networks is important for the new systems as they have to be fit for purpose. Equally, it is also important for the existing systems, which need to be upgraded due to ageing, to be expanded due to developments and raising water demands, or to become ‘smart’ by equipping them with various types of sensors, or simply because they need to be made future proof. The optimal design of those networks is extremely complex due to the large size of these systems and multiple, often competing goals, e.g., system and operational costs, service quality, system reliability and water quality.

Using multi-objective nature-based numerical methods, such as evolutionary algorithms, has received substantial attention among researchers in the past three decades. At KWR we have developed such a tool, the Gondwana optimization platform, to bridge the gap between academic research and practice. Gondwana has been successfully applied to many challenging problems, including: (i) designing network master plans and transitions from current configurations to future network blueprints, with low costs and good performance as objectives; (ii) determining optimal sectorization of networks into District Metered Areas, aimed at maximising the detectability of leaks; (iii) designing optimal valve configurations to minimize loss of supply to customers; and (iv) design sensor placement configurations (for water quality, pressure and other sensor types).

Making optimization tools even more powerful through visualization

Optimization tools require a good knowledge of the operational mechanisms to be able to take advantage of their capabilities. That means understanding how to set the optimization algorithm parameter values to achieve the best outcome, i.e., optimal design. When there are several of those parameters that potentially interact with others, i.e., influence each other in possibly unexpected ways, the task of finding the ‘optimal’ set of parameters becomes really difficult. Through collaboration with some colleagues from Guangdong University of Technology and Tsinghua University in China, we have developed a method to visualize the impact of various parameter combinations on the performance of the optimization method and select the best possible set for the problem at hand. The method is conceptualised based on ancient Chinese magnetic compass, also known as the luopan, or geomantic compass. This type of apparatus is used in Feng Shui to help individuals harmonize with their surrounding environment.

Figure 1: Luopan (Feng Shui) Compass

We have taken analogy with the concentric rings of the Feng Shui compass (Figure 1) to visualize the performance of an optimization algorithm based on combination of parameter values shown in concentric rings (Figure 2). Basically, we performed multiple runs of the optimization algorithm on the same design problem with 5 parameters within their extreme ranges (min/max).

Figure 2: Compass diagram of parametrization of optimized design of a water network

Each concentric ring represents two shades (depths) of a single colour, e.g., dark/light. Due to the stochastic character of the nature-based optimization, each combination of parameters has to be run multiple times. We then compare the results of those runs based on how often for each combination of parameters the algorithm came close to the best solution found (in grey colour and the outmost ring of the compass plot). The range of depths of the grey colour was used to show the frequency of a particular parameter combination achieving the best solution.

Our findings show that the interrelationships among the key parameters are complex and are case by case dependent. Thus, it is highly recommended to fine-tune these parameters, preferably following the method proposed in our paper, to confirm which combinations have great potential to identify the near-optimal solutions. This can save the computational budget and time substantially when dealing with larger, real-world design problems.