The importance of system knowledge and intuition in automation

Our Gondwana optimisation tool

The ‘calibration’ must be very comprehensive; in fact, it encompasses all subject, operational and management choices for a system. Over the last few years, we have built up experience in the optimisation of the design of drinking water distribution systems using Gondwana, an optimisation tool we designed ourselves. Essentially, the question was always what diameter to choose for each pipe in a network. The main condition was that every customer, at any time, must have sufficient water (at a sufficient pressure) at their taps. Water pipes with large diameters make this easy, but smaller diameters are more interesting financially – in other words, costs versus customer comfort. But there are many other objectives and conditions (they are usually interchangeable) that come to mind, such as minimal velocity (self-cleaning networks), maximum residence time (water quality) and so forth.

Optimisation is no simple matter

The number of possible network designs quickly rises beyond the imaginative capacity of humans: if you assume that for 10 pipes you have the choice of 10 different diameters, you could build up a network in 10,000,000,000 different ways. And it soon becomes even more complicated when you try to incorporate other aspects of water-company daily practice into your optimisation. Take for instance the fact that the customers in all the other sections must not suffer any inconvenience (supply assurance) in the event that a specific network section is shut down (for maintenance work or emergency repair).

Setting up and conducting such an optimisation is no simple matter. It is very important that the system and all relevant behaviours be included in the optimisation model. This calls for a complete and updated network model, with water demand patterns that are representative of the situation(s) targeted by the optimisation. This is itself no trivial matter for many water companies. But that’s only the beginning. If you want to incorporate more complex aspects into the optimisation, then you need to include the associated behaviours in the model in a representative and quantitative manner. In the case of supply assurance, for instance, this refers first of all to the valves used by water companies for network sectioning. In general, this is reasonably-to-well-known at a water company. But it is also a question of which valves and production locations are manipulated if section X shuts down, in order to supply customers in section Y with enough water. And this knowledge is located primarily in the heads of people.

Knowledge of the system in needed

To properly evaluate the results of an optimisation, you also need to know exactly what input you are working with. And that brings us back to the expert at the water company. He or she is the only person who knows! We researchers at KWR rely for the most part on the data that they supply. Therefore, even with a powerful optimisation tool you need thorough knowledge of the system, both to feed the model correctly and to interpret the results. In this process, intuition is a useful tool to help recognise and interpret unexpected situations and solutions.

Collaborating for usable results

The most important lesson from our optimisation projects so far is the great importance of collaboration in the numerical optimisation of a drinking water distribution system. The expert at the water company must be able to explain precisely to the KWR researcher what the system does, so that it is properly represented in the optimisation model. And the KWR researcher must explain precisely to the water company expert what exactly the optimisation model does, so that the company expert can understand and interpret the outcomes. Only then can one achieve usable results. Unexpected solutions can then challenge and stimulate the experts. The significance of the experts thus remains undiminished. And a considerable added benefit of this process is that the implicit knowledge that the experts have in their heads can be explicitly recorded. In other words, it becomes much easier to transfer knowledge, including to the next generation of experts.

The initial results point to possible reductions in network volume of tens of percentage points. This translates into networks that are potentially millions of euros cheaper than existing ones. Fulfilling this potential requires continuous attention to the effective collaboration between the system expert and the optimising researcher.