Andreas holding a branded mug with coffee
Learn about the future of renewable energy infrastructure
Please fill in a valid email
You are now subscribed to our newsletter!
Oops! Something went wrong while submitting the form.
Share this article

Oat Milk Latte with Andreas Malekos: What engineers need to know about evolutionary computing

We sat down for a latte (with oat milk, please) and a chat with Andreas Malekos, Head of AI at Continuum Industries, for an easy introduction into the world of evolutionary computing and algorithms.

Q: Andreas, what is evolutionary computing exactly?

A: Evolutionary computing is a set of algorithms that serve as problem-solving and optimisation superheroes inspired by nature. They take cues from the process of evolution and use them to tackle all sorts of optimisation problems. It’s pretty cool!

These algorithms view possible solutions to a problem as a population of individuals. They take them through an “evolutionary journey” towards a specific goal. With each iteration of the algorithm, the entire population moves a little bit closer to the goal. But they don’t just move in one direction. They can explore different paths simultaneously. In this way, the algorithm is able to not just optimise, but also explore the search space.

What makes evolutionary algorithms really amazing is that they treat the goal they are optimising for as a black box and focus on finding the best solution without getting caught up in the nitty-gritty details (unlike us, they do not care about the properties of the problem they are optimising). This makes them super versatile - we can throw different problems (or the same problem with significant variations) and be reasonably confident that they will do a decent job at finding a solution.

Q: What is Multi-Objective Optimisation (MOO)?

A: When it comes to optimisation, it's usually about finding that one perfect solution, right? Well, in the world of linear infrastructure design, things get a bit trickier. And that’s simply because the different people involved in the process of planning critical infrastructure care about different things.

Engineers will likely want the shortest and most feasible route, while environmental experts aim to minimise disruptions to sensitive areas. Project managers are often focused on cost and risk reduction, while local communities want to make sure their neighbourhood isn't negatively affected. See the problem? These goals are often in opposition with each other. For instance, we find that the shortest and most constructible route will rarely be the most environmentally friendly one.

All of these considerations are valid and important, but in different ways. So it’s clear that using a single objective way of thinking to tackle this problem will not work very well.

MOO is a set of techniques and algorithms that address this challenge by optimising for multiple goals at the same time. We try to find multiple solutions, each representing a different compromise between the goals that we are trying to achieve. If we want to focus on two objectives - for instance cost and “environmental friendliness”- the algorithm will give us  a solution that is the best for goal 1 and worse for goal 2, and vice versa. And we will also have a bunch of solutions that represent compromises between these two goals.

Q: Why is MOO considered to be better than other approaches?

A: Without sounding too academic, when it comes to MOO problems, some approaches try to simplify things by merging all the different goals into just one.

Now, here's the catch with these approaches:

  • Introducing bias: It's difficult to collapse all those goals without showing favouritism towards one or the other.
  • Scale struggles: Imagine trying to blend two things that are measured in totally different ways. For instance, you wouldn’t mix a "penalty" goal that's measured in thousands with a TOTEX goal that's measured in millions of currency units.
Q: How does MOO work in the context of Optioneer™?

A: MOO is completely integrated into Optioneer. When you create a configuration in Optioneer, you have the freedom to customise both costing and penalty models, giving the user total control.

Once Optioneer works its magic and returns results, you can explore the tradeoffs between the different objectives by comparing and evaluating the options generated. Optioneer’s user interface allows you to easily look at multiple options side by side and compare all sorts of metrics related to penalties or cost.

The tradeoffs involved in MOO are also apparent through heatmaps (one of our customers’ favourite features), which show a visual representation of where different solutions are distributed and allow users to identify possible corridors for routing.

With MOO at its core, Optioneer allows you to play around with multiple objectives but also ensures you have all the tools to make informed decisions.

Share this article