Around June 2024, we were contacted by a company in the renewable energy sector. They called us by phone, which is not common, and told us they had a problem with their wind farm designs.

The “Hard” Problem

Around June 2024, we were contacted by a company in the renewable energy sector. They called us by phone, which is not common, and told us they had a problem with their wind farm designs.

As you can understand, we had no idea or experience with this specific problem. We had worked in the sector on a satellite monitoring project for the construction status of onshore wind farms.

In this case, the problem is that placing wind turbines in an offshore wind farm is not trivial, and it determines how much energy they produce. The turbines must be optimally positioned so that the wind moves them and they generate energy.

The issue is that finding the optimal configuration is an NP-Hard problem, meaning there is no quick and easy solution. Many simulations and calculations must be performed to find the best way to place the wind turbines.

Our client’s approach to the problem was more of a “brute force” method. They generated many designs and evaluated them with simulation software (PyWake).

But… the optimal solution might not be in those designs. There might be a better way to place the wind turbines.

Open Innovation

This is where I wanted to get to. We don’t initially know about the problem - but we do know AI and Machine Learning. They don’t know as much about AI and Machine Learning, but they do know about the problem.

That’s how we help our clients:

• We contribute our experience in AI and Machine Learning.
• We contribute our experience in quality software development.
• We dive into the problem and study it.
• We conduct an “out of the box” analysis.

In this case, we first studied the state of the art and selected the most relevant scientific papers and different approaches in the market and academic world.

We realized that there are people who have dedicated their academic and professional careers to this problem and have published thoughtful articles about it. Humility, friends.

The Proposal

The first thing was to be honest with the client about not being able to guarantee a complete resolution of the problem. We can help them improve their design process and reduce the time it took to find the optimal solution. And we can contribute to the solution.

What we proposed:

• Use a wake simulation model based on PyWake but faster.
• Use that model to analyze production each time we place a new turbine.
• Treat the problem as a game in which an AI agent learns to optimally place turbines.

The Result

In a few months we achieved:

• Understanding the problem and its angles.
• Testing different wake simulation models.
• Launching Reinforcement Learning (RL) agents to place turbines that play millions of “games” in parallel.
• Generating improved wind farm layouts.

The Future

It remains an NP-Hard problem, but we’ve managed to reduce simulation time and improve design quality. We have AI models and new brute force models that allow us to explore the solution space more efficiently.

We’re working with real cases with other wind farm design stakeholders and seeing that the solution is scalable.

We participated as speakers at OTDChallenge2025 where we presented the project and the solution.