Offshore wind farms are among the biggest machines we build—vast arrays of towers topped with slowly turning blades. They generate megawatts of power from their giant turbines, taking up miles of space.
That means that, as green as they are, they still have an immense impact on the ecology around them, affecting birds, fish, and even the growth of kelp and other marine plants.
Managing those turbines is a big issue. We can’t look at them in isolation as much as we’d like to. Instead, we need to consider them as part of a larger system, one that includes the environment they’re part of.
Instead of optimizing those turbines for power generation, we have to be able to control them to allow migrating birds to pass, at the same time ensuring marine plants don’t affect their moorings and that fishing boats don’t damage pylons as they follow shoals of herring and other fish into the farm.
It starts with puffins
The initial impetus for the project wasn’t a digital twin as such, instead it was using AI models to count the puffins on a remote island off the Scottish coast. As SSE Renewables was building a wind farm some 200 miles from a major puffin breeding ground on the Isle of May, the company wanted to know if the turbines were influencing the puffin population.
It’s hard to count puffins; they spend eight months of the year out at sea, returning to shore to breed, only laying one egg a year.
A set of cameras near the breeding burrows capture a live stream of puffin movements, which are fed to a trained model that can track individual birds, even noting when they leave and return.
The island is one of the U.K.’s largest puffin breeding grounds with over 80,000 birds, making it an ideal place to track fluctuations in population and to try to understand if the nearby wind farm is causing any changes.
Using AI to count puffins isn’t a digital twin, but it’s one input and one technique we can use to build a large-scale model of the environment around a wind farm. No two wind farms are the same: They use different turbine types and are built in different coastal waters and wind patterns.
As a result, they’re in different bird migratory patterns and host different species of fish. Any environmental model used as part of a control system needs to be custom for each wind farm.
Managing wind farms in the cloud
Part of the approach that Microsoft and its partner Avanade are taking is to use a wide range of different sensor types to get an understanding of what is happening around the wind farm, and using that data to build a complex, near-real time view of conditions. The aim is to remove slow, manual counting techniques, much like the puffin counting service currently in use.
Modern environmental sensors can be passive, like cameras or microphones, or active, like lidar and radar. That makes them less intrusive than using nets to sample fish or sending in divers to make a count.
An array of AI-interpreted sensors gets around the limitations that come with human intervention, collecting data in all conditions and at all times of day.
SEE:Hiring Kit: Cloud Engineer (TechRepublic Premium)
Models like this can take advantage of cloud scale to run multiple simulations in parallel at an accelerated rate. If a storm is approaching, what will be the effect of slowing the turbines, and to what speed?
The results of simulations like these can be compared with actual data, adding an extra feedback loop that lets the team refine their models, so the next set of results will be more accurate. The data can then be used to train machine learning models to identify conditions that are likely to cause problems, so appropriate protections can be applied.
Working with large, complex systems
This approach will allow SSE to experiment with reducing risks to migrating birds. For example, they can determine an optimum blade speed that will allow flocks to pass safely while still generating power. By understanding the environment around the turbines, it will be possible to control them more effectively and with significantly less environmental impact.
Simon Turner, chief technology officer for data and AI at Avanade, described this approach as “an autonomic business.” Here, data and AI work together to deliver a system that is effectively self-operating, one he described as using AI to “look after certain things that you understood that could guide the system to make decisions on your behalf.”
Key to this approach is extending the idea of a digital twin with machine learning and large-scale data. Historical data can be used along with real-time data to build models of large, complex systems, which can expand out to whole environments.
As Turner notes, this approach can be extended to more than wind farms, using it to model any complex system where adding new elements could have a significant effect, such as understanding how water catchment areas work or how hydroelectric systems can be tuned to let salmon pass unharmed on their way to traditional breeding grounds, while still generating power.
There’s another aspect to the wind farm project that reflects the ethos behind Microsoft’s AI for Earth program: All of the data gathered will be shared outside SSE Renewables and will be available to marine and other environmental researchers.
The resulting dataset should be a valuable resource for planning new wind farms and for any other continental shelf infrastructure projects. This adds another feedback path, allowing scientists to add their expertise and analysis to the data.
Using existing Azure services
Azure is an ideal platform for this type of application. Most of the tools needed to build it are already in place: Azure IoT Hub to manage sensors; Data Lake to process the massive data storage requirements; and Azure’s AI tooling to build, test and use the resulting models along with its existing Digital Twins product to host and run models.
It’s an approach that’s scalable and flexible enough to support the differences between wind farms built and operating in different places. As new data points are found they can be added to the models, allowing the platform to adapt to new data and to new questions from the team running the wind farm and managing its environmental impact.
Data will need to be stored for long periods, as the impact of a wind farm is one that’s years long, so models need to work over the order of seasons and years, even decades, not just minutes and seconds.
Large scale digital twins like this are the logical next step in the industrial Internet of Things. Microsoft is already seeing interest from other customers with complex systems that need monitoring and control.
That becomes a benefit for Microsoft itself, as it has a commitment to become carbon negative, so it needs to work with innovative renewable energy providers to develop new techniques to reduce its environmental footprint.
There’s another aspect to the use of massive environmental models like this, in that their outputs could be shared with other systems, for example providing data for Microsoft’s own precision agriculture platform FarmBeats.