Part 1:
Why cloud-based analytics have underdelivered for wind energy

Have these digital investments delivered?

Over the last decade, wind energy developers and wind turbine manufacturers have invested heavily in data analytics and AI.

Countless companies and studies have claimed digital technologies (e.g. AI/ML, IoT, and predictive analytics) will detect underperformance, reduce downtime, prevent faults, increase lifespan, and so on, enabling new levels of efficiency, and oil & gas profit levels.

As a consequence, wind energy players have invested massively in large data/IT infrastructures, cloud solutions, and digital services, in order to gather terabits of data on wind speed, air conditions, blade speed and angle, power output, vibrations, temperature, etc, from thousands of wind turbines equipped with hundreds of sensors.

The increase in data has certainly grown research into applying analytics, AI & ML to wind energy. A recent study^[1] shows exponential growth in the number of publications from 2000 to 2022.

And this research has created valuable models for optimizing energy generation, such as wind farm control (e.g. wake adjustments, alarms, control systems), wind speed prediction, and the monitoring of electrical equipment, mechanical parts, and faults.^[3]^,^[2]

However, this promise in the lab has rarely led to large scale real-life deployments. A few rare, published successes include a 2.7% energy production increase from active control of yaw misalignment^[3],

and increased wind capacity per area of land of up to 22%^[4]^,^[5] from better wake steering and optimized windfarm layouts. But most studies do not mention real world impact.

Data released by industry tells a similar story. Most big players have invested in digital optimization. Publicly available data on these is limited, but published results vary from a 1% Annual Energy Production (AEP) increase for specific wake steering solutions^[6] to up to 5% for more advanced features^[7]. OEMs assert bigger numbers, including claims of up to 10% in Annual Energy Production^[8].

Even the best results – and few are hitting these numbers – are not the transformational figures that were hoped for.

So, what is not working?

Two main problems hinder progress. The first is that all these promising academic models run too slowly and require too much processing power. Wind turbines must send sensor data back to a cloud, run a model, return the data, and act on it. To offer real-time optimization, they may need to do this thousands of times, across thousands of sensors and control mechanisms.

Reviews of models highlight that, to deliver at scale, they need to significantly reduce processing time, and use faster communication technologies (like 5G) between the turbine and the cloud. Such reviews also identify a need for better protection against cyberattacks.

As such, the deployment of promising models is still mostly at the Proof of Concept (POC) or small scale deployment stage.

Is data intelligence being scaled the right way in the wind energy sector?

The second problem is that wind professionals are now overwhelmed with data.

Today, around 350,000 wind turbines are installed globally^[9]. This is expected to double by 2030^[10]. Each smart turbine contains over 300 sensors, continuously transmitting over 200 gigabytes of data per day^[11]. Most of this data is boring, reporting at high frequency that everything is normal. Sending all this information back to the cloud is wasteful and expensive.

The facts are clear. Data can enable more efficient wind energy generation. The industry has a dense academic knowledge base and hundreds of promising models. But there is a real challenge when it comes to scaling and industrializing those models. At the same time, too much data is being accumulated in data centers, with very little value produced from it.