The Importance of the Long-term Reference Curve

When we explain how turbine performance trends are tracked with our i4SEE Performance™ application it sounds easy. We compare the long-term reference curve with the latest power curve and then we calculate an energy change KPI based on any differences. It’s as simple as that! We can us this result to prioritize turbines with a significant negative change in performance, drilling down further into the data to accurately spot changes in performance, indicate root causes and help our customers to fix their turbines to increase their overall AEP.

But spoiler alert: this is not the end of the story… it’s only the beginning. Calculation of the long-term reference curve is a highly complicated process, where all algorithms need to seamlessly work together. We often discuss the diverse challenges facing the wind industry. These range from understanding new turbine technologies, through adapting to intricate operational strategies demanded by site layouts or permit stipulations, to dealing with intermittent or enduring grid restrictions. Additionally, the industry faces the influence of market forces that shape operational decisions. Collectively, these factors can combine to produce a complex, and potentially toxic cocktail, that could jeopardize the efficacy of your analytics processes.

As always, when dealing with toxic substances, it is advisable to have the right antidote at hand. With our different filtering techniques, we can address and classify every main source of underperformance, removing affected datapoints before calculating the long-term reference curve. This advanced filtering is a required step, but still not all that is required to produce a stable reference curve. There are seasonal effects to be considered, and local flow effects such as wake or turbulence intensity can also have a tremendous impact on your power curves.

Calculating a valid long-term reference curve requires a historical dataset which includes all conditions affecting the turbine One of the reasons that we run our analytics processes on a monthly basis, is to ensure that we have enough data to calculate stable power curves. However, in order to ensure all proper consideration of all long-term effects, the reference curve is continuously optimized, as our algorithms continuously learn and improve our understanding of the normal behavior of the turbine.

It is crucial to ensure seamless integration of generated insights into service activities. This will allow for swift and efficient action based on analytical insights, promptly addressing and resolving underlying issues. In our experience, performing monthly analytics processes provides the idea compromise between achieving large enough datasets for accurate results, delivering results at a rate that can be managed by service teams, and ensuring response times are adequate.

So to summarize our formula for tracking long term performance trends, here is the magic formula: gather sufficient data, perform air density correction, perform turbulence intensity correction, filter data to ensure that different operational states are correctly understood, perform binning of corrected and filtered data points, adjust to consider any new learned and normal behavior, and finally derive a valid long term reference curve.

But we’re not quite there yet…

How do we make sure that our algorithms don’t “learn” temporary underperformance as “normal”?

How do we make sure that we consider every wind turbine as a unique system, without forgetting that there are certain variations in performance that can be expected from a certain turbine type?

All of these questions are considered while creating the long-term reference curve in our performance application. But why go through all that trouble?

Because the long-term reference curve is the key to monitoring your wind turbine’s performance. It allow us to precisely see when there is a negative development in efficiency and it clearly Informs us about any changes to the turbine control or health status.

To put it in simple words, it will tell us when something is wrong.

Others call it a “digital twin”. We call it long-term reference curve and our approach, driven by domain expertise, is the essence of i4SEE Augmented Intelligence™.