Power forecasting for wind farm maintenance scheduling optimisation
Today we will introduce how wind fleet owners can unlock multi-million euros cost savings with Jungle’s product offerings. Did we get your attention? Read on!
Currently, most wind portfolio owners have their wind turbines under maintenance contracts. There are different agreement scopes which range from a standard 97% uptime guarantee by the maintenance provider (usually the OEM) to energy-based availability which also come with a much heavier cost. Peace of mind comes at a cost.
On top of that 3% of yearly production, service contracts reserve a number of maintenance hours (around 80 h/year per turbine) for which the service provider is not penalised. This means that the energy production lost in these time periods is a loss taken up by the wind farm owner. Therefore, it is of the highest interest to the owner that turbine maintenance visits occur when the power production loss is minimised (disregarding emergency situations).
The most commonly used solution is to have teams at the wind fleet command centre trying to coordinate with the maintenance teams with the hopes that the maintenance interventions coincide as much as possible with low wind periods. This is, however, a very difficult task for several reasons.
It is a labour-intensive job to keep track of different wind forecasts for all turbines, even for reasonably sized wind fleets. Furthermore, this process is not free of human biases and certainly does not leverage all available data, e.g. distinct weather forecast sources, different atmospheric measurements such as temperature and pressure and historical wind turbine power production data. A lot of parameters and correlations in the data for any human to cope with.
Jungle’s approach to power forecasting
Jungle’s power production forecasting solution allows wind fleet managers to gain access to high-quality forecasts that can be used for different applications. Two of the main applications that are unlocked with good power forecasts are energy bidding and power production maximisation. Today we will be focusing on the second one.
Our approach makes use of state-of-the-art ML algorithms to ensure that the maximum prediction capability is used. It is fully automated and purely data-driven with no human biases involved in the process. Our models have been battle-tested and beat other players in this area by having lower average errors and better calibration (more details here).
It is very easy to get started since our users only need to provide historical data of the turbines power production. Jungle makes sure that the best weather forecasts are used for the specific wind farm locations.
In the following image, we see a three-day ahead power production forecast for a wind farm. We can see that our users are informed of the best possible maintenance windows for the coming days. Our users are also provided with probabilistic outputs that display the gamut of predictions made by the ML algorithms. With these well-calibrated prediction intervals, our users can mitigate the risk involved in maintenance planning.
Unprecedented detailed power forecasts
The standard approach is to predict power production at a wind farm aggregated level. However, this approach does not give insights specific enough for wind farms that span across large and challenging geographies. Wind turbines belonging to the same farm may experience very different wind conditions. This means that putting them all together in the same forecasting model does not give enough information to decision-makers. Jungle’s prediction pipeline is highly flexible and scalable allowing us to provide power production forecasts for clusters of turbines and, if needed, at turbine level.
Real-world case example
The next image shows several SCADA sensor measurements over a two-month period with three important time periods. The first one is a maintenance intervention around October 20th. The maintenance team brought the wind turbine offline to perform the maintenance activities, missing however a period of a few days with low wind speeds which led the wind turbine to not produce any power.
What’s more, this maintenance intervention caused a previously non-existing issue in the generator’s slip-ring. This was immediately detected by our normality models (please see our blog post for more information). Unknowingly, the turbine ran for 1.5 months with the slip ring at double of the normal temperature.
When the maintenance team returned at the end of November (period 3 in the image) the turbine was approximately 30 hours stopped to perform a fault find and repair. This intervention alone caused more than EUR 5 thousand revenue loss for the wind turbine owner since it was a high wind.
An excellent maintenance window presented itself in the first days of November (period 2) which could have avoided:
- power production loss since there was not enough wind to produce energy;
- the generator from running at 200% of the expected temperature for an extra month.
In this example we can clearly see the advantages that were missed by not having a holistic maintenance strategy in place:
- The first maintenance intervention would have been performed a few days earlier;
- Our normality models would have immediately identified an abnormal temperature of the slip ring as soon as the turbine started. The impact on the generator lifetime could potentially be much lower;
- The second maintenance would potentially have been undertaken during a low wind period;
- The duration of the second maintenance could have been much shorter if the maintenance team had already known where the issue was by having access to the insights of our normality models.
Wind fleet impact
The real-world case shown above is an example of the many that happen at wind fleet scale. In fact, the energy production losses can easily scale to the millions of euros. The plot below shows the average yearly revenue loss (in million EUR) due to uncovered turbine downtime. We used the European average capacity factor of 26% in 2019, the typical 3% uncovered downtime and an extra 80-hour service package which is also a service contract standard.
It is important to note that in this analysis we are disregarding the fact that penalties applied to the maintenance provider, in case the 97% uptime is not achieved, do not cover all losses incurred by the owner. Furthermore, we are also not accounting for downtime due to maintenance works undertaken at the wind farm substation which is a single point of failure that when taken offline, it disconnects the entire wind farm from the energy grid. Having a power forecasting system in place would also benefit the wind farm owners in these situations.
We can see that an average yearly loss of EUR 5 million is expected for a 1GW wind portfolio at a 60 EUR/MWh energy price. For 4GW or higher capacities, losses quickly rise to tens of millions of EUR/year zone.
Jungle can help our users reduce this impact. One way is by providing our users with power forecasting solutions which allow them to create optimised maintenance schedules. North of 10 million EUR/year for large fleets are recoverable when uninsured power production loss is reduced by 30% as shown below.
Another level of cost reduction may be unblocked with a combination of predictive monitoring and power production forecasting. Knowing if it is needed to have a maintenance intervention and when it should be performed!
Main advantages for our users
- Very low entry barriers
- Powerful state-of-the-art AI models specifically tailored for your applications
- Scalable solution to any wind turbine model and wind farm location
- Holistic approach: power forecasting and predictive maintenance
Get in touch
If you enjoyed our approach to power forecasting, feel free to reach out! Let us help you reduce uncertainty from your operations! Send us a message to hello@jungle.ai and we will be in touch shortly after! It’s that simple.
