Probabilistic wind power forecasting is the most recent addition to our technological offering. We develop these forecasts tailored to our customers’ needs. In this article, we explain why it makes sense for Jungle to have built this service.
While working on a predictive maintenance project, our customer mentioned that when scheduling turbine maintenances they would estimate generation in the next few days to understand when it would have the least impact on performance. Their approach was quite simple and mostly based on visual inspection of the weather forecast for the next few days. But even so, they were noticing significant savings since they started doing this. That’s when it became clear.
If simple methods like visual inspection can create so much value, then bringing our technology to wind power forecasting can be disruptive.
We knew then that applying our technology to wind power forecasting could bring immense value to the industry. Not only is it valuable for maintenance planning by allowing asset managers to schedule maintenance on the periods with least impact on performance, but wind power forecasting is also essential for optimal energy trading, allowing the trader to both maximise revenue while minimising imbalance costs and penalties. Both of these are aligned with Jungle's objectives of minimising downtime and optimising performance.
When the opportunity came to start a wind power forecasting project with one of the largest wind asset owners in the world we couldn’t refuse. The project focused on day-ahead wind power forecasting for Italian and Romanian wind farms, where the imbalance costs were quite significant, and as such improving the quality of their current forecast could translate directly to yearly savings.
Fast forward three months and we had developed a model that was 15% more accurate than the average forecaster in their benchmark. Not only that, but we had beaten their main contracted forecaster.
In three months we were able to beat a company that has been doing it for more than 20 years. This was when we recognized the disruptive power of our technology.
One of the reasons behind this success was that we were standing on top of all the previous work developed for our normality modelling framework. The deep learning technology that powers our probabilistic modelling framework seamlessly translated to the wind power forecasting paradigm.
We use proprietary deep learning models with probabilistic distributional modelling trained on Numerical Weather Prediction (NWP) forecasts and historical asset data to forecast generation for the next hours or days.
Best of all, we can tailor all the forecasting characteristics to our customer needs:
The best part is that all we need from you is historical generation data and the location of the wind farms. You can count on us to do the rest.
You may also be wondering: There are many companies that do wind power forecasting, so why Jungle? Well, bigger companies tend to produce models in a formatted pipeline. At Jungle, we like to think that we handcraft models. We do this because we know that different customers will have different requirements. We’re agile and we know machine learning, this means that we can offer highly personalised and robust solutions that really answer your questions.
Our state-of-the-art probabilistic wind power forecasting technology can be tailored to your needs. We are looking forward to scale our wind power forecasting technology to new customers and to implement new features. In fact, our technology can just as well adapt to several other use cases such as solar power forecasting and load forecasting.
So, whether you are an asset manager, wind farm operator or energy trader, our technology can empower you to perform your work even better.