The Challenge
In 2019 the Greek electricity market was heading into regulatory changes that made short-term accuracy in wind generation more valuable. Wind farm operators needed to know how much power their parks would produce hour by hour for the coming day and a half, and when weather would allow safe maintenance on the turbines.
The ENTEKA wind park asked Ex Machina to pilot both: a short-term power forecast, and a reliable local weather forecast for maintenance planning.
Our Approach
The pilot combined two Ex Machina services.
- ForecastXM is our machine-learning forecasting service. It ingests the wind farm’s own production data together with weather data, trains several competing models, and selects or blends the best-performing ones to produce an hourly power forecast for the next 36 hours. Its results are delivered back in whatever form suits the client’s systems, such as files, secure HTTP posts or direct database writes.
- WeatherXM provides the weather forecast for the specific site. Paired with a weather station installed at the wind farm, it supplies the best available local forecast for planning turbine maintenance, and adds a dashboard on our IoT platform for monitoring the data.
For weather and power forecasting, this was also the first project in which Ex Machina worked with Meteoblue as a forecast data source.
What We Learned
The pilot gave us our first direct evidence of how much a ground station matters. Forecasts built on regional model data alone left a large error at the turbine. Adding measurements from a station at the site improved the forecast substantially, and brought our error down to the level that was the industry standard at the time.
The lesson was to put ground-truth measurements at the point of interest first, and build models on top of them.
Why It Matters
Better forecasts of wind output let a wind farm operator plan with less uncertainty, and a reliable local weather forecast keeps maintenance crews from working in unsuitable conditions.