Modular deep learning approach for wind farm power forecasting and wake loss prediction
<p>Power production of offshore wind farms depends on many parameters and is significantly affected by wake losses. Due to the variability in wind power and its rapidly increasing share in the total energy mix, accurate forecasting of the power production of a wind farm becomes increasingly im...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-04-01
|
| Series: | Wind Energy Science |
| Online Access: | https://wes.copernicus.org/articles/10/779/2025/wes-10-779-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>Power production of offshore wind farms depends on many parameters and is significantly affected by wake losses. Due to the variability in wind power and its rapidly increasing share in the total energy mix, accurate forecasting of the power production of a wind farm becomes increasingly important. This paper presents a novel data-driven methodology to construct a fast and accurate wind farm power model. The deep learning model is not limited to steady-state situations but captures the influence of temporal wind dynamics and the farm power controller on the power production of the wind farm. With a multi-component pipeline, multiple weather forecasts of meteorological forecast providers are incorporated to generate farm power forecasts over multiple time horizons. Furthermore, in conjunction with a data-driven turbine power model, the wind farm model can also be used to predict the wake power losses. The proposed methodology includes a quantification of the prediction uncertainty, which is important for trading and power control applications. A key advantage of the data-driven approach is the high prediction speed compared to physics-based methods, enabling its use in applications that require forecasting multiple scenarios in real time. It is shown that the accuracy of the proposed power prediction model is better than for some baseline machine learning models. The methodology is demonstrated for two large real-world offshore wind farms located within the Belgian–Dutch wind farm cluster in the North Sea.</p> |
|---|---|
| ISSN: | 2366-7443 2366-7451 |