# Wind power predictions from nowcasts to 4-hour forecasts: a learning approach with variable selection

**Authors:** Dimitri Bouche, Rémi Flamary, Florence d'Alché-Buc, Riwal Plougonven, Marianne Clausel, Jordi Badosa, Philippe Drobinski  
**arXiv:** [2204.09362](https://arxiv.org/abs/2204.09362)  
**Date:** 2022-04-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2204.09362v2/

## Key points

- The study focuses on short-term prediction of wind speed and wind power within a 4-hour time frame with 10-minute intervals.
- Accurate forecasts for wind speed and wind power are important in mitigating negative impacts of wind farms' intermittent production on energy systems and markets.
- Researchers use a machine learning approach that combines outputs from numerical weather prediction models with local observations to enhance accuracy.
- Well-known methods that can handle large volumes of data are used to ensure practicality for practitioners.
- Variable selection techniques using both linear and nonlinear approaches are explored to identify relevant variables for accurate predictions.
- Linear models versus nonlinear models are compared in forecasting wind speed and wind power.
- Two approaches for predicting wind power, indirect (using a power curve) and direct (without intermediate calculations), are evaluated.
- The research aims to improve short-term predictions by leveraging machine learning techniques and combining information from numerical weather prediction models with local observations.

## Summary

The study focuses on the short-term prediction of wind speed and wind power, specifically within a 4-hour time frame with 10-minute intervals. Accurate forecasts for these variables are essential in mitigating the negative impacts of wind farms' intermittent production on energy systems and markets. To achieve this, the researchers employ a machine learning approach that combines outputs from numerical weather prediction models with local observations. By integrating information from numerical weather prediction models, which provide insights into higher-scale dynamics, and local observations, which offer fresher and location-specific data, the researchers aim to enhance the accuracy of their predictions. The study emphasizes using well-known methods that can effectively handle large volumes of data to ensure practicality for practitioners. The research initially explores variable selection techniques using both linear and nonlinear approaches. This analysis helps identify the most relevant variables for predicting wind speed and wind power accurately. Building upon these findings, the researchers then compare the performance of linear models versus nonlinear models in forecasting wind speed and wind power. In addition to comparing different modeling techniques, the study also evaluates two approaches for predicting wind power specifically: indirect and direct methods. The indirect approach involves passing wind speed predictions through a power curve to estimate wind power output. On the other hand, the direct approach directly predicts wind power without relying on intermediate calculations. Overall, this research aims to improve short-term predictions of wind speed and wind power by leveraging machine learning techniques and combining information from numerical weather prediction models with local observations. By refining variable selection methods and comparing linear versus nonlinear modeling approaches, this study provides valuable insights into enhancing forecast accuracy in renewable energy systems.