Accurate wind power predictions are vital for the efficient and safe operation of power systems, especially with a high share of wind generation. Additionally, precise forecasts reduce the need for costly balancing energy from reserve markets, ensuring higher profitability and overall value for wind power producers.
This work presents a new method for improving the accuracy of wind power forecasting. The key aspect of this method is the optimal combination of a large number of meteorological parameters using statistical and machine learning approaches, such as dimensionality reduction and feature selection algorithms, prior to applying several regression algorithms calibrated for different weather regimes. Technical metrics, such as normalised root mean square error (NRMSE) are discussed in this work.
The methodology is applied individually to forty wind power plants in Portugal, as well as for the aggregated wind power in Germany. A traditional approach from a forecast provider was used as the benchmark.
The proposed approach shows a performance improvement when compared to the benchmark (on average, the NRMSE reduced 5.5% for the forty wind power plants). Results demonstrate the importance of selecting the most relevant meteorological features for each power plant or aggregated country to maximize the accuracy of the power forecast.
Within this enlightening white paper, you'll notice that the paper frequently references Smartwatt Intelligence rather than Enlitia. This is because, at the time of publication, Enlitia hadn't officially launched (through a spin-off from Smartwatt Intelligence).
Nevertheless, many contributors to this paper have been instrumental in shaping Enlitia from day one. So, while the paper may bear the Smartwatt banner, it embodies the knowledge and innovation that Enlitia brings to the renewable energy sector. It's a testament to our constant growth and evolution in this ever-changing industry.