Abstract:
Wind power is having a significant effect on the global energy landscape, providing sustainable
choices that address environmental, economic, and social concerns. Forecasting wind
power is of the utmost importance for the successful incorporation of wind power into the
network of electrical power distribution systems. Accurate forecasts empower electric grid
operators to foresee variations in wind energy generation. In order to accurately anticipate
wind power, meteorological data, weather occurrences, wind turbine performance, and grid
limits are all taken into consideration. In this context, we have proposed a novel approach
called DBSCAN-RFE-XGBoost a two-stage process. The initial stage, we have proposed a
clustering algorithm that uses density-based spatial clustering (DBSCAN) to automate the
EPS value that is required for DBSCAN clustering by using K-dist plot and Knee Point Detection
Algorithm. Additionally, we have removed outliers from the dataset. In the second
stage, we applied two fold scheme, first we enginered temporal features and then applied
recursive feature elimination to the preprocessed dataset which identifies best suited features
to feed into the XGBoost algorithm to predict wind power. To determine effectiveness of
proposed model, we have utilized SCADA dataset obtained from a wind farm in Pakistan.
In comparison to existing benchmarking approaches, our suggested model achieves a performance
that is 38.89% higher (RMSE), and its effectiveness is demonstrated by an R2 value
of roughly 5.10%.
