Enhancing Wind Energy Management: predictive modelling and data analysis insights
Kujala, Ari-Pekka (2024)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2024053018703
https://urn.fi/URN:NBN:fi:amk-2024053018703
Tiivistelmä
Wind power output prediction was researched to optimize energy management and sales in the renewa-ble energy sector. The primary objective was to develop accurate predictive models for forecasting wind power output over the next 36 hours using machine learning techniques, benefiting turbine owner Tuuliveikot Oy by optimizing energy sales and management.
A comprehensive dataset from a wind turbine in Kauhava, including variables such as wind speed, nacelle direction, air temperature, and power output, was utilized. Data from the Norwegian Meteorological Institute (Met.no) was integrated into the predictive models. Implementation involved iterative development, including data cleaning, feature engineering, and the application of machine learning and neural network models to improve prediction accuracy.
The models developed demonstrated a mean absolute error of 300 kW for a 2,5 MW turbine, indicating that while the predictions were not perfectly accurate, they significantly reduced errors compared to simpler methods. Wind speed transformation was found to be crucial for enhancing model performance, and the models generated valuable insights despite challenges in data alignment and model complexity.
The research highlighted the complexities in wind power prediction and the importance of understanding underlying physical phenomena. The iterative process provided valuable insights that guided the refinement of methodologies. Although the models did not achieve perfect accuracy, they demonstrated the potential for significant improvements in wind power forecasting, contributing valuable tools for the renewable energy sector.
A comprehensive dataset from a wind turbine in Kauhava, including variables such as wind speed, nacelle direction, air temperature, and power output, was utilized. Data from the Norwegian Meteorological Institute (Met.no) was integrated into the predictive models. Implementation involved iterative development, including data cleaning, feature engineering, and the application of machine learning and neural network models to improve prediction accuracy.
The models developed demonstrated a mean absolute error of 300 kW for a 2,5 MW turbine, indicating that while the predictions were not perfectly accurate, they significantly reduced errors compared to simpler methods. Wind speed transformation was found to be crucial for enhancing model performance, and the models generated valuable insights despite challenges in data alignment and model complexity.
The research highlighted the complexities in wind power prediction and the importance of understanding underlying physical phenomena. The iterative process provided valuable insights that guided the refinement of methodologies. Although the models did not achieve perfect accuracy, they demonstrated the potential for significant improvements in wind power forecasting, contributing valuable tools for the renewable energy sector.