Ultra-short-term wind power prediction method combining financial technology feature engineering and XGBoost...

Area of Science:

• Renewable Energy Systems
• Machine Learning Applications
• Time Series Analysis

Background:

• Existing wind power prediction models struggle with feature representation and slow convergence due to deep learning methods.
• These limitations hinder practical application in dynamic production environments.

Purpose of the Study:

• To develop an efficient and accurate ultra-short-term wind power prediction model.
• To address the limitations of current deep learning-based prediction methods.
• To enhance the interpretability and reduce reliance on expert experience in feature engineering.

Main Methods:

• Proposed an XGBoost model incorporating feature engineering from financial technical indicators.
• Utilized a variational ant colony algorithm for optimizing indicator parameters.
• Applied the model to the Tennet wind power dataset for validation.

Main Results:

• Achieved a mean absolute error (MAE) of 0.859 and a root mean square error (RMSE) of 1.329.
• Demonstrated rapid prediction times, completing forecasts in just 244 milliseconds.
• The novel feature engineering effectively condensed potential relationships within time series data.

Conclusions:

• The proposed model offers a significant advancement for ultra-short-term wind power forecasting.
• The integration of financial technical indicators and bionic algorithms enhances prediction accuracy and efficiency.
• This approach provides a viable alternative to traditional deep learning models in operational settings.