Short-term wind farm cluster power point-interval prediction based on graph spatio-temporal features and S-Stacking...

Area of Science:

• Renewable Energy Systems
• Power System Operations
• Computational Intelligence

Background:

• Wind energy integration requires precise forecasting to minimize operational costs and maximize consumption.
• Existing wind power forecasting models often overlook spatial correlations within wind farm clusters and objective sub-learner selection, increasing operational risks.
• Interval forecasting for wind power clusters is underdeveloped, hindering reliable grid management.

Purpose of the Study:

• To develop an advanced forecasting method for wind power clusters that enhances accuracy and reliability.
• To address the limitations of existing models by incorporating spatial characteristics and objective sub-learner selection.
• To provide a robust tool for power generation planning in wind farm clusters.

Main Methods:

• A novel "decomposition-aggregation-multi-model parallel prediction" approach is proposed.
• Data preprocessing involves decomposition-aggregation strategy and spatial feature extraction.
• A Stacking model with bootstrap-selected parallel sub-learners is employed for point and interval forecasting.

Main Results:

• The proposed method demonstrated superior accuracy and reliability in both point and interval wind power forecasting compared to existing models.
• Achieved a root mean square error (RMSE) of 7.47 and an average F value of 1.572.
• Experimental validation was performed using 15-min resolution wind power data from a wind farm cluster in northwest China.

Conclusions:

• The developed method offers a significant improvement for wind power forecasting in clustered environments.
• The approach effectively captures spatial characteristics and optimizes sub-learner selection for enhanced prediction.
• Results provide a reliable reference for optimizing power generation planning and grid integration of wind energy.