Spatiotemporal wind speed forecasting using conditional local convolution and multidimensional meteorology features.

Area of Science:

• Meteorology
• Renewable Energy Systems
• Artificial Intelligence

Background:

• Accurate wind speed prediction is vital for efficient wind power forecasting and operational cost reduction, especially in complex highland terrains.
• Traditional forecasting methods and existing deep learning models struggle with localized meteorological variations due to uniform influence weight assumptions.

Purpose of the Study:

• To introduce an enhanced Conditional Local Convolution Recurrent Network (CLCRN) for improved spatiotemporal wind speed forecasting.
• To address the limitations of uniform influence weight structures in deep learning models for wind speed prediction.

Main Methods:

• Developed an enhanced CLCRN model incorporating multidimensional meteorological inputs (temperature, pressure, dew point, wind components).
• Redesigned convolution kernels to capture local meteorological features and integrate multiple influencing factors, overcoming uniform weight issues.
• Validated the model using meteorological station data from 2019 to 2021 for prediction intervals of 3, 6, 9, and 12 hours.

Main Results:

• The enhanced CLCRN model consistently achieved lower Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) compared to other models across all prediction intervals.
• Spatial distribution of local convolution weights demonstrated alignment with local wind velocity patterns, enhancing model interpretability.
• The model's performance was validated using real-world meteorological data.

Conclusions:

• The enhanced CLCRN model significantly improves spatiotemporal wind speed forecasting accuracy in challenging highland regions.
• The model's ability to capture local features and its enhanced interpretability offer practical benefits for renewable energy planning and wind dynamics simulation.