A new intelligent prediction method for low-altitude short-term wind fields is reported

Recently, Professor Lin Lianlei’s research team from the School of Electronics and Information Engineering has made significant progress in the field of intelligent wind speed prediction. Innovatively constructing a Multivariate Meteorological Data Fusion Wind Forecast Network (MFWPN), the team has achieved high-precision intelligent prediction of low-altitude short-term wind fields. Relevant research results were published in Nature Communications under the title A machine learning model for hub-height short-term wind speed prediction. This achievement can support the development of smart wind power and guarantee the flight safety of low-altitude aircraft.

As a clean energy source, wind energy plays a crucial role in the future energy structure. However, the randomness of wind power and its inability to be stored lead to uncertainties in wind power supply. In addition, in the low-altitude economy sector, strong convection and complex meteorological conditions in low-altitude airspace pose a serious threat to the flight safety of low-altitude aircraft. Therefore, accurate short-term wind speed prediction (WSP) is essential for the dispatching, operation and maintenance of wind power generation, as well as ensuring the safety of low-altitude aircraft.

Schematic Diagram of Wind Speed Prediction

To address the above issues, Professor Lin Lianlei’s team proposed a new method for fine-grid vector wind speed prediction based on the Multivariate Meteorological Data Fusion Wind Forecast Network (MFWPN). This study uses a CNN-Transformer architecture to extract spatiotemporal features of wind speed and a spatiotemporal fusion module to integrate multivariate meteorological information. Results show that within the first 6 hours, MFWPN outperforms the European Centre for Medium-Range Weather Forecasts (ECMWF) model in vector wind speed prediction accuracy. Transfer experiments demonstrate that MFWPN has excellent generalization performance and can be rapidly applied to wind speed prediction in different regions. Efficiency experiments indicate that MFWPN only requires 18 milliseconds to predict the 24-hour fine-grid vector wind speed in Northeast China. With its accuracy and efficiency, MFWPN can serve as an effective tool for vector wind speed prediction, facilitating ultra-short-term and short-term deployment planning of wind power centers and providing strong support for path planning of low-altitude aircraft.

Rose Diagram of Predicted Wind Fields