Multi-task prediction of wind speed and time-varying wind shear coefficient using dynamic graph interactive neural network
Wind power is the main form of wind energy utilization and has developed rapidly in recent years. The power of a wind turbine is closely related to the cube of the wind speed. Current research usually relies on wind speed at the hub height of wind turbines or equivalent wind speed for wind assessment. Therefore, the vertical wind profile within a wind farm is crucial as it demonstrates the variation of wind speed and direction with height.
Generally, wind towers at the same height as the wind turbine hub are required to collect corresponding data. However, the installation and maintenance of these tall wind towers are costly, especially in offshore wind farms. A more feasible strategy is to build a high wind tower and supplement it with several low auxiliary towers, as shown in Figure 1. In this way, the wind speed of the vertical wind profile is extrapolated to the required height by using the wind shear model and low-altitude wind speed.
In current research, power law or logarithmic law is usually employed to characterize wind shear phenomena and models. The power law often adopts a fixed empirical value of 1/7 or a specific constant. However, the wind shear coefficient is not static but varies dynamically with low-altitude wind speed and different meteorological conditions, such as temperature gradient, time of day, and atmospheric stability. Although there are many studies on wind shear phenomena, there is a significant shortage of research specifically on the dynamic tracking and prediction of wind shear coefficients.
To address this issue, especially the scarcity of research on dynamic wind shear coefficient prediction, the research team innovatively constructed a multi-task learning method based on graph neural networks, as shown in Figure 2. This method simultaneously predicts wind speed and wind shear coefficient, thereby overcoming the coupling challenge between wind shear phenomenon and wind speed.
The accuracy of the proposed method was verified through the actual wind speed and wind shear coefficient of the offshore wind farm, demonstrating its potential in engineering applications. Some of the comparative verification results are shown in Figure 3.
The relevant research results are titled “Multi-task prediction of wind speed and time-varying wind shear coefficient using dynamic graph interactive neural network”. Published in Information Fusion on July 5, 2025.
