A physics-guided deep learning methodology for wind farm power modeling under active yaw control

Accurate calculation of power in wind farms under various wind and yaw-controlled conditions is critical for wind farm flow control. However, due to the time-delayed nature of the flow field, current wind farm power modeling approaches rely on extra wind prediction to estimate power during practical applications, introducing increased uncertainty. Additionally, the modeling data employed lacks measurements under active yaw control, limiting model generalization. In this regard, this study develops a physics-guided deep learning model, termed the PGNN-based power model, utilizing field-measured data that encompasses measurements under active yaw control. During model construction, wind prediction is integrated to eliminate the need for additional predictive modeling in engineering applications. Furthermore, the relative positions among wind turbines (WTs) and wake velocity deficits derived from the Gaussian wake model were incorporated into the neural network, providing essential prior knowledge. To validate its effectiveness, a case study was conducted on 15 wind turbines from an offshore wind farm located in China. The results show that the proposed model effectively captures the overall power trends of WTs and wind farms under varying wind and yaw-controlled conditions, exhibiting significant engineering application value due to its independence from additional wind prediction during practical applications.