A multi-wind turbine collaborative optimization control application technology based on data-driven modeling

Aiming at the problem that hard sensor failure in wind turbines can easily lead to shutdown and power limitation, this paper proposes a multi-machine collaborative optimization control technology based on data-driven modeling. It can be mounted as an independent module on the SCADA system of the first area of the wind farm. By leveraging the existing communication and data advantages of the platform, it integrates field-level data to achieve parameter reconstruction, establishes a soft sensor data prediction model, covers nearly 70 types of hard sensors, and uses K-means clustering, random forest, GNN, etc., to meet the prediction performance in real-time. The trend prediction uses an Attention-LSTM sequence model based on time-frequency decomposition. On the PLC side of the wind turbine, a dual arbitration of physical constraints and statistical confidence is established. When the hard sensor fails, it automatically switches to the soft sensor data and optimizes the control strategy according to the predicted trend of the soft sensor. This study integrates predictive maintenance and active fault-tolerant control, providing a new idea for smart operation of wind power, which has practical reference value.