Wind forecasting-based model predictive control of generator, pitch, and yaw for output stabilisation – a 15-megawatt offshore

As wind energy continuously expands its share in power generation, the grid has a higher requirement for stable wind production. This study aims for a wind forecasting-based turbine control to mitigate power fluctuation caused by wind uncertainties. Firstly, a compass-vector transformation supports a wind model on direction forecasting besides velocity. Wind modelling adopts a general network structure of learning-shaping to learn the transformed vector series. Wind speed and direction averaged from prediction determine the three-degree-of-freedom (3-DOF) reference as the control objective and update the system configuration. Subsequently, the model predictive control (MPC) solves real-time regulation by sparse quadratic programming (QP). Besides, the control loop integrates generator control, speed compensation, and output buffer to coordinate the generator, pitch servo, and yaw servo. According to the simulation, the long short-term memory (LSTM) ensures a mean accuracy of over 0.997 on a 30-s prediction window. Its performance is more stable than the dense (DNN), convolutional (CNN), and CNN-LSTM. Compared to the baseline control, the proposed MPC can reduce 7% output oscillation and 12% peak-to-peak. Wind forecasting improves rotation and power stability by 44% at high wind. The proposed turbine control is proven to contribute to better wind power quality.