Abstract:

During the rapid frequency regulation process of wind turbine units, the transient active power release can induce load fluctuations in aerodynamic, transmission, and tower components. In order to reasonably characterize the fluctuation characteristics and serve the optimization of frequency regulation control, this paper presents a fast dynamic modeling method for wind turbine units that takes into account the coupling characteristics of blades, main shaft, generator, and control systems. Firstly, a wind farm-turbine coordinated primary frequency regulation control strategy is set up, and a rapid frequency regulation controller at the unit level is developed for both below and above the rated wind speed based on a refined 5MW wind turbine model. And then, the Spilman correlation analysis algorithm is used to select input and output variables with consideration of input and output delay orders, and the operational domain partitioning is completed, enabling adaptive identification and switching between operation regions both above and below the rated wind speed. Thirdly, based on balanced sampling of simulation operating data under discrete operating conditions, and guided by physical prior information, subspace identification and deep neural network algorithms are employed to conduct multi-input-multi-output modeling and simulation verification of the unit's primary frequency modulation dynamics across the full range of operating conditions. The results show that the state space model obtained has good interpretability, but the model structure inherently limits its approximation accuracy to a finite degree; in comparison, the temporal neural network model demonstrates a better ability to capture dynamic characteristics, providing a robust model foundation for subsequent optimization control of the unit's primary frequency modulation.

Key words: wind turbine unit, primary frequency regulation, load dynamics, LSTM neural network, subspace identification