Equivalent method for DFIG wind farms based on modified LightGBM considering voltage deep drop faults
Xuecheng Liu, Peixiao Fan, Jun Yang, Ke Song, Binyu Ma, Yangzhou Pei, Jian Xu
Abstract
• Construction of complex operational scenarios considering partial trip-off of wind turbines. • Establishment of a wind turbine operational states identification model based on the mLightGBM. • Proposal of a two-stage clustering method driven by a data-model hybrid approach. To address the challenges of inadequate accuracy in identifying the Crowbar action state and incomplete consideration of operating scenarios in existing methods for Doubly Fed Induction Generator (DFIG) wind farms, a DFIG wind farm equivalent method based on modified Light Gradient Boosting Machine (mLightGBM) considering voltage deep drop faults is proposed. First, the low voltage ride through process of wind turbines is analysed, with particular consideration given to the scenario of partial wind turbines tripping off due to voltage deep drop faults in the wind farm. The factors influencing the Crowbar action state and trip-off state of wind turbines are identified, and a feature vector for wind turbine operating states is constructed. Second, based on simulations to obtain sample data of wind turbine operating states under different operating scenarios in wind farms, a classification model based on mLightGBM is established. Different weights are assigned to various samples and hyperparameter optimization is conducted to enhance the model’s classification accuracy. Finally, a two-stage clustering method driven by a data-model hybrid approach is proposed. Under specific operating conditions, wind turbine clusters are divided sequentially into two stages based on the mLightGBM classification results and the wind speed range of the turbines. The final equivalent model is derived through parameter calculations. Simulation results demonstrate that the proposed DFIG wind farms equivalent model, compared to traditional methods, not only can adapt to a broader range of operating scenarios but also achieves superior identification of wind turbine operating states, making the equivalent method rational and effective.