基于LightGBM和可解释代理模型的风电场并网振荡稳定分析与控制

Oscillation stability analysis and control of grid-connected wind farms based on LightGBM and interpretable surrogate model

针对目前风电场并网振荡稳定评估的研究未考虑串补和黑箱模型无法充分被信任的问题,提出一种考虑串补特征的基于轻量级梯度提升机(light gradient boosting machine,Light GBM)和可解释代理模型的风电场并网振荡稳定在线评估和控制方法。利用模式分析法构造训练数据集,建立基于LightGBM的振荡稳定评估模型;提出基于广义加性模型的局部可解释模型的控制方法。在Simulink中搭建多风电场并网系统的仿真模型,验证了所提方法能有效地进行振荡稳定评估和抑制振荡。相比传统线性回归作为可解释代理模型,该方法能够充分拟合复杂模型,并对系统进行有效的针对性控制。

At present,the researches on the stability evaluation of wind farm grid-connected oscillation do not consider series compensation,and black box model cannot be fully trusted.Aiming at the problems,this paper proposes an online evaluation and control method for wind farm grid-connected oscillation stability based on light gradient boosting machine(LightGBM)and interpretable surrogate model,taking into account the characteristics of series compensation.The training data set is constructed by modal analysis,and a oscillation stability assessment model based on LightGBM is established.A control method of locally interpretable model is proposed based on generalized additive model.The simulation model of multi-wind farm grid-connected system is built in Simulink,which verifies that the proposed method can effectively evaluate oscillation stability and suppress oscillation.Compared with the traditional linear regression interpretable surrogate model,the proposed method can fully fit the complex model and effectively control the system.