基于特征选择和XGBoost的风机叶片结冰预测

Wind Turbine Blade Icing Forecast Based on Feature Selection and XGBoost

风电场风机叶片积冰会造成风电机组的效率降低,严重时会导致叶片断裂,严重威胁风电场的正常生产运行。提出了一种基于Relief的特征选择和XGBoost的风机叶片结冰预测方法,能够根据风机运行的SCADA数据对叶片结冰的早期过程进行精确预测,并采用某风电场的2台风机数据进行模型验证和对比试验,取得了较好的预测精度,提升了预测速度,能够有效预测早期叶片结冰故障的发生,从而为降低风机的效率损失和风机的运行风险提供数据支撑。

Icing of wind turbine blades in a wind power plant may reduce the efficiency of the wind turbine set and even cause blade breakage in severe cases,thus threatening normal production and operation of the plant.This paper proposed a method for turbine blade icing forecast,based on Relief feature selection and XGBoost,which could complete exact forecasting of the early process of blade icing according to the SCADA data of wind turbine operation.Furthermore,data on 2 sets of turbines in a certain wind power plant was adopted to make model verification and contrast experiment with good forecasting accuracy and higher forecasting speed.The proposed approach could effectively predict the occurrence of early blade icing,so as to provide data support for reduction of turbine efficiency loss as well as its operational risk.