基于BP_Adaboost算法的风电机组叶片结冰检测

Wind turbine blade ice detection based on BP_Adaboost algorithm

精准预测冬季风电机组是否处于叶片结冰状态成为亟待解决的技术难题。Adaboost算法具有预测精度高、可使用简单弱分类器等优点。文章在传统BP神经网络算法的基础上,提出了BP_Adaboost算法精准预测风电机组是否处于叶片结冰状态。首先,将SCADA历史监测的相关信息进行重采样;其次,运用BP_Adaboost算法对叶片状态进行预测;最后,选择6台风电机组的历史数据进行实验验证。实验结果表明,由BP_Adaboost算法构建的强分类器在检测风电机组是否处于叶片结冰故障时,比BP神经网络构成的弱分类器平均得分高12%左右。BP_Adaboost算法已在部分风电场进行了实际应用。

It is an urgent technical problem to accurately predict the state of blade icing in wind turbines.Due to the advantages of Adaboost algorithm,such as high prediction accuracy and simple weak classifier,a BP_Adaboost algorithm is based on the traditional BP neural network algorithm and proposed to accurately predict the state of blade icing in wind turbines.Firstly,the relevant information of SCADA historical monitoring was resampled.Secondly,BP_Adaboost algorithm was used to predict the blade state.Finally,the historical data of 6 wind turbines are selected for experimental verification.The results show that the strong classifier built by BP_Adaboost algorithm has about 12%higher average score than the weak classifier built by BP network when detecting if the wind turbines are in blade icing failure.The research content of this paper has been applied in some wind farms.