基于混合模型的风电机组异常数据识别方法

Identification of Abnormal Wind Turbine Data Based on Hybrid Model

结合风电机组异常数据的分布特征,采用参数模型与非参数模型结合的算法对异常数据进行识别。首先,在水平功率方向将风电机组运行数据以一定间隔分层,采用非参数模型扩散核密度估计建立不同水平功率区间内运行数据的数字概率密度曲线。然后,采用参数模型混合威布尔分布拟合概率密度曲线,利用威布尔分布的模型参数来准确描述不同水平功率区间复杂异常数据整体分布特征。最后,采用平均置信区间法识别和剔除异常数据。以2台风电机组的复杂异常数据为实例进行验证,结果表明该方法能够克服单一参数模型或非参数模型的局限性,可实现对风电机组异常数据的有效识别。

By analyzing the distribution characteristics of abnormal data of wind turbines,the algorithm combining parametric model and non-parametric model is used to identify the abnormal data.Firstly,all wind turbine operational data are placed at a certain interval in the horizontal power direction,and non-parametric model diffusion-based kernel density method is used to successively establish the digital probability density curve of operational data in different horizontal power intervals.Secondly,parametric model mixture Weibull distribution is invoked to fit probability density curves,and Weibull distribution model parameters can accurately describe the overall distribution characteristics of complex abnormal data in different horizontal power intervals.Finally,an average confidence interval method is proposed to identify and remove abnormal data.Taking two wind turbines with complex abnormal data as examples,the results show that this method can overcome the limitation of single parameter model or non-parameter model,and can effectively identify abnormal data of wind turbines.