风机异常及缺失数据的填补方法研究

风机异常数据和缺失数据的识别和填补对于风机运行状态的评估和未来风速的预测具有重要意义。本文考虑到SCADA数据中某些风机可能存在异常数据和大量缺失数据的情况，首先对数据进行错误数据的识别剔除，再对缺失数据进行分类，对于个别不连续点缺失的情况进行均值填补；对于连续缺失并有旁侧风机数据参考的情况下，基于同时间段临近风机数据，先建立风向填补模型，绘制连续完整的风向数据，再分风向区间分别使用SVM方法建立风速填补模型；对于无旁侧风机参考状态下的缺失数据，使用NAR神经网络进行逐点填补。本文采用某风场实测数据进行数据验证，并与其他几种传统神经网络填补方法进行比较，测试结果表明本文提出的方法性能优于其他模型。

Research%20on%20filling%20method%20of%20abnormal%20and%20missing%20data%20of%20wind%20turbines

The%20identification%20and%20filling%20of%20wind%20turbine%20anomaly%20data%20and%20missing%20data%20is%20of%20great%20significance%20for%20the%20assessment%20of%20the%20operating%20state%20of%20the%20wind%20turbine%20and%20the%20prediction%20of%20future%20wind%20speed.%20This%20paper%20considers%20that%20some%20wind%20turbines%20in%20SCADA%20system%20may%20have%20abnormal%20data%20and%20a%20large%20amount%20of%20missing%20data.%20Firstly,%20the%20data%20is%20identified%20and%20excluded,%20and%20then%20classified.%20For%20the%20case%20of%20missing%20individual%20missing%20points,%20fill%20with%20the%20mean%20of%20adjacent%20data;%20In%20the%20case%20of%20continuous%20missing%20and%20side%20wind%20turbine%20data%20reference,%20first%20establish%20the%20wind%20direction%20filling%20model,%20draw%20continuous%20and%20complete%20wind%20direction%20data,%20and%20then%20use%20the%20SVM%20method to establish%20the%20wind%20speed%20filling%20model%20in%20each%20wind%20direction%20interval%20respectively%20based%20on%20the%20adjacent%20wind%20turbine%20data%20at%20the%20same%20time.%20For%20the%20missing%20data%20without%20the%20side%20wind%20turbine%20reference,%20the%20NAR%20neural%20network%20is%20used%20for%20point-by-point%20wind%20speed%20filling.%20In%20this%20paper,%20the%20measured%20data%20of%20a%20certain%20wind%20field%20is%20used%20for%20data%20verification,%20and%20compared%20with%20other%20traditional%20neural%20network%20filling%20methods.%20The%20test%20results%20show%20that%20the%20proposed%20method%20outperforms%20other%20models.