基于PKFF-Transformer的风力发电机故障预测  

Wind Turbine Fault Prediction Based on PKFF-Transformer

作　　者：杨伟新 赵洪山 张扬帆 张一波 林诗雨 YANG Weixin;ZHAO Hongshan;ZHANG Yangfan;ZHANG Yibo;LIN Shiyu(State Grid Jibei Electric Power Research Institute,Beijing 100045,China;School of Electrical and Electronic Engineering,North China Electric Power University(Baoding),Baoding Hebei 071000,China)

机构地区：[1]国网冀北电力有限公司电力科学研究院,北京100045 [2]华北电力大学(保定)电气与电子工程学院,河北保定071000

出　　处：《机床与液压》2025年第4期221-229,共9页Machine Tool & Hydraulics

基　　金：国网冀北电力有限公司电力科学研究院科学技术项目(52018K22001P)。

摘　　要：为了准确预测风电机组故障,提出一种基于PKFF-Transformer风力发电机故障预测模型。针对风电数据高维复杂特性,提出基于皮尔逊核特征融合(PKFF)的特征工程法;通过皮尔逊相关系数(PCC)筛选与机组状态强相关的特征,再采用核主成分分析(KPCA)对筛选数据进行非线性特征融合;将健康状态下的融合特征输入到Transformer模型中构建风电机组温度预测模型;采用滑动窗口法统计预测残差动态特性并确定故障预警阈值;最后,将风电机组实时运行数据输入训练好的PKFF-Transformer模型进行故障预测。采用我国北方某风电场风力发电机数据进行验证。结果表明:PKFF-Transformer模型能够提前5.6 h预测到故障,且在机组健康状态下没有误报现象;此外PKFF-Transformer温度预测模型的均方误差也比Transformer模型提高了97.39%。In order to accurately predict wind turbine faults,a wind turbine fault prediction model was proposed based on PKFF-Transformer.For the high-dimensional complexity of wind power data,a feature engineering method was proposed based on Pearson-Kernel feature fusion(PKFF);the features strongly correlated with the state of the turbine were screened through the Pearson correlation coefficient(PCC)and then the kernel principal components analysis(KPCA)was used to fuse the nonlinear features of the screened data.Then the fused features under the healthy state were input into the Transformer model to construct the wind turbine temperature prediction model;and the sliding-window method was used to predict the residual dynamics and determine the threshold of the fault early warning.Finally,the real-time operation data of the wind turbine was input into the trained PKFF-Transformer model for fault prediction.The wind turbine data from a wind farm in the north of China was used for validation.The results show that the PKFF-Transformer model is able to predict the faults 5.6 h in advance,and there is no false alarm phenomenon under the healthy state of the turbine;in addition,the mean squared error of the PKFF-Transformer temperature prediction model is also improved by 97.39%compared with that of the Transformer model.

关 键 词：风力发电机 Transformer模型 核主成分分析(KPCA) 故障预测 皮尔逊相关系数 

分 类 号：TM315[电气工程—电机]