Abstract: To detect the generator abnormality of wind turbine and reduce the occurrence of unit shutdown caused by fault,a condition monitoring approach based on SCADA data analysis and One-Class Reduced Kernel Extreme Learning Machine (OC-RKELM) modeling was proposed.The features related to the generator health status were selected,and the data were cleaned by referring to the characteristics of wind turbine generator and combining with the local outlier factor algorithm.The historical normal behavior data of wind turbine were used to train the OC-RKELM,and then the constructed OC-RKELM model was used to explore the behavior of the wind turbine generator and to issue warning when the generator was wrong.Other one-class classification methods were also used to analyze the SCADA data,their performances were compared with OC-RKELM.The t-distributed stochastic neighbor embedding (t-SNE) technology was used to verify and evaluate the performance of OC-RKELM in wind turbine generator condition monitoring.Results showed that the proposed approach could detect generator abnormalities effectively and could issue warning earlier than other single classification methods.

Key words: wind turbines, condition monitoring, supervisory control and data acquisition data, one-class modelling, extreme learning machine

摘要： 为了检测风电机组发电机异常,减少由故障引起的机组停机事件的发生,提出一种基于数据采集与监控系统(SCADA)数据分析和单分类简化核极限学习机(OC-RKELM)建模的发电机状态监测方法。首先,在风电机组SCADA数据中选取与发电机健康状态相关的特征,参照风电机组的性能特性同时结合局部异常因子算法对数据进行清洗;其次,利用机组历史正常的行为数据训练OC-RKELM模型,探明机组正常工作的行为规律,进而基于该模型在线监测发电机运行状态,当其工作异常时及时实现预警;最后,与其他单分类方法进行了对比分析,并利用t-SNE可视化技术对所提方法的分析结果进行了验证与评估。结果表明:OC-RKELM具有较好的健康状态监测效果,比其他单分类方法能更早发现风电机组发电机工作异常。

关键词: 风电机组, 状态监测, SCADA数据, 单分类模型, 极限学习机