基于门控循环网络融合多源数据的风电齿轮箱状态预警方法
STATE WARNING OF WIND TURBINE GEARBOX BASED ON GATED RECURRENT UNIT NETWORK FUSING MULTI-SOURCE DATA
-
摘要: 多源数据融合是风电齿轮箱状态预警的有效方法,针对现有机器学习方法在融合风电多源数据时未考虑时序信息导致状态预警精度不高的问题,提出基于门控循环(GRU)网络融合多源数据的风电齿轮箱状态预警方法。首先,选择对环境影响不敏感的风电齿轮箱油液压力作为状态预警模型的预测量,采用相关系数法选择与油液压力关联密切的数据采集与监控(SCADA)参数作为预警模型的输入量;然后,通过GRU的链式结构和门函数对SCADA参数的时序特征进行融合,得出预测值并计算残差;最后,根据残差的变化趋势进行风电齿轮箱状态预警。某风场运行数据状态预警结果表明所提方法的有效性。Abstract: Multi-source data fusion is an effective method for state warning of wind turbine gearbox. Aiming at the low precision problems of state warning caused by lack of consideration of time series information in existing machine learning methods,this paper proposesa a new state warning method of wind turbine gearbox based on the gated recurrent unit（GRU）network fusing multi-source data. Firstly,the oil hydraulic pressure of wind turbine gearbox which is insensitive to the environment is chosen as the predictor of the state warning model. While the supervisory control and date acquisition（SCADA）parameters closely related to the oil hydraulic pressure are selected as the input of the early warning model by the correlation coefficient method. Then,the characteristics of time series of SCADA parameters are fused by the chain structure and gate functions of GRU,the predicted value is obtained and the residual is calculated. Finally,the state of wind turbine gearbox is pre-warned according to the trend of residual. The effectiveness of state warning is proved by the wind farm operation data
下载:
