考虑数据特征差异的风电齿轮箱群组健康状态预测研究

Health Status Prediction of Wind Turbine Gearbox Cluster Considering Data Distribution Discrepancy

随着风电装机规模的日益增长,同时风电场中现有历史故障数据不足且各风电齿轮箱之间的监测数据分布存在差异,给风电运维后市场带来了巨大挑战。为了能够将现有故障数据样本用于其他风电齿轮箱的健康状态预测以优化运维策略,提出一种基于长短时记忆网络和模糊综合的状态标定方法,以及基于迁移学习与动态加权的风电齿轮箱群组健康状态预测方法。并利用我国山西盘道梁风电场中某5台风电机组共计2年的历史监测数据对所提方法进行验证。结果表明,该组合状态标定方法能够提前检测到风电齿轮箱的故障信息,可实现单个风电齿轮箱健康状态评估。此外,剔除远离“风速-功率”曲线的样本数据后,基于深度迁移学习网络的跨设备状态预测平均准确率为92.06%,表明该方法能够将现有故障数据样本用于其他风电齿轮箱的状态监测。该研究对风电装备健康状态预测具有重要的理论价值与工程实际意义。

Wind turbine operation and maintenance are becoming more and more challenging due to the increasing installed capacity of wind turbines,the lack of existing historical fault data,and its distribution discrepancy among wind turbine gearboxes.To monitor the operational status of the wind turbine gearbox for optimizing the operation&maintenance strategies,a combined method is proposed for wind turbine gearbox cluster operational state prediction,based on the long short-term memory,fuzzy synthesis,transfer learning,and dynamic weighting function.Case applications are performed by using the monitoring data from 5 wind turbines of Pandaoliang wind farm in Shanxi,China.As a result,the operational state calibration method can sensitively detect potential fault information in advance.The average accuracy of state prediction based on deep transfer learning(TL)networks is as high as 92.06%when the sample data deviating from the"wind-power"curve are removed.It demonstrated that the proposed method can accurately narrow the data discrepancy among each wind turbine gearbox,and make full use of existing monitoring data with fault characteristics to predict the operational state of other wind turbine gearboxes.Meanwhile,it has important theoretical value and engineering practical significance for wind power equipment operational state prediction.