基于WGAN-GP的风电机组传动链故障诊断

传动链负责将风电机组叶轮的能量传递至发电机,若传动链中的任一部件,如齿轮、轴承发生异常,风电机组将面临巨大的安全隐患.现有基于深度学习的风电机组故障诊断大多需要人为选择目标变量,所识别故障与所选变量关联性大、通用性不足.梯度惩罚Wasserstein生成对抗网络(WGAN-GP)采用Wasserstein距离作为量度生成数据与真实数据的代价函数,具有训练结果稳定的优势.文中基于数据采集与监控(SCADA)系统提出两步数据预处理方法进行数据筛选,并基于WGAN-GP设计风电机组传动链异常状态分数,进而识别传动链故障.所提方法运用通用SCADA参数,无须人为挑选目标变量,可稳定识别风电机组传动链中的非特定故障,具有识别结果准确、泛化能力强等优点.9台双馈风电机组的状态识别结果验证了所提方法的有效性,可以辅助指导风电场的运行维护.

Fault Diagnosis of Wind Turbine Drivetrain Based on Wasserstein Generative Adversarial Network-Gradient Penalty

The drivetrain is responsible for the energy transfer from rotor hub to generator in wind turbines. If any part of the drivetrain, such as gears and bearings, is abnormal, the wind turbine will face a huge safety hazard. Now most of the current wind turbine fault diagnosis based on the deep learning need to select target parameters artificially, and the identified fault has a close correlation with the selected variables, resulting in insufficient versatility. Wasserstein generative adversarial network-gradient penalty (WGAN-GP) uses Wasserstein distance between the generated data and the real data as a measurement for the cost function, which has the advantage of stable training results. This paper proposes a two-step data preprocessing method for data screening based on the supervisory control and data acquisition (SCADA) system, and designs anomaly state score of the wind turbine drivetrain based on the WGAN-GP model to identify the drivetrain faults. The proposed method uses common SCADA parameters, does not need to manually select target variables, and can stably identify non-specific faults in the wind turbine drivetrain, which has the advantages of accurate identification results and strong generalization ability. The status identification results of nine doubly-fed wind turbines verify the effectiveness of the proposed method, which can assist in guiding the operation and maintenance of wind farms.