基于混合输入模糊神经网络的风力发电机异常监测

Abnormaly Monitoring Based on Fuzzy Neural Network with Hybrid Input for Wind Turbine Generator

风力发电机工作环境恶劣致使故障频发,常规异常监测方法存在监测参数单一、误报率高等问题。将风力发电机运行工况作为分类变量引入混合输入模糊神经网络,建立风力发电机关键参数的正常行为模型,计算在线运行数据与正常行为模型的残差,通过残差建立多元高斯分布模型,并利用高斯概率密度的等高线设定异常状态阈值。实验以5 MW海上风力发电机基准模型为例,建立多输入多输出模型对发电机输出功率和转子转速进行异常监测。对比实验结果表明,该方法的发电机输出功率和转子转速异常状态识别正确率优于其他对比方法。

The poor working environment of wind turbine generators causes frequent failures. Conventional anomaly monitoring methods have problems such as single monitoring parameter and high false alarm rate. The proposed abnormaly monitoring introduces operating conditions as categorical variables into fuzzy neural network to establish the normal behavior model for monitoring key parameters of wind turbine generators. And the residuals between online operation data and the normal behavior model are employed to establish multivariate Gaussian distribution model, then the threshold of the abnormaly parameter range is set by the contour of Gaussian probability density. The experiment takes 5 MW offshore wind turbine benchmark model as an example, and establishes a multiple-input multiple-output model to monitor the abnormal condition of generator output power and rotor speed. The experimental results show that the proposed method is better than the other methods in abnormaly identification for generator output power and rotor speed.