基于功率曲线分析与神经网络的风电机组故障预警方法

Wind turbine fault early warning method based on power curve analysis and neural network

为尽早发现风电机组早期故障,减少风电场的运维成本,提出一种基于功率曲线分析与神经网络的故障预警方法。首先,依据功率特性曲线,使用最小二乘与离散度分析结合的算法对SCADA数据进行筛选,以保留符合机组正常工况的数据作为建模的训练数据,从而提高模型的精度。然后,使用随机森林算法筛选模型的输入参数以简化模型结构,并通过对比隐含层的数目建立风电机组的神经网络模型。最后,结合滑动窗口模型,提出一种能反映机组实时运行状态的指标,并通过非参数估计法确定该指标的阈值,以实现状态预警及在线监测。该方法充分利用SCADA数据,且不需要对风电机组复杂的物理特性进行分析。将该方法分别用于某风电场的变桨系统和偏航系统的故障预警,实验结果是分别提前18.5 h和28.4 h出现预警信号,进一步证明方法的有效性。

In order to detect the early fault of wind turbine as soon as possible and reduce the operation and maintenance cost of wind farm,a fault early warning method based on power curve analysis and neural network is proposed.First of all,according to the power characteristic curve,the algorithm of least square and dispersion analysis is used to filter the SCADA data to retain the data in accordance with the normal working conditions of the unit as the training data for modeling,so as to improve the accuracy of the model.After that,the random forest algorithm is used to filter the input parameters of the model to simplify the model structure,and the neural network model of wind turbine is established by comparing the number of hidden layers.Finally,combined with the sliding window model,an index which can reflect the real-time running state of the unit is proposed,and the threshold of the index is determined by non-parametric estimation method to realize state early warning and on-line monitoring.This method makes full use of SCADA data and does not need to analyze the complex physical characteristics of wind turbines.The method is applied to the fault early warning of pitch control system and yaw system in a wind farm,and the experimental results show that the early warning signals appear 18.5 h and 28.4 h in advance,respectively,which further proves the effectiveness of the method.