基于批标准化的堆叠自编码网络风电机组变桨系统故障诊断

FAULT DIAGNOSIS OF PITCH SYSTEM OF WIND TURBINE BASED ON STANDARDIZED STACKED AUTOENCODER NETWORK

为了提高风电机组变桨系统故障诊断的准确性,提出一种基于批标准化的堆叠自编码(SAE)网络故障诊断模型。针对SAE网络在特征学习过程出现的梯度硬饱和问题,选用PReLU激活函数,在SAE网络中加入批标准化(BN)层进行优化,通过输出层的Softmax函数,得到变桨系统各部件故障发生概率。以均方误差最小化为目标,采用Adam算法迭代训练数据,使模型参数得到更新。在风电机组变桨系统数据采集与监视控制(SCADA)系统中的数据集中,对优化前后的SAE网络通过改变迭代次数、样本数量进行实验,结果表明,优化后的SAE网络模型具有更好的识别精度;另外,在不同样本数量的实验中,与其他传统模型相比,优化后的SAE网络模型故障识别率也更高,表明其在风电机组故障诊断领域有一定的应用价值。

In order to improve the accuracy of fault diagnosis of wind turbine pitch system, a fault diagnosis model based on batch normalization of stacked auto-encode(SAE) network is proposed. Aiming at the problem of hard gradient saturation in the feature learning process of the SAE network,the PReLU activation function is selected,and the batch normalization(BN) layer is added to the SAE network for optimization. Through the Softmax function of the output layer,the failure probability of each component of the pitch system is obtained. With the goal of minimizing the mean square error,the Adam algorithm is used to iterate the training data to update the model parameters. In the data set of wind turbine pitch system supervisory control and data acquisition(SCADA) system,the SAE network before and after optimization is tested by changing the number of iterations and the number of samples. The results show that the optimized SAE network model has better recognition accuracy. In addition, in the experiments with different sample numbers,compared with other traditional models, the fault recognition rate of the optimized SAE network model is also higher,indicating that it has certain application value in the field of wind turbine fault diagnosis.