基于深度信念网络的风机主轴承状态监测方法

提出了一种基于深度信念网络（DBN）的风电机组主轴承状态监测方法。为了降低建模难度并减少训练时间，首先利用相关系数法选取建模变量，进而建立主轴承正常行为的DBN温度模型并用于主轴承温度预测。该模型克服了传统神经网络随机初始化网络权重、易陷入局部最小值等缺点，能有效提高主轴承温度的预测精度。然后采用指数加权移动平均法（EWMA）对主轴承温度残差序列进行分析，并利用核密度估计方法确定故障阈值。最后基于实测的数据采集与监视控制（SCADA）系统数据对主轴承故障进行模拟。结果表明，与传统预测方法相比，该方法能有效地实现主轴承的异常状态监测。

Condition Monitoring Method for Wind Turbine Main Bearings Based on DBN

A condition monitoring method of wind turbine main bearings was proposed based on DBN. To reduce the modeling difficulties and decrease the training time, a correlation coefficient method was firstly applied to select the modeling variables. Further, a DBN temperature model of the normal behaviors of the main bearings was established and the main bearing temperature was predicted. This model overcomes the shortcomings of traditional neural network that randomly initializes the network weights and easily to fall into the local minimum thus may effectively improve the prediction accuracy of main bearing temperature. Then exponentially weighted moving average was used to analyze the residual errors of main bearing temperature and the failure threshold was determined by kernel density estimation method. Lastly, the main bearing failure was simulated based on the measured SCADA data. The experimental results show that the proposed method may effectively realize the abnormal condition monitoring of the main bearings than that of the traditional prediction methods.