基于卷积神经网络的风电机组齿轮箱状态监测方法

A method of gearbox condition monitoring for wind turbine based on convolutional neural network

为了实现对风电机组齿轮箱的状态监测,文章提出了一种基于卷积神经网络的风电机组齿轮箱状态监测方法。首先,提取风电机组数据采集与监视控制(SCADA)数据和振动信号作为参数,组成齿轮箱状态矩阵。其次,建立了一种卷积神经网络模型,该模型针对输入数据设计了特定结构和池化层规则,提高了计算效率,能够从齿轮箱状态信息中提取特征并判断其状态。最后,利用实际运行的风电机组数据对卷积神经网络模型进行了训练和验证,最终取得了96.3%的识别精度。同时,将该模型应用于对同一风场其他机组的状态监测,结果验证了卷积神经网络模型对齿轮箱状态监测的有效性。

In order to monitor the state of wind turbine gearbox, a method of state monitoring of wind turbine gearbox based on convolutional neural network is proposed. First, supervisory control and data acquisition(SCADA)data and vibration signals are extracted as parameters to form the matrix of gearbox state. Secondly, a convolutional neural network model is established, which designs a specific structure and pooling layer for improves computational efficiency and can extract features from the matrix of gearbox state. Then,the convolution neural network model was trained with the data of wind turbine gearbox operation, and the final identification accuracy of the gearbox state reached 96.3%. At the same time, the model is applied to the monitoring of other wind turbine gearbox, the results verify the effectiveness of the model for the condition monitoring of gear box.