基于Bi-RNN的风电机组主轴承温度预警方法研究

Research on Temperature Prediction Method for Main Bearing of Wind Turbine Based on Bi-RNN

主轴承是风电机组能量传递的关键设备,本文以双馈风力发电机组主轴承为研究对象,首先采用高斯混合模型(gaussian mixture model,GMM)对机组工况进行辨识;其次在各个子工况空间内建立基于双向循环神经网络(bi-directional recurrent neural network,Bi-RNN)的风电机组主轴承温度模型;然后,采用随机森林算法对主轴承温度模型残差进行建模与预测,从而实现机组主轴承故障预警;最后以某大型风电场机组为对象建模并开展仿真研究.结果表明,基于工况辨识的Bi-RNN神经网络算法结合随机森林算法对主轴承故障预警具有较强的实用性和较高的准确率.

The main bearing was the key equipment for energy transmission of wind turbines. In this paper, the main bearing of doubly fed wind turbines were examined. Firstly, the working conditions of wind turbines were identified by using Gauss mixture method. Secondly, the temperature model of main bearing of wind turbines based on Bi-directional Recurrent Neural Network (Bi-RNN) was established in each sub-working condition space. Thirdly, random forest was used to establish and predict residual temperature model of main bearing, so as to realize the fault warning of main bearing of wind turbines. Finally, algorithm model were used to carry out and simulate in a large wind farm. The results showed that the Bi-RNN neural network based on conditions identification which combined with random forest algorithm had strong practicability and high accuracy for main bearing early warning.