基于数据挖掘方法的风力涡轮机状态监测技术研究

A Data Mining Approach for Wind Turbine State Monitoring

目前风力涡轮机的故障模式预测成为了风力发电站发展的重要任务;提出了一种基于数据挖掘算法的涡轮机故障状态预测方法;这种方法包括3个主要的步骤:涡轮机状态抽象,算法训练,状态预测;首先利用先验知识将涡轮机的初始状态进行分类,选择建立预测模型的参数;为了降低计算难度,采用数据挖掘算法进行模型参数的选择;最终采用发电机转速、变速箱速度、温度枢纽、叶片螺距角这些参数进行预测模型的建立;建立预测模型的过程分为3个阶段:预测任意故障;预测系统的特殊故障;确定未知故障;通过对各种数据挖掘算法基于大量风力涡轮机数据的性能分析,选择了性能最优的随机森林算法模型;这种模型的预测准确率能够达到98%;同时还能够预测训练数据没有包含的故障类型;通过在实际风力涡轮机数据的验证,表明了这种模型的稳健性。

As the rapid development of wind farms, it becomes important for wind turbine monitoring and maintenance. As the operating of wind turbine, the state may change from normal to fault. The prediction of fault modes is important for the maintenance of wind turbine. In this paper, we proposed a wind turbine fault modes prediction based on a data mining method. The prediction model contains three steps： prediction of random fault; prediction of special fault; prediction of unseen fault. We chose an optimal random forest algorithm as the data mining approach based on the comparative analysis on the data collected at a large wind farm. The prediction accuracy of the model can a- chieve 98~, and at the same the model can predict fault modes which are not contained in the training data. Based on the practical wind tur- bine data, the robustness of the model is verified.