基于极端随机森林的大型风电机组发电机故障检测

Fault Detection Based on Extreme Random Forest for Large Wind Turbine Generators

针对风电机组海量运行数据中故障检测率低和实时性差的问题,提出基于极端随机森林的大型风电机组发电机故障检测方法。该方法先利用Pearson相关性分析剔除线性相关性极弱的变量和非主要特征中的冗余变量,降低样本维度。利用最大信息系数获取主要特征参数的相关系数,消除冗余变量,从而提高计算效率和故障检测精度。将基于极端随机森林的分类方法用于大型双馈风力发电机的故障检测。实验结果表明,与经典随机森林方法相比,在风电机组发电机海量数据集上,该方法具有更低的漏报率、误报率和更好的实时性。

For the problem of low fault detection rate and poor real-time performance in the massive operation data of wind turbines, a fault detection method using extreme random forest for large wind turbine generators is proposed. The method first uses Pearson correlation analysis to eliminate weakly related variable and redundant variable in non-primary feature, and the dimensions of samples are reduced. Then the maximum information coefficient is used to obtain the correlation coefficient of the main feature parameters, and the redundant feature is eliminated, thereby improving the calculation efficiency and the fault detection accuracy. A classification method using extremely random forests is used for fault detection of large doubly-fed wind turbine. The experimental results show that compared with the classical random forest method, the method has lower false negative rate and false positive rate and better real-time performance on the massive data set of wind turbine generator.