结合C-均值聚类的自适应共振神经网络在风电机组齿轮箱故障诊断中的应用

Application of ART2 Neural Network Combined with C-means Clustering in Fault Diagnosis of Wind Turbine Gearbox

提出了一种结合C-均值聚类的自适应共振(Adaptive Resonance Theory 2,ART2)神经网络无监督学习分类方法,用于风电机组齿轮箱设备群的故障诊断.利用某风电场齿轮箱运行数据,采用ART2神经网络对样本数据进行初步分类,再采用C-均值聚类算法对神经网络分类结果进行修正,得到最终诊断结果,并与ART2神经网络分类结果进行了比较.结果表明:所提出的方法解决了原始神经网络算法存在"硬竞争"导致分类精度下降的问题,准确度高于传统的ART2神经网络,可以准确识别出故障齿轮箱.

An unsupervised learning classification method by using adaptive resonance theory 2 （ART2） neural network combined with C-means clustering was proposed for fault diagnosis of wind turbine gearbox group. For operation data of gearboxes in a wind farm, ART2 neural network was used to preliminarily classify the samples, and subsequently C-means clustering algorithm was used to modify the classification results of ART2 neural network, and the final classification results were compared with those of ART2 neural network. Results show that the proposed method has higher accuracy than traditional ART2 neural network, since it solves the problem of low classification accuracy caused by hard competition of ART2, which therefore can identify the faulted gearbox accurately.