多特征狼群优化模糊C-均值聚类感应电机无监督故障检测

Induction motor fault detection based on multi feature wolf swarm optimization FCM unsupervised clustering

感应电机是舰船系统中的重要推进动力,为提高感应电机故障检测算法的精度,提出基于多特征狼群优化模糊C-均值无监督聚类(fuzzy C-means algorithm,FCM)的感应电机故障检测方法。首先,为了去除冗余或无关信息,提出了充分考虑类间散布矩阵和类内散布矩阵之间的距离的最优特征选择方法;其次,引入狼群算法对FCM算法的聚类中心进行初始化,并通过狼群搜索优化过程对FCM算法中心进行优化,获得最佳的聚类中心设定值,实现了聚类过程的无监督执行;最后,利用MATLAB工具箱中nDexample测试集对所提狼群优化模糊C-均值无监督聚类算法的有效性进行验证,并通过对真实的感应电机故障系统的测试,显示本文算法得到的结果分类错误率等于0,显著优于前向多层神经网络算法的21%和标准FCM算法的27%,验证所提故障检测算法的性能优势。

Induction motor is an important propulsion force in ship system.To improve the accuracy of induction motor fault detection,a method based on multi-feature wolf optimized fuzzy C-mean unsupervised clustering（FCM）is proposed.Firstly,to remove redundant or irrelevant information,an optimal feature selection method is adopted.In the method,the distance between the class scatter matrix and the intra class scatter matrix is taken into account.Secondly,the wolf swarm optimization is used to initialize and optimize the center of clustering in FCM,get the best value of cluster center and implement unsupervised clustering process.Finally,the data set nDexample in MATLAB toolbox is adopted to verify the validity of unsupervised fuzzy C-mean clustering algorithm.Through the test of induction motor fault system,the results show that the proposed classification error rate is equal to 0,under 21%forward propagation multi layers neural networks and 27% standard FCM.