等距映射和局部线性嵌入算法集成的转子故障数据集降维方法

Method for the dimension reduction of rotor fault data sets by using ISOMAP and LLE

经数据分析途径实现机器智能的故障决策引发出了关于故障数据集的降维问题。通过将等距映射算法(Isometric Mapping,ISOMAP)、局部线性嵌入(Locally Linear Embedding,LLE)算法的优缺点进行互补,提出一种适用于非线性数据集降维的核框架下等距映射与局部线性嵌入相结合的KISOMAPLLE算法。该算法能够同时满足全局距离保持性和局部结构保持能力的数据降维基本要求。用典型的人工数据集和转子故障数据集进行的降维验证结果表明,该算法能够继承ISOMAP、LLE两种算法的各自优良性能,具有能够显著提高典型非线性数据集分类精度的性能。

The data set for fault diagnosis and decision based on machinary intelligence gives rise to the requirement of dimension reduction in data processing. The algorithms of Isometric Mapping （ ISOMAP） and Locally Linear Embedding （ LLE） were introduced simultaneously to mutually complement their strong points and weak points, and a new KISOMAPLLE algorithm was proposed. The algorithm can satisfy the requirement of both global distance preserving and local structure preserving ability, and has been used to reduce the dimension of typical artificial data sets and rotor fault data sets. The proposed algorithm inherits the excellent performances of ISOMAP and LLE, and can improve the classification accuracy of typical nonlinear data sets.