Identification of Catenary Performance Degradation Based on Gath Geva Clustering and Improved Support Vector Date Description

• Published in: Arabian journal for science and engineering (2011) 2022, Vol.47 (11), p.13765-13780
• Main Authors: Yi, Lingzhi, Sun, Tao, Zhao, Jian, Xu, Xiu, Jiang, Ganlin, Liu, Jiangyong
• Format: Article
• Language: English
• Subjects: Algorithms; Catenaries; Clustering; Engineering; Humanities and Social Sciences; Hyperspheres; Identification methods; Machine learning; multidisciplinary; Particle swarm optimization; Performance degradation; Research Article-Electrical Engineering; Science
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-021-06393-x

In view of the low utilization of catenary detection data, the imperfect research on catenary performance degradation caused by the joint action of multiple parameters, and most of the data are unsupervised learning data, which is not conducive to research and analysis. A catenary performance degradation identification method based on the improved support vector data description model (ISVDD) optimized by Gath Geva (GG) clustering and particle swarm optimization (PSO) algorithm is proposed. Firstly, select the semi supervised learning data, cluster the unlabeled data with t-distributed stochastic neighbor embedding dimension reduction method and GG clustering method, reduce the original data to two-dimensional space, and get the best number of clusters. Then, a PSO optimized ISVDD catenary performance degradation model is constructed using labeled data, which contains as much normal data and as little abnormal data as possible, and the radius of the established hypersphere model is as small as possible. Finally, the distance between the clustered sample data and the spherical center of the hypersphere is calculated, and the distance and radius are compared to determine the performance degradation degree of catenary. The simulation results show that the PSO-ISVDD proposed in this paper can judge the performance degradation of catenary more accurately than SVDD and ISVDD, and the accuracy can be as high as 99.33%, which provides a powerful reference for the identification of catenary performance degradation.