【摘要】 对于动车组转向架装配线,电机传动系统是其关键部分,由于其运行状态的实时监测数据具有无标签性,采用K-Means聚类等无监督式学习算法是解决这类系统健康状态评估的常用方法。针对传统K-Means算法受噪声及孤立点影响较大的缺陷,提出基于局部异常因子算法的优化方法,有效地去除数据噪声点的影响;针对中心点选取过于随机性的缺陷,提出一种基于样本密度的初始中心点选取方法,并且中心点更新是选取距簇中其他样本点方差最小的点,从而改善了聚类效果;利用实际电机传动系统的运行数据对优化的K-Means算法进行验证。实验结果表明:优化后的算法有效提高了聚类质量,很好解决了实际应用环境下的电机传动系统健康状态评估问题。更多还原

【Abstract】 The motor drive system is a key part of the assembly line of EMU bogie. Due to the label-free real-time monitoring data of its operating status, cluster analysis using unsupervised learning K-Means algorithm is one of the feasible methods to evaluate the health condition of the motor drive system. Aiming at the defect that traditional K-Means algorithm is greatly affected by noise and isolated point, an optimization method based on local abnormal factor algorithm was proposed to effectively remove the influence of data noise point. Aiming at the defect of randomness in selecting the central point, a method of initial center point selection based on the sample density was proposed, and the update of the central point is to select the point with the least variance from the other sample points in the cluster, thus improving the clustering effect. The optimized K-Means algorithm was validated by the running data of the actual motor drive system. The experimental results show that the optimized algorithm can effectively improve the clustering quality and solve the problem of motor drive system health assessment in practical application.更多还原