改进模糊C均值聚类法的车辆实际行驶工况构建  

Vehicle Real Driving Cycle Construction Based on Improved Fuzzy C-means Clustering Algorithm

作　　者：高建平[1] 高小杰[1] 

机构地区：[1]河南科技大学车辆与交通工程学院,河南洛阳471003

出　　处：《河南科技大学学报（自然科学版）》2017年第6期21-27,共7页Journal of Henan University of Science And Technology:Natural Science

基　　金：河南省科技攻关计划基金项目(152102210073);河南省高等学校青年骨干教师计划基金项目(2015GGJS-046);河南科技大学第六届研究生创新基金项目(CXJJ-2016-ZR01)

摘　　要：针对模糊C均值(FCM)聚类法的性能依赖于初始聚类中心、迭代容易陷入局部极值、不能确保FCM收敛于一个最优解的问题,利用多岛遗传算法(MIGA)与序列二次规划法(SQP)组合优化,对FCM聚类的初始聚类中心进行优化,从而使聚类结果更加接近最优聚类。采用主成分分析和改进的FCM聚类分析,将运动学片段的特征值进行降维和分类处理,构建出基于大样本、符合郑州市交通特征的行驶工况。与试验数据对比表明:所构建的乘用车行驶工况与试验数据特征参数平均相对误差仅为2.097%,速度-加速度联合分布差异(SAFD_(diff))仅为1.74%,行驶工况拟合精度较高,更能综合反映郑州市交通真实状况。Since the performance of Fuzzy C-means( FCM) clustering algorithm depended on the initial clustering center,the iterative was inclined to fall into the local extremum and could not ensure that FCM converges to an optimal solution. Therefore,the initial cluster center of FCM was determined by optimization algorithm,and combination of multi-island genetic algorithm( MIGA) and sequential quadratic programming( SQP) method in order to get closer to the global optimal clustering. Dimension reduction treatment was performed on the characteristic parameters of kinematics sequences by principal component analysis and improved FCM clustering technique. Thus,a driving cycle based on large sample and traffic characteristics of Zhengzhou was constructed. The driving cycle was compared with the measured data. The relative average error of characteristic parameters between the synthetic cycle of passenger cars and measured data is 2. 097%,and the joint difference of velocity-acceleration( SAFD_(diff)) is only 1.74%. The clustering technique has higher accuracy in driving cycle fitting and can more comprehensively reflect the real situation of Zhengzhou.

关 键 词：FCM聚类 聚类中心 主成分分析 行驶工况 

分 类 号：U46[机械工程—车辆工程]