K均值聚类改进与行驶工况构建研究

Research on Improved K-Means and Driving Cycle Construction

采集了广州市2 800万个样本数据,采用相关指标比较分析K均值聚类、K中心点聚类、模糊聚类、高斯混合聚类4种方法,并以90%置信区间作为初始中心选取范围提高K均值聚类稳定性。运用改进后的K均值聚类构建广州市行驶工况,平均相对误差小于6%,并与美国、欧洲、日本、中国等地区的典型行驶工况进行比较。结果表明,广州市行驶工况具有车辆加减速频繁、怠速与低速段工况占比高的特点,与国内现行NEDC以及中国QC/T 759-2006工况存在一定差异。

With 28 million sample data of Guangzhou, K-means, K-medoids, FCM and GMM are compared and analyzed by related index, 90% confidence interval is chosen as the initial centre selection range and to improve the stability of K-means. The improved K-means method is used to construct the Guangzhou driving cycle, average relative error is less than 6%, Guangzhou cycle is compared with the typical driving cycles in the U.S., EU, Japan and China, etc.The results show that Guangzhou driving cycle is characterized with frequent acceleration and deceleration, high proportion of idle speed and low speed section, which is different from the current domestic driving cycle NEDC and QC/T 759-2006 cycle.