为提高网约车运输服务水平并制定合理的运营调度计划，采用聚类分析法识别不同时段和日期内网约车需求量的变化规律。针对现有的K均值（K-means）算法存在初始聚类中心随机设置的不足，提出一种动态调整惯性权重的粒子群优化K-means算法（Hybrid Particle Swarm Optimization K-means with Dynamic Adjustment of Inertial Weight, IDWPSO-K-means）来优化初始聚类中心，而后基于时段特征和日特征对网约车需求进行聚类分析，并与K-means算法及粒子群优化K均值算法（Particle Swarm Optimization K-means, PSO-K-means）进行对比分析。结果表明：IDWPSO-K-means聚类算法可有效识别不同数据模式下网约车需求量时间序列变化的相似性，基于时段特征将网约车需求量聚为2类，基于日特征将网约车需求量聚为4类;相比于PSO-K-means算法和K-means算法，IDWPSO-K-means算法的误差平方和与迭代次数这两个聚类评价指标值均更优，且IDWPSO-K-means聚类算法基于时段特征和日特征的误差平方和分别比PSO-K-means聚类算法减小了1.63%和10.93%，证明该方法可更好地识别网约车需求时变特征。

In order to improve the service level of online car-hailing and formulate a reasonable operation scheduling plan, the cluster analysis method was used to identify the variation law of online car-hailing demand in different time periods and dates. The existing k-means algorithm has the disadvantage of random setting of initial clustering centers. Aiming at this problem, a IDWPSO-K-means (Hybrid Particle Swarm Optimization K-means with Dynamic Adjustment of Inertial Weight) method was proposed to optimize the initial clustering center. The online car-hailing demand was clustered and analyzed based on time period characteristics and daily characteristics, then compared with K-means and Particle Swarm Optimization K-means (PSO-K-means) algorithm respectively. The results showed that the IDWPSO-K-means algorithm could effectively identify the similarity of time series changes of online car-hailing demand under different data modes. The online car-hailing demand was clustered into two categories based on time period characteristics and into four categories based on daily characteristics. Compared with PSO-K-means algorithm and K-means algorithm, the values of two cluster evaluation indexes which were sum of squares due to errors and iteration times of IDWPSO-K-means algorithm were better, and the sum of squared due to errors of IDWPSO-K-means clustering algorithm were 1.63% and 10.93% less than those of PSO-K-means clustering algorithm in time period characteristics and daily characteristics. It shows that the algorithm can better identify the time-varying characteristics of online car-hailing demand.