为解决公交线网布局的动态性与传统充电站选址静态性之间的矛盾，以影响线网布局与运力配置的关键因素——客流起讫点（OD）为切入点，提出一种基于时空客流分区的公交充电站选址方法。首先，利用公交刷卡数据识别市域公交出行连接网络，构建面向公交充电站服务分区的社区发现模型。其次，在完成服务范围分区的基础上，建立以充电需求为权重的选址模型，结合各分区内充电需求的聚集特征确定站址，最终形成公交复杂网络的充电站服务分区选址方案。最后，以重庆市中心城区2 900万条公交刷卡数据为样本进行实例验证，通过社区发现模型将城市划分为13个公交充电站分区，各分区内采用基于充电需求权重的重心法完成公交充电站选址，并将结果与AP聚类方法及P-中位选址方法进行对比。结果表明，该方法能有效兼顾大时间尺度下公交车的充电需求。考察基于惩罚系数修正后的平均充电距离综合指标，本方法较AP聚类方法降低14.51%，较P-中位选址方法降低18.33%，对应的充电调度距离成本更低，在宏观服务分区与候选站点识别方面具有良好的应用潜力。

To address the contradiction between the dynamic nature of bus network layouts and the static nature of conventional bus charging station siting, this paper takes the passenger flow OD (origin-destination), a key factor influencing network layout and fleet allocation, as the entry point, and proposes a bus charging station location method based on spatiotemporal passenger flow partitioning. Firstly, bus smart card data are used to identify the urban bus travel connection network, and a community detection model is constructed for bus charging station service partitions. Secondly, based on the delineated service partitions, a location model weighted by charging demand is established, and the aggregation characteristics of charging demand within each partition are used to determine station locations, thereby generating a service-partition-based location scheme for bus charging stations within a complex bus network. Thirdly, on empirical study is conducted using 29 million bus smart card records from the central urban area of Chongqing. The community detection model divides the city into 13 bus charging station service partitions. Within each partition, the charging-demand- weighted gravity method is applied to determine the station location. The results are then compared with those obtained using the Affinity Propagation (AP) clustering method and the P-median location method. The results show that the proposed method effectively accommodates bus charging demand over a large temporal scale. Based on the composite indicator of average charging distance adjusted by a penalty coefficient, the proposed method reduces this indicator by 14.51% compared to the AP clustering method and by 18.33% compared to the P-median location method, implying lower charging dispatch distance costs. The proposed method thus exhibits considerable potential for macro-level service partitioning and candidate site identification.