文章编号： 1672-5328（2026）03-0113-10

戢晓峰1, 2，杨岚1, 2，邓若凡1, 2，关昊天1, 2，李武1, 2，陈方2, 3
(1. 昆明理工大学交通工程学院，云南昆明650504；2. 西南综合交通运输发展研究院，云南昆明650504；3. 昆 明理工大学马克思主义学院，云南昆明650504)

摘要： 为识别学校周边通学关键路段的交通流集聚特性，选取昆明市典型小学周边路段为研究对 象，基于无人机视频与YOLOv8-DeepSORT目标识别结果，构建集聚概率、集聚强度与安全阈值指 标，并采用长短期记忆网络(LSTM)模型开展短时交通流预测。结果表明：空间上，学校周边呈现 高强度交通集聚，上学时段车流密度均值达放学时段的2.5 倍，而放学时段集聚强度峰值超出安全 阈值的幅度分别达140%和134%；非机动车穿插进一步压缩儿童通行空间，加剧人车冲突风险。时 间维度上，交通流呈双周期波动：上学前集聚概率以每5 min 约14%的速率递增至饱和；放学后增 长率放缓至约13%，并呈现双峰脉冲特征，车辆滞留现象突出，进一步压缩儿童避险时间。此外， 集聚强度动态演化特征分析显示，不同时段通学关键路段交通流表现出差异化的集聚与消散模式， 延长了儿童暴露于危险环境的时长。研究发现，LSTM模型预测优势显著，能够精准捕捉通学交通 流的峰谷波动及双阶段峰值特征，其预测精度优于随机森林(RF)和支持向量机(SVM)。

关键词： 通学安全；通学交通流；交通集聚特性；时间序列分析；长短期记忆网络(LSTM)

中图分类号： U491.1+12

文献标识码：A

Traffic Flow Aggregation Characteristics of Key School Commuting Routes: A Safety Management Perspective

Ji Xiaofeng1, 2, Yang Lan1, 2, Deng Ruofan1, 2, Guan Haotian1, 2, LiWu1, 2, Chen Fang2, 3
(1. Faculty of Transportation Engineering, Kunming University of Science and Technology, Kunming Yunnan 650504, China; 2. Southwest Institute of Integrated Transportation Development, Kunming Yunnan 650504, China; 3. School of Marxism, Kunming University of Science and Technology, Kunming Yunnan 650504, China)

Abstract： To identify the traffic flow aggregation characteristics of key school commuting routes, this paper selects road sections surrounding a typical elementary school in Kunming as the research subject. Based on drone video footage and YOLOv8-DeepSORT object recognition results, indicators of aggregation probability, aggregation intensity, and safety thresholds are established. Additionally, a Long Short-Term Memory (LSTM) model is employed to conduct short-term traffic flow forecasting. The results indicate that, spatially, high-intensity traffic aggregation occurs around schools, with the average traffic density during the morning commute period reaching 2.5 times that of the afternoon dismissal period. Furthermore, during the two afternoon dismissal periods, the peak aggregation intensity exceeds the safety threshold by 140% and 134%, respectively. The weaving of non-motorized vehicles further compresses the space available for children to pass, exacerbating the risk of conflicts between pedestrians and vehicles. In the temporal dimension, traffic flow exhibits a dual-period fluctuation: the probability of aggregation before school increases at a rate of approximately 14% every 5 minutes until saturation; after school, the growth rate slows to about 13%, exhibiting a double-peak pulsation pattern, with prominent vehicle queuing that further compresses children's time to avoid hazards. Furthermore, analysis of the dynamic evolution of aggregation intensity reveals that traffic flow on key school commuting routes exhibits distinct patterns of aggregation and dissipation during different time periods, prolonging the duration of children's exposure to hazardous environments. The findings show that the LSTM model demonstrates significant predictive advantages, accurately capturing peak- trough fluctuations and two- phase peak characteristics of school commute traffic flows, with prediction accuracy superior to that of Random Forest (RF) and Support Vector Machines (SVM).

Keywords： school commuting safety; school commuting traffic flow; traffic aggregation characteristics; time series analysis; Long Short-Term Memory (LSTM)