随着机动车保有量的日益增加，道路交通压力与日俱增，快速识别城市交通状况，不仅能够为交通拥堵控制决策提供有效的支持，也为居民出行路线提供指导。近年来大数据、机器学习等计算机技术的快速迭代，车载GPS等移动交通监测设备的广泛普及，为识别城市交通状况提供了新的技术路径。本文基于出租车GPS轨迹数据，使用数据挖掘、统计分析和可视化技术，对城市交通状况变化趋势进行研究，分析城市交通拥堵的时空演化特征，应用加权K-means聚类算法，建立城市交通状况识别模型。具体的研究内容为以下几个方面：（1）基于平均速度的成都市城市交通时空分析。对原始出租车GPS数进行预处理后，使用KDTree算法将其与路网数据进行融合，实现地图匹配。基于出租车平均速度数据，通过傅里叶变换，发现成都市城市交通流量存在明显的早晚高峰期。利用栅格化技术，研究不同时段平均速度在研究区域的空间分布情况，发现平均速度在空间上的分布不均匀，早高峰时期拥堵从外围蔓延至中心区域，晚高峰则相反。使用K-means聚类、模糊C均值聚类、DBSCAN聚类算法对具有平均速度和经纬度相似性的栅格进行划分，以识别拥有相似交通状况的区域，结果表明K-means聚类算法的聚类结果较好。（2）基于加权K-means聚类的交通状况识别模型。使用Pearson相关系数分析主干道路段之间的关联关系，并根据总相关性为每个路段分配权重系数，以反映其在城市交通系统中的代表程度。以平均速度和行程时间比TTI为特征指标，结合权重系数，改进传统K-means聚类算法优化目标的计算公式，让聚类结果更准确地反映在权重大的数据点上。该交通状况识别算法成功将主干道路段的所有时间窗切片序列识别为畅通、基本畅通、轻度拥堵、中度拥堵和严重拥堵这五种不同的交通状况。（3）基于交通状况识别模型的交通拥堵识别。该识别模型能在多个场景开展应用，包括识别早晚高峰期主干道交通状况、识别常发性拥堵路段以及识别瓶颈路段等，可为交通参与者和城市交通管理部门提出对策建议，以缓解交通拥堵、减少行程延误，提高城市交通效率。
With the increasing number of motor vehicles, road traffic pressure is increasing. Rapid identification of urban traffic conditions can not only provide effective support for traffic congestion control decisions, but also provide guidance for residents‘ travel. In recent years, the rapid iteration of computer technologies such as big data and machine learning, as well as the widespread use of mobile traffic monitoring devices such as vehicle GPS, have provided new technical paths for identifying urban traffic conditions. This study uses taxi GPS trajectory data and applies data mining, statistical analysis, and visualization techniques to study the trend of urban traffic condition changes, analyze the spatiotemporal evolution characteristics of urban traffic congestion, and apply the weighted K-means clustering algorithm to establish a model for identifying urban traffic conditions. Specifically, the research content is as follows: (1) Spatial-temporal analysis of urban traffic in Chengdu based on average speed. After preprocessing the original taxi GPS data, the KDTree algorithm is used to fuse it with road network data for map matching. Based on average speed data, the Fourier transform is used to discover the obvious peak traffic periods in Chengdu. Using rasterization technology, the spatial distribution of average speed at different times in the research area is studied, and it is found that the spatial distribution of average speed is uneven, spreading from the periphery to the central area during the morning peak period and the opposite during the evening peak period. The K-means clustering, fuzzy C-means clustering, and DBSCAN clustering algorithms are used to partition grids with similar average speed and latitude and longitude to identify areas with similar traffic conditions, and the results show that the K-means clustering algorithm has better clustering results. (2) Traffic condition recognition algorithm based on weighted K-means clustering. Pearson correlation coefficients are used to analyze the correlation between the main road sections, and weight coefficients are assigned to each road section based on the total correlation to reflect their representation degree in the urban traffic system. Using average speed and travel time index (TTI) as feature indicators and combining weight coefficients, the calculation formula for optimizing the traditional K-means clustering algorithm‘s objective function is improved to make the importance of data points clearer and to more accurately reflect the clustering results on data points with higher weight. The traffic condition recognition algorithm successfully identifies all time-windowed slice sequences of main road sections into five different traffic conditions: smooth, basically smooth, mild congestion, moderate congestion, and severe congestion.(3) Traffic congestion recognition based on the traffic condition recognition algorithm. This recognition algorithm has multiple application scenarios, including identifying the traffic conditions of main road sections during peak periods, identifying frequently congested road sections, and identifying bottleneck road sections. It can provide suggestions for traffic participants and city traffic management departments to alleviate traffic congestion, reduce travel delay, and improve urban traffic efficiency.
