Research on urban traffic anomaly detection method based on collective anomaly mining

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Sept.2023, Vol.46 No.9: 1237-1246

DOI: 10.3969/j.issn.1003-5060.2023.09.014

Authors

HUANG Xiaodi $ ^{1} $, ZHU Xiaoxi $ ^{2} $, WU Shuhui $ ^{2} $, HU Zhongfeng $ ^{1} $

(1. School of Economics and Management, Hefei University, Hefei 230601, China; 2. School of Management, Hefei University of Technology, Hefei 230009, China)

Abstract and Keywords

Abstract: Aiming at the problem of abnormal state detection of urban traffic in multi-source data environment, an integrated detection model of offline fitting and online detection based on collective anomaly mining is proposed. The model uses distance-density-weight k-medoids (DDWK-medoids) algorithm, which is measured by distance-density-weight, to determine the number and location of traffic pivot points according to urban traffic situation adaptively. Based on this, traffic flow anomaly detection and trajectory anomaly detection are combined by fitting multi-source traffic data at different resolutions. In the detection model, neither the temporal nor spatial attributes of the data are directly involved in the computation in the form of numerical values, which effectively reduces the computational complexity. Experimental results show that the runtime of this algorithm is significantly better than that of the comparison algorithm for test datasets of different sizes. Especially for large datasets with sufficient samples, there is clear advantage in terms of detection accuracy.

Keywords: urban traffic; anomaly detection; collective anomaly; distance-density-weight k-medoids (DDWK-medoids) algorithm; traffic pivot point

作者信息

黄晓地 $ ^{1} $，朱晓曦 $ ^{2} $，吴淑慧 $ ^{2} $，胡中峰 $ ^{1} $

(1. 合肥学院经济与管理学院, 安徽合肥 230601; 2. 合肥工业大学管理学院, 安徽合肥 230009)

摘要和关键词

摘要: 针对时空数据环境下的城市交通异常检测问题, 文章提出一种基于集体离群点挖掘的“线下拟合-线上检测”一体化检测模型。该模型采用以距离-密度-权重为度量的改进聚类（distance-density-weight k-medoids, DDWK-medoids）算法, 根据城市交通态势自适应确定交通枢纽点的数量和位置, 通过对数据进行不同分辨率拟合, 将交通流量异常检测与交通轨迹异常检测相结合。在该检测模型中, 数据的时间属性与空间属性均未以数值的形式直接参与计算, 有效降低了运算复杂度。实验结果表明, 该模型算法对于不同规模数据集的处理速度均显著优于对比算法, 尤其是对于样本充足的大规模数据集, 检测的准确率更具有明显优势。

关键词: 城市交通; 异常检测; 集体离群点; 基于距离-密度-权重度量的改进聚类(DDWK-medoids)算法; 交通枢纽点

Funding

国家自然科学基金资助项目（72271075）；安徽省高等学校优秀青年人才基金资助项目（2022AH051774）

合肥工业大学学报(自然科学版)

Navigation