基于边缘计算的危险品运输车辆跟踪预警方法

Tracking and Warning Technology of Hazmat Transport Vehicles in Edge Computing Environment

边缘计算作为一种新兴的分布式计算模型,可提升信息系统的响应速度、节省网络带宽资源。将边缘计算应用于危险品运输车辆的跟踪预警,提出一种边缘智能路侧节点的协作式跟踪算法,在保障实时性的前提下解决了GNSS失锁导致轨迹缺失问题;针对车辆跟踪时的异常状态预警问题,将智能路侧感知的多源信息与车辆运行状态相结合,基于机器学习提出一种融合自车、周围车辆、道路和自然环境等特征因素的预警算法,有效提升了异常检测精度。使用SUMO交通仿真器分析了跟踪算法的性能,结果表明,边缘计算较传统云计算方式时延平均降低了90%,带宽消耗平均减少了68%。基于美国交通部网联车开放数据,通过参数调优分别建立了基于SVM,KNN,Adaboost的单车动力学变量与多因素变量的6种异常检测模型,实验表明,多因素变量检测模型优于单车动力学变量模型。基于SVM的多因素变量模型性能最优,其准确率为0.972,召回率为0.98,AUC为0.974。

As a new distributed computing paradigm,edge computing can improve system response speed and network bandwidth.In this paper,edge computing is applied for tracking and early warning of hazmat transport vehicles.A collaborative tracking algorithm for edge intelligent roadside units is proposed,which solves the problem of incomplete trajectories caused by GNSS outage and ensures the real-time performance at the same time.For anomaly warning during tracking of hazmat transport vehicles,an early warning algorithm is proposed based on machine learning to fuse vehicle status with multi-source information from the intelligent roadside units.The algorithm that integrates factors such as vehicles,surrounding vehicles,roads and natural environment,can effectively improve the accuracy of anomaly detection.Simulations for tracking algorithm performance are carried out within SUMO traffic simulator.Compared with cloud computing,the results show that edge computing reduces the delay by 90%and bandwidth consumption by 68%on average.Use the connected vehicle pilot open data published by the U.S.Department of Transportation,six anomaly detection models based on SVM,KNN,and Adaboost are created.The results show that multi-source information from intelligent roadside units can improve anomaly detection performance compared with models created using only variables of vehicle dynamics.Among the models created with multi-source information,SVM has the best performance,with the accuracy of 0.972,recall of 0.98,and AUC of 0.974.