【作者】 王磊；

【导师】 赵建东；

【作者基本信息】 北京交通大学 ， 交通运输规划与管理， 2021， 硕士

【摘要】 随着全国机动车辆的与日俱增,道路上的交通拥堵成了亟待解决的问题。相比于其他种类车辆,货车在行驶过程中易形成停留点,且在某些区域停留时间较长,这些停留点扩散后会对周边交通产生影响,造成局部区域拥堵,为交通事故埋下隐患。因此,本文基于货车GPS轨迹数据,提取货车停留点,从而在宏观上对货车热点区域进行识别。在微观上挖掘热点区域内货车交通变化规律,对区域内货车活跃等级进行判断研究,进而实现对热点区域内的货车交通状态的预测,为道路管理部门对热点区域内部及周边实行交通管控提供依据,减少区域交通拥堵的发生,提高货物运输效率。主要内容如下:(1)针对货车轨迹数据中存在异常数据的问题,制定了异常数据处理规则。首先,对货车GPS数据采集原理进行描述。其次,对北京市六环区域内货车GPS数据进行筛选。然后,针对五种具体的货车GPS异常数据形式,制定了相应的数据清洗原则,并绘制出GPS异常数据清洗流程图。(2)构建了货车热点区域识别模型。首先,提出了货车停留点提取算法对货车停留点进行提取。其次,搭建了基于DBSCAN密度聚类算法的货车热点区域识别模型,在工作日与周末两种场景下对货车热点区域进行识别。然后,又搭建了基于核密度估计算法的货车热点区域识别模型进行识别对比,从而证明了所识别出的货车热点区域的准确性。(3)提出了热点区域内货车活跃度的概念,并划分出货车活跃等级。首先,针对热点区域内部的货车流量与平均速度两类参数,提出了货车活跃度的概念。其次,参考北京市道路交通拥堵指数分级,将货车活跃度划分为五个等级。然后,以新发地与空港物流园区域为例,对两个热点区域内部的货车流量与平均速度进行统计分析,从而对其工作日与周末的货车活跃度与活跃等级进行可视化展示。(4)构建了基于GRU的热点区域货车流量与速度预测模型。首先,利用梯度下降优化算法Adam对GRU模型权重参数进行优化。其次,利用APSO算法对GRU模型的超参数进行寻优,搭建了基于APSO-GRU的热点区域货车流量与速度预测模型。然后,以货车热点区域新发地为例,对预测模型进行验证。分别在工作日与周末两种场景下对新发地区域内的货车流量与平均速度进行预测,进而实现对货车活跃等级的判断。最后,将APSO-GRU模型的预测结果与GRU、SVR、ARIMA模型进行对比,在货车流量预测方面,预测精度分别提升了0.75%、2.44%和4.05%;在货车速度预测方面,预测精度分别提升了0.51%、2.31%和3.83%.更多还原

【Abstract】 With the increasing number of motor vehicles,traffic congestion on the road has become an urgent problem.Compared with other types of vehicles,freight cars are prone to form stop points in the driving process,and stay longer in some areas.These stop points will affect the surrounding traffic by the means of causing local regional congestion and leading to potential traffic accidents after diffusion.Therefore,based on the truck GPS trajectory data,this study extracts the truck stop point to identify the hot spot area of the truck.Mining the change laws of truck traffic in the hot area,judging the active level of truck in the area,and then realizing the prediction of truck traffic state in the hot area,providing the basis for road management departments to implement traffic control in and around the hot area,reducing the occurrence of regional traffic congestion and improving the efficiency of freight transportation.The main contents are as follows:(1)Aiming at the abnormal data in truck trajectory,the abnormal data cleansing rules are proposed.Firstly,the collection principle of truck GPS data is described.Secondly,the GPS data of freight cars in Beijing six ring area are screened.At last,according to five specific forms of truck GPS abnormal data,the corresponding data cleaning rules are proposed,and the flow chart of GPS abnormal data cleaning process is drawn.(2)The hot spot area recognition model of truck is constructed.Firstly,the freight car stop point extraction algorithm is proposed.Secondly,a hot spot area recognition model based on DBSCAN density clustering algorithm is built to identify the hot spot area of freight cars on working days and weekends.Then,the identification model of freight car hotspots based on kernel density estimation algorithm is built for identification and comparison.By using this identification model,the accuracy of the identified freight car hotspots is proved.(3)The concept of the degree of truck activity in hot spots is proposed,and the truck activity level is divided.Firstly,the concept of the degree of truck activity is proposed under the parameters of truck flow and average speed in hot spot area.Secondly,referring to the classification of road traffic congestion index in Beijing,the degree of truck activity is divided into five levels.Then,taking the area of Xinfadi and Airport Logistics Park as an example,the freight traffic flow rate and average speed in these two hot spots are statistically analyzed,and the degree of freight traffic activity and its activity level on a certain working day and weekend are visualized.(4)The prediction model of truck flow and speed in hot spot area based on GRU is constructed.Firstly,the weight parameters of GRU model are optimized by gradient descent algorithm Adam.Secondly,the improved PSO algorithm is used to iteratively optimize the hyper-parameters of the GRU model.With the improved PSO algorithm,a prediction model of freight traffic and speed in hot spots based on APSO-GRU is built.Then,by using the newly identified hot spots of freight cars,the prediction model is verified.The truck flow and average speed in the new area are predicted on working days and weekends,respectively,so as to judge the level of truck activity.Finally,the prediction results of APSO-GRU model are compared with those of GRU,SVR and ARIMA models,and its prediction accuracy is increased by 0.75%,2.44% and 4.05%,respectively in the prediction of freight traffic flow rate.In terms of truck speed prediction,the prediction accuracy is increased by 0.51%,2.31% and 3.83%,respectively.更多还原