基于时空特征序列匹配的交通流状态估计方法

An Estimation Method of Traffic Flow State Based on Matching of Temporal-spatial Feature Sequences

为了针对无交通流检测器路段更好地进行交通流状态估计,提高估计精度,研究了基于时空特征序列匹配的交通流状态估计模型。通过交通运行指数的计算方法预设城市道路中有交通流参数路段的交通流状态;分析影响城市道路运行条件的各项因素,引入交通流参数与道路参数、路网拓扑参数等时空多维度参数特征,提取3个维度8个特征1个附加维度组成交通流时空特征,构建城市道路交通流DNA特征序列对交通流状态进行描述;将各个特征的值归一化处理,利用WH-KNN匹配方法,得到全路网中与待估计路段最近的交通流状态。实验选取武汉市中环快速路编号为10468、10483以及8816的路段1周数据,假定路段数据缺失,通过所述方法进行交通流状态估计,将估计结果与原始数据结果进行对比。研究表明,模型不仅能够得到无检测数据路段的交通流状态,其状态估计结果的准确率保持在88%以上,且误判结果在1个运行指数等级之内。

An estimation model of the traffic flow state based on matching of temporal-spatial feature sequences is studied to better estimate the traffic flow state for the road section without a traffic flow detector and improve the estimation accuracy.The model firstly uses the calculation method of the traffic-operation index to preset the traffic-flow state of the urban-road section with traffic flow data.Various factors affecting the operating conditions of urban roads are analyzed,with the introduction of the characteristics of time and space multi-dimensional parameters such as traffic flow parameters,road parameters,and road network topology parameters.The temporal and spatial characteristics of traffic flow form by extracting3 dimensions,8 features,and1 additional dimension,thus constructing the DNA feature sequence of urban-road traffic flow.After normalizing the value of each feature,the WH-KNN matching method is used to obtain the traffic-flow state closest to the road section to be estimated in the whole road network.The experiment selects the data of one week in road sections10468,10483,and8816 of Wuhan Zhonghuan Expressway.Assuming that the road section data is missing,the traffic flow state is estimated by the method described above,and the estimated results are compared with the original data results.The results shows that the model can obtain the traffic flow state of the road section without detection data as well as maintain the accuracy rate of the state estimation result above88%.The misjudgment is within a performance-index level.