Abstract: This paper proposes a travel time calculating method based on the traffic state of the grid by mining trajectory data of taxis. With the gridding of a selected region, this paper utilizes the GPS data of taxis to construct the macroscopic fundamental diagram and to fit the flow- density function of the macroscopic fundamental diagram. By applying the Gaussian mixture clustering method, the traffic states are divided into three categories: free flow, mild congestion, and severe congestion. Through analyzing the travel time of grids with the three traffic states, we found that the travel time of grids reveals different characteristics of the distribution. Specifically, Gamma distribution, Weibull distribution, and lognormal distribution correspond to the states of free flow, mild congestion, and severe congestion, respectively. Furthermore, this paper deduces the probability density model for travel time by approximating the joint probability density distribution of travel time in different grids. The empirical results illustrate that the proposed method can quickly calculate the travel time with specific reliability, and the mean absolute error is between 1% and 16%. The proposed method in this paper can provide technical support for traffic guidance and enhanced navigation.

Key words: intelligent transportation, travel time, gridding, traffic state identification, macroscopic fundamental diagram

摘要： 在出租车轨迹数据挖掘的基础上，本文提出基于网格交通状态的行程时间计算方法。在区域网格化的基础上，利用出租车全球定位系统(Global Positioning System, GPS)数据构建区域网格宏观基本图，并对宏观基本图的流量-密度关系进行拟合；进而使用高斯混合聚类法，将区域交通状态分类为畅通、轻度拥堵和重度拥堵。对不同交通状态网格的行程时间进行挖掘分析，发现3类交通状态下网格行程时间表现出不同的分布特征，畅通、轻度拥堵和重度拥堵的最佳行程时间分布分别为Gamma分布、Weibull分布和对数正态分布；通过不同状态网格行驶时间联合概率密度分布的近似拟合推导出路径网格行程时间概率密度模型。本文提出的方法可以快速计算一 定可靠度条件下的行程时间，对不同线路和时间内的案例分析结果表明，该方法对路径行程时间估计的平均绝对误差在1%~16%，可以为交通诱导与未来导航提供技术方法支撑。

关键词: 智能交通, 行程时间, 网格化, 交通状态辨识, 宏观基本图