Research On Key Technologies Of Urban Road Travel Time Estimation

The problem of traffic congestion has always been a troublesome problem which seriously affects the pace of urban development and the qualities of residents' lives.The problem is very difficult but urgent to be solved.Urban road travel time as visual reflection of road traffic congestion has attracted a large number of scholars to study the estimation problem on it,but has not been solved well due to the following difficulties: the data is high-dimensional and highly sparse,and different driving characteristics of users lead to differences in travel time,and road traffic conditions are affected by various factors and change dynamically.To solve the above problems,this paper studies the key technology of urban travel time estimation,and proposes a real-time traffic time estimation method based on driving characteristics and traffic conditions.The main work of this paper includes:(1)clustering users using spectral clustering algorithm with local scaling to distinguish different driving characteristics where the users are represented by low-dimensional dense representations learnt by an auto-encoder,and clustering the user travels into different traffic conditions of the roads based on the aforementioned driving-characteristic clustering to make a general travel time tensor for traffic under different driving characteristics and traffic conditions;(2)utilizing the CP tensor decomposition method to fill the missing tensor entities and make general travel time estimations for drivers of different driving characteristics and under different traffic condition,and using an extra MLP to make personal estimations;(3)applying the general estimation model and personal estimation model to a dataset containing GPS trajectories of two months and in Beijing,and compare the effectiveness and robustness of the proposed methods with that of some baseline methods on several popular metrices.The research results of urban road travel time estimation have certain reference value for solving urban traffic congestion.It can not only help related departments to understand road congestion conditions and find traffic bottlenecks,but also provide data support for road traffic congestion prevention and response measures in advance,which is also conducive for travelers to choose appropriate travel routes in advance to save travel time and avoid road congestion.