Data-Based Evaluation And Prediction Of Traffic Congestion

With the number of vehicles growing rapidly, most of domestic cities, especially large city traffic congestion problem becomes more and more serious. Traffic congestion of cities seriously affects the daily work and life of citizens. Subjected to many conditions, roads develop slowly. As a new way to control the traffic congestion, intelligent transportation has become the key work of traffic management departments. Various sensors installed in city roads can obtain large amounts of data every day. How to use these data to help solve the problem of traffic congestion has become a hot research field of transportation.Traffic congestion is not only an important basis for the service level of road traffic, but also the premise of traffic management and control. Differentiating traffic congestion state accurately is very important for forecasting road congestion, guiding traffic and planning the best route. However, actually, because the road traffic data is difficult to gain, and the level of information sharing is very low. Most of researches still stay in the stage of theoretical modeling and simulation. There is a large space to improve the traffic data analysis of real roads. Based on a large number of real roads data(card slot data, microwave date and GPS data), this paper studies the traffic congestion evaluation and prediction method. Three achievements have been completed as follows:(1) After detail analyzing on data collected by multi-source sensors which are easily seen on road, and design a license plate hash algorithm to remove redundant data, in view of the advantages and disadvantages of various sensors, a method of multi-source traffic data fusion has been put forward and effectively revised the singular data in the original data combining with the quality of the real data.(2) Evaluation of city road traffic congestion state. Based on the real traffic data, the congestion state have been evaluated by the K-means clustering method. Traffic congestion state is the key point of traffic state analysis. However, traffic congestion state has a low proportion of traffic state, so using the K-means clustering method can’t divide the traffic congestion state. Therefore, to solve this problem, an evaluation method of city road traffic congestion state, based on density, has been put forward. It can divide the traffic congestion state effectively.(3) Research on forecast method of city road traffic congestion state. To solve the lag problem of forecasting traffic congestion state by one-order Markov model, a forecast method of short-time traffic congestion state based on high-order Markov model has been put forward. The method make accuracy of forecasting traffic congestion state improve to 92.7%, and solve the lag problem effectively.