Research On Link Travel Time Estimation And Prediction Based On Data Driven

As an important traffic variable, the link travel time plays an important role in transportation analysis and control. An accurate estimation and prediction of the link travel time thus becomes a key to the traffic management and guidance. It does a lot of help in many aspects such as easing the traffic congestion, avoiding the waste of social resources, reducing the environmental pollution and reducing the economic loss.With the development of intelligence and informatization in transportation, a series of processing methods of massive traffic data, such as recording, storage or extraction, is no longer a technical problem. Due to the huge amount of data, some model-based estimating or predicting methods will face many problems in traffic estimation or prediction, such as too many parameters involved, complex model structure or low accuracy. However, some data-driven based methods achieve better results and reflect internal relationships of the data well. In this paper, we use data-driven methods to estimate and predict the link travel time.A Section of an urban expressway in Beijing is chosen as the target link of this paper. Some data-driven methods are used to estimate and predict the link travel time of the target link. In the travel time estimation stage, a simulation Scenario of the target link based on VISSIM is established to achieve the detailed vehicle data. Then the true link travel time and the simulated probe vehicle data are both obtained from it. An optimized BP neural network based on the genetic algorithm is established to estimate the link travel time. Based on the effectiveness analyze of the BP neural network, which is proved by the simulated data, the real probe data is substituted into the network to obtain the estimated the link travel time in the real case.In the travel time prediction stage, some historical databases are established based on the estimated link travel time, which is the output of the stage of the travel time estimation. Two categories of the data-driven methods, i.e. the nonparametric regression and the nonparametric regression combined with a nonlinear fitting, are used to predict the link travel time. Some numerical experiments are conducted by varying the input parameters to test the effectiveness of the two types of methods. The input is respectively set to be the moment when the vehicle runs into the link, the average vehicle velocity of the link at the moment, and the combined the above two types of information. The prediction evaluations of various combinations of the predicting methods and input are carried out through calculating the average percentage error (APE) and the mean percentage absolute error (MAPE). The result shows that when the moment when the vehicle runs into the link and the average vehicle velocity are both treated as the input information, the prediction performs best. On the other hand, the nonparametric regression combined with a nonlinear fitting method performs better than the pure nonparametric regression method.