Study On The Traveling Time Prediction For Urban Road

Traffic jam can be directly reflected by road travel time which is the important foundation of developing intelligent transportation system.Predicting the travel time of crucial road timely can not only help the traffic management department take measures for traffic control aiming at some potential congested roads,but also provide real-time road information for people and help people to make the optimal travel plan.Nowadays,while the traffic networks is taking shape,using data-driven method to construct intelligent transportation is the main means to alleviate and prevent traffic jam.Because the road travel time is affected by many environmental factors,how to use traffic history data set to accurately and real-time predict the travel time of each road is a difficult point in the study of intelligent traffic system.In the existing prediction algorithms,either the prediction models are too simple to fully combine the space-time characteristics of the road network to extract key features,or the established models are limited in the real-time aspect of the prediction.Considering these problems,an integrated algorithm(i.e.random forest)based on regression tree are chosen as the prediction method of urban road travel time in this thesis.Because the random forest algorithm can adapt to the uncertainty and complexity of the road network better,and it has a strong anti-interference ability,so the prediction performance of the algorithm will not be affected by abnormal values and missing values,and the training time of the model is short,which is more suitable for the prediction scenario with strong real-time.The main content of the thesis includes three parts: Firstly,the maximum-relevancy and minimum-redundancy algorithm(mRMR)is used to select features for the prediction model,mRMR can improve the performance of the prediction model by reducing or eliminating redundant features in the feature subsets.Secondly,when constructing the prediction model,each regression tree in the random forest was combined with the DBSCAN(Density-Based Spatial Clustering of Applications with Noise)algorithm to avoid the influence of outliers in the leaf nodes on the prediction results,and the experimental results show that the improved random forest model has better prediction accuracy than orther model.Finally,according to the upstream and downstream relationship of the road section,we abstract the road network structure as a time-dependent traffic network diagram,and propose a calculation method of the optimal path that suitable for the network environment by combining the road travel time,road grade and path length.