Travel Time Prediction Of Urban Arterial Road Based On License Plate Recognition Data

Travel time prediction of urban road is not only an indispensable part of urban traffic information service system,but also an important support for urban intelligent traffic control and management system.Accurately predicting travel time of urban road can not only offer real-time and useful traffic information to traffic participants,but also provide decision-making basis for the traffic management authorities to formulate strategies for traffic control and management.The dissertation studies travel time prediction of urban road problem based on license plate recognition data,and the following lists the contents of the study.(1)This dissertation utilizes the time series ARIMA method to make short-term prediction of short links and long links and then proposes an algorithm of combining travel time prediction results of short links to obtain the travel time prediction of long links.Afterwards,the results of direct prediction method and combined prediction method are compared and analyzed which show that the combined prediction method has higher accuracy and better practicability in short-term travel time prediction.(2)This dissertation applies Principle Component Analysis(PCA)-Gradient Boosting Decision Tree(GBDT)method to propose a model considering temporal feature for travel time prediction of urban road whose prediction accuracy and algorithm practicability is validated using real data of Fuzhou south road,Qingdao.The Results indicate that compared to the conventional k-Nearest Neighboring(kNN)method,time series ARIMA model and support vector machine(SVM)approach,the proposed PCAGBDT algorithm proves to better capture the nonlinear nature of traffic flow and the characteristics of travel time changes of urban road.Therefore,it can better realize the travel time prediction and can be applied to the actual travel time prediction.(3)This dissertation proposes GM-kNN model considering spatial and temporal features for travel time prediction of urban road.Meanwhile,three different approaches of calculating distance similarity are put forward which are tested and verified by real data of Fuzhou south road,Qingdao.At the same time,in order to consider the characteristics of the early and late peaks of the traffic flow,the hierarchical clustering analysis method is used to classify the traffic flow in different periods and different parameters are used for different periods.It turns out to be that GM-kNN model considering spatial-temporal features outperform conventional kNN model with higher prediction accuracy and better performance.(4)This dissertation proposes three types of long short-term neural network(LSTM)architecture based on deep learning to make travel time prediction of Fuzhou south road,Qingdao with and without considering spatial correlation,respectively.Results indicate that considering spatial correlation in travel time prediction can improve the performance of LSTM,which can be used for travel time prediction of urban road.