Travel Time Estimation Based On Data Correlation

Probability theory and mathematical statistics are widely used in physics,chemistry,engineering,biology,economy,sociology and other fields.In this thesis,travel time estimation problem is investigated using the theories in mathematical statistics,optimization methods,machine learning and so on.The travel time is an important parameter that directly reflect the performance of traffic condition.Travel time estimation using the sensor data has always been considered an important issue in the field of transportation management.However,in reality,due to the limitations of physical conditions(sensor location,quantity,accuracy,environmental impact,etc.),it is difficult for the sensor systems to cover the whole network.In view of this,how to efficiently locate the sensors in the transportation network so as to estimate the travel time of the whole road network is a valuable and important problem.Firstly,this paper studies the travel time correlation based on machine learning methods.On this basis,sensor location models and related network-wide travel time estimation models are proposed.Numerical experiments are then carried out to verify the effectiveness of the proposed models.The summaries of each chapter are as follows:The first chapter introduces the research background and significance of the traffic travel time estimation based on data correlation,and discusses the up-to-date research progress of travel time estimation and sensor location problems.A brief introduction of the user equilibrium traffic assignment model is also presented.In Chapter 2,by using the method of random forest in machine learning,a correlation matrix is generated to quantify the travel time correlation of different links in spatial manner.On the basis of the correlation matrix,a sensor location model is proposed for small-scale networks,and a two-stage sensor location model is proposed for large-scale networks,which is formulated as an integer programming problem.In Chapter 3,the network-wide travel time estimation model is given.Owing to the results of sensor location models in Chapter 2,the set of the observed links with the strong correlation in the traffic network is obtained.According to the observed travel time data of these links,an artificial neural network is established to estimate the travel time of the unobserved links,so as to obtain the travel times of all links in the whole road network.The effectiveness of the proposed model is demonstrated by two numerical examples.In Chapter 4,the sensor location model in Chapter 2 is extended to a new sensor location model based on estimating error,which aims to minimize the estimating error of the estimated travel time.The corresponding model is formulated as an integer programming problem.The objective function of this model is the minimization of the total estimating errors subject to the bound constraint of the total number of sensors.In the numerical experiment,the results of this newly proposed model are compared with the results of the model in Chapter 2 to show the effectiveness.Chapter 5 concludes the thesis with suggestions for further studies.The thesis includes 12 figures,16 tables and 60 references.