Freeway Traffic Flow Quantities Estimation Based On Heterogeneous Data Sources

For road traffic system, the main principle of mitigating the traffic congestion is to take both traffic supply and traffic demand in consideration. Particularly, the capacity of the traffic infrastructure shall be fully utilized and the traffic demand shall be controlled in a sense to keep the traffic system running in a driver-satisfied level of service. Now, the soaring development of sensing technology, communication technology, information technology and computer technology takes traffic surveillance technology into the threshold of multi-sensors based detecting period. These traffic surveillance technologies can produce massive data sources, which can be used to advancethe efficiency of traffic system by modern traffic management or traffic flow organization. The advance of efficiency of traffic system will make the capacity of infrastructureto be maximized. However, caused by the economy cost, privacy concerns and bad weather condition, etc., the traffic sensing devices could not be installed enough to fully cover the traffic network. Thus, the traffic state estimation technology is used to know the traffic state based on the limited partly covered data sources.Based on the traffic kinematic wave theory, this dissertation studies the traffic state estimation problem in the highway or freeway segment by following aspects:(1)Estimating the microscopic and macroscopic traffic states using the famous three-detector theory. Be different from the classified kinematic wave theory, which adopts the flow or density as the system sate variable, the Three-detector theory adopts the cumulative flow count as the system state variable. By doing this, the traffic state inside the boundary and along the boundary can be explicitly modeled, rather than by a cell-by-cell transfer framework like the cell transmission model. Under the perfect boundary condition, this study not only can estimate the macroscopic traffic states such as link based travel time as well as the evolution of flow or density along the segment of interest, but also the microscopic traffic states such as individual vehicle trajectory as well as the vehicle fuel consumption or emission;(2)Proposing a stochastic boundary based three-detector model. In the real world, the traffic state measurement error is inevitable and the sampling frequency is limited, which imply that the assumed perfect boundary is hard to satisfy in current sensing condition. Thus, this imperfect boundary, namely the measurements are discontinuous and noised, is necessary to modeled in the traffic state estimation framework. By assuming the measurement noise as a white noise and designing a linear interpolation algorithm, this study models the imperfect boundary a stochastic boundary, which is constructed by a cumulative flow count vector and error variance-covariance matrix. In this stochastic boundary condition, the system process equation is modified as the minimization of two normal distributed random variables, rather than the minimization of two deterministic variables. The classical Probit model, which is extensively used in the field of random route choice, can produce the minimization or maximization of multiple random variable based the Clark’s approximation theory. This study employs the notion of Probit model and Clark’s approximation to approximate the cumulative flow count at any point inside the boundary as a normally distributed random variable, the mean and variance of which are calculated by the mean and variance of the related cumulative flow count at the boundary. This modeling notion is very plausible to quantify the estimation uncertainty.(3)Modeling the observations of heterogeneous data sources as a series of linear measurement equations in a Kalman filter framework. The considered heterogeneous data sources are includingthe flow, occupancy and instantaneousspeed of the loop detectors, the travel time observation from the automatic vehicle identification surveillance technology, and the semi-continuous vehicle trajectory from the global position system based surveillance technology. Under the framework of the proposed stochastic three-detector model, all the aforementioned observations can be modeled or transferred as cumulative flow count observations. Along this way, the linear measurement equation of each kind of observation can be built. In the Kalman framework, based on the measurement equation and the system process equation (from the stochastic three-detector model), an optimal Kalman gain can be calculated and further the prior traffic state estimation can be updated to a posterior traffic state estimation including an optimal traffic state estimation and a posterior error variance-covariance matrix. The posterior error variance-covariance describes the range of estimation uncertainty. Theoretically, the more the observations are, the less estimation uncertainty is.(4)Estimating the value of information of each observation based on the informationtheory.Above studies build the relationship between each observation and the traffic state estimation uncertainty, such as density estimation uncertainty, and the uncertainty can be quantified using the trace of the error variance-covariance matrix. The inverse of the uncertainty can be used to calculate the value of information. Thus, we can calculate the value of information for each observation, which is very useful for the traffic engineer to decide the sensing configurations. Particularly, this study focus on the value of information of different sampling frequency of the middle loop detector, the market penetration rate of automatic vehicle identification based surveillance technology, and the market penetration rate of global position system based surveillance technology.(5)Modeling the traffic state estimation problem of a typical freeway corridor as a non-linear optimization problem under a constrained least squares framework. The minimization function of the three-detector theory can be equally transformed as two inequalities constraints to describe the traffic flow propagation, particularly the forward and backward shockwave propagation. Based on the flow conservation at merge or diverge point, the total inflow and total outflow are modeled as an equality constraint. The capacity limit of any point along the corridor can be model as an inequality constraint. The objective of the optimization problem is constructed by the total least squares of the difference between measurements and system variable in the notion of cumulative flow count. In addition, the travel time observation can be weighted into the objective to reflect the real link based traffic states.