Chris Tampère

An Extended Kalman Filter Application for Traffic State Estimation Using CTM with Implicit Mode Switching and Dynamic Parameters

This paper presents a traffic state estimation and prediction model based on the cell transmission model (CTM). The nonlinear CTM is transcribed in a closed analytical state-space form for use within a general extended Kalman filtering framework. The state-space CTM switches implicitly between numerous possible linear modes. The paper provides measurement models for the traffic state and model parameters for automatically estimating traffic conditions and model parameters in an online context. The applicability of the approach is illustrated in a real and a simulated case study.

Methodology for Identifying Vulnerable Sections in a National Road Network

Under the authority of the Flemish Traffic Centre, a study was carried out to identify road sections that are vulnerable to major incidents. The primary objective of the study was to develop a methodology capable of rapidly scanning a large network for the most vulnerable sections. The methodology builds on a number of vulnerability indicators based on an analysis of the typical stages of an incident. With these indicators, the most important bottlenecks in the network can be selected and subjected to a closer examination. The methodology developed in the project was tested on a study area comprising the network in the corridor between the two large cities of Brussels and Ghent, Belgium. For this network, an analysis of empirical data was performed to gain insight into the extent of the reliability problems on the Flemish road network.

New Gradient Approximation Method for Dynamic Origin-Destination Matrix Estimation on Congested Networks

In origin–destination (O-D) estimation methods, the relationship between the link flows and the O-D flows is typically approximated by a linear function described by the assignment matrix that corresponds with the current estimate of the O-D flows. However, this relationship implicitly assumes the link flows to be separable; this assumption leads to biased results in congested networks. The use of a different linear approximation of the relationship between O-D flows and link flows has been suggested to take into account link flows being nonseparable. However, deriving this relationship is cumbersome in terms of computation time. In this paper, the use of marginal computation (MaC) is proposed. MaC is a computationally efficient method that performs a perturbation analysis, with the use of kinematic wave theory principles, to derive this relationship. The use of MaC for dynamic O-D estimation was tested on a study network and on a real network. In both cases the proposed methodology performed better than traditional O-D estimation approaches, and thereby showed its merit.

Dynamic origin–destination estimation in congested networks: theoretical findings and implications in practice

In this study we analyse the impact of congestion in dynamic origin–destination (OD) estimation. This problem is typically expressed using a bi-level formulation. When solving this problem the relationship between OD flows and link flows is linearised. In this article the effect of using two types of linear relationship on the estimation process is analysed. It is shown that one type of linearisation implicitly assumes separability of the link flows, which can lead to biased results when dealing with congested networks. Advantages and disadvantages of adopting non-separable relationships are discussed. Another important source of error attributable to congestion dynamics is the presence of local minima in the objective function. It is illustrated that these local minima are the result of an incorrect interpretation of the information from the detectors. The theoretical findings are cast into a new methodology, which is successfully tested in a proof of concept.

Marginal Incident Computation: An Efficient Algorithm to Determine Congestion Spillback due to Incidents

In studies on the influence of incidents on travel time, researchers rely on Monte Carlo simulation. Because this procedure is demanding computationally, the research scope is limited. This paper presents a highly efficient method for approximately quantifying congestion spillback due to incidents: marginal incident computation (MIC). MIC superimposes the effect of an incident on a single base simulation run (without incidents) instead of carrying out a complete dynamic network loading with the incident, which would involve many calculations identical to the base simulation (e.g., before or far away from the incident). Whereas the results obtained with MIC vary only slightly from the outcome of a complete dynamic network loading, the gain in computation time is significant: a factor > 1,100 for a case study of the Sioux Falls, South Dakota, benchmark network.

Macroscopic effects of reaction time on traffic flow characteristics

Reaction time is defined as a physiological parameter reflecting the period of time between perceiving a stimulus and performing a relevant action. In the traffic flow theory literature, the effects of reaction time on string stability have been described using the microscopic modeling approach. This paper presents a distinct approach to investigate how reaction time influences traffic flow stability using a macroscopic model. In the paper, the distinction between string stability and flow stability is defined. The flow stability conditions are derived based on the macroscopic model, which is developed from a gas-kinetic principle. From linear analysis, we find that at macroscopic scale the reaction time influences how instabilities propagate but does not contribute to whether those (linear) instabilities occur. Nevertheless, nonlinear analysis might give a different view on the impact of reaction time on traffic flow stability, but the effect is nonlinear. We argue that the findings provide a better understanding of the effects of reaction time on traffic flow characteristics.

Gas-Kinetic Traffic Flow Modeling Including Continuous Driver Behavior Models

A modeling technique is presented that analytically bridges the gap between microscopic behavior of individual drivers and the macroscopic dynamics of traffic flow. The basis of this approach is the (gas-) kinetic or mesoscopic modeling principle that considers the dynamics of traffic density and generalizations thereof as a probability density function of vehicles in different driving states. In contrast to traditional kinetic models, deceleration of individual vehicles due to slower traffic is treated as a continuous adaptive process rather than a discrete event. An analytic procedure is proposed to aggregate arbitrarily refined individual driver behavior to a macroscopic expected acceleration or deceleration of flow as a whole that can be used in macroscopic differential equations for traffic flow. The procedure implicitly accounts for the anisotropy of information flow in traffic, for anticipation behavior of drivers, and for the finite space requirement of vehicles, as long as these properties have been specified at the level of individual driver behavior. The procedure is illustrated for a simple car-following model with overtaking opportunity. The results show that the procedure yields micro-based aggregate traffic flow models that capture the essential properties of traffic dynamics. The techniques presented can contribute to the development of traffic flow models with driver behavior and driver psychology as important explanatory factors of congestion formation and propagation. Moreover, the approach allows building macroscopic traffic flow equations from future traffic flows for which no empirical speed–flow–density relations are available yet.

Dynamic Origin–Destination Matrix Estimation on Large-Scale Congested Networks Using a Hierarchical Decomposition Scheme

Despite its ever-increasing computing power, dynamic origin–destination (OD) estimation in congested networks remains troublesome. In previous research, we have shown that an unbiased estimation requires the calculation of the sensitivity of the link flows to all origin–destination flows, in order to incorporate the effects of congestion spillback. This is, however, computationally infeasible for large-scale networks. To overcome this issue, we propose a hierarchical approach for off-line application that decomposes the dynamic OD estimation procedure in space. The main idea is to perform a more accurate dynamic OD estimation only on subareas where there is congestion spillback. The output of this estimation is then used as input for the OD estimation on the whole network. This hierarchical approach solves many practical and theoretical limitations of traditional OD estimation methods. The main advantage is that different OD estimation method can be used for different parts of the network as necessary. This allows applying more advanced and accurate, but more time-consuming, methods only where necessary. The hierarchical approach is tested on a study network and on a real network. In both cases the proposed methodology performs better than traditional OD estimation approaches, indicating its merit.

Two-Step Approach for Correction of Seed Matrix in Dynamic Demand Estimation

In this work, deterministic and stochastic optimization methods are tested for solving the dynamic demand estimation problem. All the adopted methods demonstrate difficulty in reproducing the correct traffic regime, especially if the seed matrix is not sufficiently close to the real one. Therefore, a new and intuitive procedure to specify an opportune starting seed matrix is proposed: it is a two-step procedure based on the concept of dividing the problem into small problems, with a focus on specific origin–destination (O-D) pairs in different steps. Specifically, the first step focuses on the optimization of a subset of O-D variables (the ones that generate the higher flows or the ones that generate bottlenecks on the network). In the second step the optimization works on all the O-D pairs, with the matrix derived from the first step as starting matrix. In this way it is possible to use a performance optimization method for every step; this technique improves the performance of the method and the quality of the result with respect to the classical one-step approach. The procedure was tested on the real-world network of Antwerp, Belgium, and demonstrated its efficacy in combination with different optimization methods.

Continuous Traffic Flow Modeling of Driver Support Systems in Multiclass Traffic With Intervehicle Communication and Drivers in the Loop

This paper presents a continuous traffic-flow model for the explorative analysis of advanced driver-assistance systems (ADASs). Such systems use technology (sensors and intervehicle communication) to support the task of the driver, who retains full control over the vehicle. Based on a review of different traffic-flow modeling approaches and their suitability for exploring traffic-flow patterns in the presence of ADASs, kinetic traffic-flow models are selected because of their good representation on both the aggregate level (congestion dynamics) and the level of the individual vehicle (vehicular interactions either directly or through intervehicle communication). The human-kinetic modeling approach is presented. It is a multiclass variant of kinetic traffic-flow models that is strongly based on individual driver behavior, i.e., on fully continuous acceleration/deceleration behavior and explicit modeling of the activation level of the driver. The strength of this modeling approach is illustrated by application to a driver-assistance system that uses intervehicle communication. It warns drivers when approaching sharp decelerations in a queue tail. The explorative analysis shows that the system results in safer and smoother transition from free-flowing to congested traffic. It also avoids compression of the queue tail, thus preventing the emergence of stop-and-go congestion patterns.