Hans van Lint

Predicting Urban Arterial Travel Time with State-Space Neural Networks and Kalman Filters

A hybrid model for predicting urban arterial travel time on the basis of so-called state-space neural networks (SSNNs) and the extended Kalman filter (EKF) is presented. Previous research demonstrated that SSNNs can address complex nonlinear spatiotemporal problems. However, SSNN models require off-line training with large sets of input-output data, presenting three main drawbacks: (a) great amounts of time and effort are involved in collecting, preparing, and executing these training sessions; (b) as the input-output mapping changes over time, the model requires complete retraining; and (c) if a different input set becomes available (e.g., from inductive loops) and the input-output mapping has to be changed, then retraining the model is impossible until enough time has passed to compose a representative training data set. To improve SSNN effectiveness, the EKF is proposed to train the SSNN instead of conventional approaches. Moreover, this network topology is derived from the urban travel time prediction...

Network-Wide Traffic State Estimation Using Loop Detector and Floating Car Data

In real-time traffic management and intelligent transportation systems (ITS) applications, an accurate picture of the prevailing traffic state in terms of speeds and densities is critical, for which traffic state estimation methods are needed. The most popular and effective techniques used are so-called model-based traffic state estimators, which consist of a dynamic traffic flow model to predict the evolution of the state variables; a set of observation equations relating sensor observations to the system state; and data-assimilation techniques to combine the model predictions with the sensor observations. Commonly, both process and observation models are formulated in Eulerian (space–time) coordinates. However, recent studies show that (first-order) macroscopic traffic flow models can be formulated and solved more efficiently and accurately in Lagrangian (vehicle number–time) coordinates (which move with traffic stream) than in Eulerian coordinates (which are fixed in space). In this article such a Lagrangian system model for state estimation is used. The approach uses the extended Kalman filtering technique, in which the discretized Lagrangian kinematic wave model with an extension (node models) for network discontinuities is used as the process equation and the average relation between vehicle spacing and speed (the fundamental diagram) is used as the observation equation. The Lagrangian state estimator is validated and compared with its Eulerian counterpart based on ground-truth data from a microscopic simulation environment. The results demonstrate that network-wide Lagrangian state estimation is possible and provide evidence that the Lagrangian estimator outperforms the Eulerian approach.

State Space Neural Networks for Freeway Travel Time Prediction

The highly non-linear characteristics of the freeway travel time prediction problem require a modeling approach that is capable of dealing with complex non-linear spatio-temporal relationships between the observable traffic quantities. Based on a state-space formulation of the travel time prediction problem, we derived a recurrent state-space neural network (SSNN) topology. The SSNN model is capable of accurately predicting experienced travel times - outperforming current practice by far - producing approximately zero mean normally distributed residuals, generally not outside a range of 10% of the real expected travel times. Furthermore, analyses of the internal states and the weight configurations revealed that the SSNN developed an internal models closely related to the underlying traffic processes. This allowed us to rationally eliminate the insignificant parameters, resulting in a Reduced SSNN topology, with just 63 adjustable weights, yielding a 72% reduction in model-size, without loss of predictive performance.

Lagrangian Formulation of Multiclass Kinematic Wave Model

The kinematic wave model is often used in simulation tools to describe dynamic traffic flow and to estimate and predict traffic states. Discretization of the model is generally based on Eulerian coordinates, which are fixed in space. However, the Lagrangian coordinate system, in which the coordinates move with the velocity of the vehicles, results in more accurate solutions. Furthermore, if the model includes multiple user classes, it describes real traffic more accurately. Such a multiclass model, in contrast to a mixed-class model, treats different types of vehicles (e.g., passenger cars and trucks or vehicles with different origins or destinations, or both) differently. The Lagrangian coordinate system is combined with a multiclass model, and a Lagrangian formulation of the kinematic wave model for multiple user classes is proposed. It is shown that the advantages of the Lagrangian formulation also apply for the multiclass model. Simulations based on the Lagrangian formulation result in more accurate solutions than simulations based on the Eulerian formulation.

Tokyo Virtual Living Lab: Designing Smart Cities Based on the 3D Internet

The Tokyo Virtual Living Lab is an experimental space based on 3D Internet technology that lets researchers conduct controlled driving and travel studies, including those involving multiple users in the same shared space. This shared-use feature is crucial for analyzing interactive driving behaviors in future smart cities. The labs novelty is two-fold: it outputs a semantically enriched graphical navigation network using free map data as input, and it includes a navigation segment agent that coordinates a multiagent traffic simulator. This simulator, which is based on the navigation network, supports the integration of user-controlled vehicles. The labs approach can significantly reduce the effort of preparing realistic driving behavior studies. To demonstrate this, the authors built a 3D model of a part of Tokyo to perform experiments with human drivers in two conditions: normal traffic and ubiquitous eco-traffic.

Anisotropy in generic multi-class traffic flow models

Traffic flow models and simulation tools are often used for traffic state estimation and prediction. Recently several multi-class models based on the kinematic wave traffic flow model have been introduced. These multi-class models take into account the heterogeneity of both vehicles and drivers. We analyse two important properties of these models: hyperbolicity and anisotropy. Both properties relate to the propagation speed of disturbances, as can be observed in real traffic. We discuss the importance of traffic flow models to be hyperbolic and anisotropic. Moreover, we develop a framework to analyse whether traffic flow models have these properties. Therefore, we derive a generic formulation of multi-class kinematic wave traffic flow models, rewrite it in the Lagrangian formulation and apply eigenvalue analysis to the resulting system of equations. Our analysis shows that most multi-class kinematic wave traffic flow models are indeed hyperbolic and anisotropic under certain modelling conditions.

Reliability of Travel Time: Effective Measures from a Behavioral Point of View

The reliability of the travel time on a route is widely regarded as one of the dominant factors affecting the route and departure time choices of travelers. There is still a lack of consensus on the quantitative measures that best reflect travel time reliability. This paper addresses this question from a travelers point of view. It was concluded from a large stated preference-revealed preference route choice experiment that a measure of the asymmetry of a travel time distribution (λskew) was an important reliability measure. Put simply, people prefer a route that is usually fast but sometimes quite costly over a route that can have any travel time over a long range. In terms of reliability there is a discrepancy between user-optimal choices and system-optimal choices, because the recent literature reveals that skewed travel time distributions may yield much higher (collective) costs than symmetric distributions. Although the results indicate that the degree of skew in the travel time distribution of a route largely affects travel choices, it is strongly argued that there is no one best measure, because what can be regarded as “best” is contingent on the goal that must be reached. In line with this view, it was found in more detailed analyses that travel information and travel purpose influenced travelers’ route choices. Further research should focus on elaborating the contingency view and on working with more individual-based reliability measures for route choice situations.

Macroscopic Modeling Framework Unifying Kinematic Wave Modeling and Three-Phase Traffic Theory

Modeling breakdown probabilities or phase-transition probabilities is an important issue when assessing and predicting the reliability of traffic flow operations. Looking at empirical spatiotemporal patterns, these probabilities clearly are a function not only of the local prevailing traffic conditions (density, speed) but also of time and space. For instance, the probability that a start-stop wave occurs generally increases when moving upstream away from the bottleneck location. A simple partial differential equation is presented that can be used to model the dynamics of breakdown probabilities, in conjunction with the well-known kinematic wave model. The main assumption is that the breakdown probability dynamics satisfy the way information propagates in a traffic flow, that is, they move along with the characteristics. The main result is that the main characteristics of the breakdown probabilities can be reproduced. This is illustrated through two examples: free flow to synchronized flow (F-S transition) and synchronized to jam (S-J transition). It is shown that the probability of an F-S transition increases away from the on ramp in the direction of the flow; the probability of an S-J transition increases as one moves upstream in the synchronized flow area.

A modular approach for exchangeable driving task models in a microscopic simulation framework

We present a structure for driver models regarding different driving tasks in microscopic simulation. This structure is part of an open and extendable simulation framework which facilitates development and research into ITS applications and driver behavior. The structure deals with typical difficulties of providing a high level of flexibility while allowing a high level of overview and user-friendliness. For driver behavior this is not a trivial task as different driving tasks depend on each other and as one requires a consistent driver, e.g. an aggressive driver is usually aggressive regarding all aspects of driving. This results in a high degree of interaction and dependency of driver models where we would like to segregate as much as possible in order to make them individually exchangeable. Our model structure allows a number of ways in which models may interact with varying degrees of generality. Inevitably this leads to concessions where we favor flexibility over user-friendliness. Specifically, the structure does not force appropriate use of model interactions. Appropriate use is thus up to the user.

Design of open source framework for traffic and travel simulation

For the evaluation, design, and planning of traffic facilities and measures, traffic simulation packages are the de facto tools for consultants, policy makers, and researchers. However, the available commercial simulation packages do not always offer the desired work flow and flexibility for academic research. In many cases, researchers resort to designing and building their own dedicated models, without an intrinsic incentive (or the practical means) to make the results available in the public domain. To make matters worse, a substantial part of these efforts pertains to rebuilding basic functionality and, in many respects, reinventing the wheel. This problem not only affects the research community but adversely affects the entire traffic simulation community and frustrates the development of traffic simulation in general. For this problem to be addressed, this paper describes an open source approach, OpenTraffic, which is being developed as a collaborative effort between the Queensland University of Technology, Australia; the National Institute of Informatics, Tokyo; and the Technical University of Delft, the Netherlands. The OpenTraffic simulation framework enables academics from geographic areas and disciplines within the traffic domain to work together and contribute to a specific topic of interest, ranging from travel choice behavior to car following, and from response to intelligent transportation systems to activity planning. The modular approach enables users of the software to focus on their area of interest, whereas other functional modules can be regarded as black boxes. Specific attention is paid to a standardization of data inputs and outputs for traffic simulations. Such standardization will allow the sharing of data with many existing commercial simulation packages.