Michael Balmer

Agent-Based Demand-Modeling Framework for Large-Scale Microsimulations

Microsimulation is becoming increasingly important in traffic demand modeling. The major advantage over traditional four-step models is the ability to simulate each traveler individually. Decision-making processes can be included for each individual. Traffic demand is the result of the different decisions made by individuals; these decisions lead to plans that the individuals then try to optimize. Therefore, such microsimulation models need appropriate initial demand patterns for all given individuals. The challenge is to create individual demand patterns out of general input data. In practice, there is a large variety of input data, which can differ in quality, spatial resolution, purpose, and other characteristics. The challenge for a flexible demand-modeling framework is to combine the various data types to produce individual demand patterns. In addition, the modeling framework has to define precise interfaces to provide portability to other models, programs, and frameworks, and it should be suitable for large-scale applications that use many millions of individuals. Because the model has to be adaptable to the given input data, the framework needs to be easily extensible with new algorithms and models. The presented demand-modeling framework for large-scale scenarios fulfils all these requirements. By modeling the demand for two different scenarios (Zurich, Switzerland, and the German states of Berlin and Brandenburg), the framework shows its flexibility in aspects of diverse input data, interfaces to third-party products, spatial resolution, and last but not least, the modeling process itself.

Towards Truly Agent-Based Traffic and Mobility Simulations

Traveling is necessary and desirable; yet, it imposes external costs on other people. Quantitative methods help finding a balance. Multi-agent simulations seem an obvious possibility here. A real world traffic simulation consists of many modules, all requiring different expertise. The paper discusses how such modules can be coupled to a complete simulation system, how such a system can be made fast enough to deal with real-world sizes (several millions of travelers), and how agent memory can be introduced. A real-world case study is presented, which says that multi-agent methods for traffic are mature enough to be used alongside existing methods. Finally, some outlook into the near future is given.

Large-Scale Multi-Agent Simulations for Transportation Applications

In many transportation simulation applications including intelligent transportation systems (ITS), behavioral responses of individual travelers are important. This implies that simulating individual travelers directly may be useful. Such a microscopic simulation, consisting of many intelligent particles (= agents), is an example of a multi-agent simulation. For ITS applications, it would be useful to simulate large metropolitan areas, with ten million travelers or more. Indeed, when using parallel computing and efficient implementations, multi-agent simulations of transportation systems of that size are feasible, with computational speeds of up to 300 times faster than real time. It is also possible to efficiently implement the simulation of day-to-day agent-based learning, and it is possible to make this implementation modular and essentially “plug-and-play.” Unfortunately, these techniques are not immediately applicable for within-day replanning, which would be paramount for ITS. Alternative techniques, which allow within-day replanning also for large scenarios, are discussed.

Capturing Human Activity Spaces: New Geometries

“Activity space,” defined as the local areas within which people move or travel during the course of their activities during a specified time period, is a measure of an individuals spatial behavior that captures individual and environmental differences and offers an alternative approach to studying the spatial reach of travelers. The shape and the area of activity space are a product of how it is conceptualized and measured. This paper enlarges the set of geometries that can be used to describe activity space. It tests four parametric geometries (ellipse, superellipse, Cassini oval, and bean curve), which are identified as those capturing a specific share of all locations visited (i.e., 95%) while minimizing the area covered. They are estimated for a number of long-duration data sets while distinguishing among trip purposes. This paper presents a flexible, easily adaptable method for calculating activity spaces of different shapes and a qualitative comparison of the four shape types on the basis of the given surveys. The choice of an appropriate shape representing an individuals activity space is highly dependent on the spatial distributions and frequencies of the locations visited by the person in the given time period.

Location Choice Modeling for Shopping and Leisure Activities with MATSim

The activity-based multiagent simulation toolkit MATSim adopts a coevolutionary approach to capturing the patterns of peoples activity scheduling and participation behavior at a high level of detail. Until now, the search space of the MATSim system was formed by every agents route and time choice. This paper focuses on the crucial computational issues that have to be addressed when the system is being extended to include location choice. This results in an enormous search space that would be impossible to explore exhaustively within a reasonable time. With the use of a large-scale scenario, it is shown that the system rapidly converges toward a systems fixed point if the agents’ choices are per iteration confined to local steps. This approach was inspired by local search methods in numerical optimization. The study shows that the approach can be incorporated easily and consistently into MATSim by using Hägerstrands time–geographic approach. This paper additionally presents a first approach to improving the behavioral realism of the MATSim location choice module. A singly constrained model is created; it introduces competition for slots on the activity infrastructure, where the actual load is coupled with time-dependent capacity restraints for every activity location and is incorporated explicitly into the agents location choice process. As expected, this constrained model reduces the number of implausibly overcrowded activity locations. To the authors’ knowledge, incorporating competition in the activity infrastructure has received only marginal attention in multiagent simulations to date, and thus, this contribution is also meant to raise the issue by presenting this new model.

Reconstructing the 2003/2004 H3N2 influenza epidemic in Switzerland with a spatially explicit, individual-based model.

BackgroundSimulation models of influenza spread play an important role for pandemic preparedness. However, as the world has not faced a severe pandemic for decades, except the rather mild H1N1 one in 2009, pandemic influenza models are inherently hypothetical and validation is, thus, difficult. We aim at reconstructing a recent seasonal influenza epidemic that occurred in Switzerland and deem this to be a promising validation strategy for models of influenza spread.MethodsWe present a spatially explicit, individual-based simulation model of influenza spread. The simulation model bases upon (i) simulated human travel data, (ii) data on human contact patterns and (iii) empirical knowledge on the epidemiology of influenza. For model validation we compare the simulation outcomes with empirical knowledge regarding (i) the shape of the epidemic curve, overall infection rate and reproduction number, (ii) age-dependent infection rates and time of infection, (iii) spatial patterns.ResultsThe simulation model is capable of reproducing the shape of the 2003/2004 H3N2 epidemic curve of Switzerland and generates an overall infection rate (14.9 percent) and reproduction numbers (between 1.2 and 1.3), which are realistic for seasonal influenza epidemics. Age and spatial patterns observed in empirical data are also reflected by the model: Highest infection rates are in children between 5 and 14 and the disease spreads along the main transport axes from west to east.ConclusionsWe show that finding evidence for the validity of simulation models of influenza spread by challenging them with seasonal influenza outbreak data is possible and promising. Simulation models for pandemic spread gain more credibility if they are able to reproduce seasonal influenza outbreaks. For more robust modelling of seasonal influenza, serological data complementing sentinel information would be beneficial.

Modeling station-based and free-floating carsharing demand

Carsharing, in any form, is still growing around the world. One of the effects is the increasing number of cities in which multiple carsharing operators are competing. The carsharing industry has never been as competitive as it is now: the present is a good time for researchers to invest efforts in providing tools for the assessment and planning of carsharing programs. Nevertheless, efforts in this direction are still scarce, in particular for some of the newest forms in which carsharing has been implemented, such as free-floating carsharing. This paper reports on a study that made use of MATSim, an agent-based simulation software that had already been used to model station-based carsharing, to evaluate different carsharing scenarios for the city of Berlin. The main findings are the existing high potential to extend carsharing services further in Berlin and the apparent complementarity of station-based and free-floating carsharing. On the methodological level, the work introduces a new tool for the modeling of free-floating carsharing along with improvements of the previously existing station-based carsharing model.

PRELIMINARY RESULTS OF A MULTI-AGENT TRAFFIC SIMULATION FOR BERLIN

This paper provides an introduction to multi-agent traffic simulation. Metropolitan regions can consist of several million inhabitants, implying the simulation of several million travelers, which represents a considerable computational challenge. We reports on our recent case study of a real-world Berlin scenario. The paper explains computational techniques necessary to achieve results. It turns out that the difficulties there, because of data availability and because of the special situation of Berlin after the reunification, are considerably larger than in previous scenarios that we have treated.

Route choice sets for very high-resolution data

With the increasing use of global positioning systems (GPS) in transport survey analysts are facing numerous new possibilities to model transport behavior but also several new challenges. For instance, information about chosen routes is now available with a high level of spatial and temporal accuracy. However, advanced postprocessing is necessary to make this information usable for route choice modeling. Out of the many research issues, this paper focuses on the generation of choice sets for car trips extracted from GPS data. The aim is to generate choice sets for about 36,000 car trips from 2,434 persons living in and around Zurich, Switzerland, on the Swiss Navteq network, a very high-resolution network. This network resolution is essential for an accurate identification of the chosen routes. However, it substantially increases the requirements for the choice set generation algorithm regarding the performance as well as the choice set composition. This paper presents a route set generation based on shortest path search with link elimination. The proposed procedure combines a \emph{Breadth First Search} with a \emph{topologically equivalent network reduction} to ensure a high diversity between the routes as well as computational feasibility for large-scale problems like the one described above. To demonstrate the usability of the algorithm, its performance and the resulting route sets are compared to those of a stochastic choice set generation algorithm.

Performance improvements for large-scale traffic simulation in MATSim

In contrast to aggregated macroscopic models of traffic simulation, multi-agent microscopic models, such as MATSim, enable modeling of individual behavior and facilitate more detailed traffic analysis. However, such detailed modeling also leads to an increased computational burden, such that simulation performance becomes critical.