A framework for investigating the impact of PHEVs
Paper will be presented at IAMF 2009


The paper proposes a framework to investigate impacts of wide-scale Plug-In Hybrid Electric Vehicle (PHEV) utilization on power- and transport systems. It consists of two interacting parts, a power systems- and a transport simulation. The latter one finds an equilibrium representing user behaviour as an input for power system simulations incorporating distribution power network bounds. These results are fed back for renewed transport simulation until system convergence.

Wirkungen der Westumfahrung Zürich: Eine Analyse mit einer agentenbasierten Mikrosimulation



Im Jahr 1971 ergänzte die Bundesversammlung das Nationalstrassennetz mit der nordwestlichen Umfahrung von Zürich (N20). Am 13. September 1996 wurde schlussendlich mit dem Spatenstich für den Abschnitt Umfahrung Birmensdorf der Baubeginn der Westumfahrung von Zürich gefeiert. Eines der erklärten Ziele dieses Werkes ist es, den Verkehr durch die Stadt Zürich, im Speziellen entlang derWesttangente, auf den Ring zu verlagern. Dazu sind verschiedene bauliche und verkehrslenkende Veränderungen, so genannte flankierende Massnahmen, geplant. Im Rahmen dieses Projektes werden die Auswirkungen jeder einzelnen Massnahme, sowie auch deren Kombinationen anhand einer agentenbasierten Mikrosimulation auf verkehrsplanerische Ziele für die Grossregion Zürich untersucht. Der Fokus wird dabei auf folgende Massnahmen gelegt:

* Bau der Westumfahrung und des Autobahnabschnitts A4
* bauliche Massnahmen Westtangente / Aussersihl: Rückbau der Nord-Süd-Achse Weststrasse zur Quartierstrasse, Gegenverkehr auf der bisherigen Süd-Nord Achse Seebahnstrasse
* Verkehrsbeeinflussung durch Anpassungen der Grünphasenanteile der Lichtsignalanlagen in Wollishofen

Als weiterer Aspekt werden die Auswirkungen auf den Verkehr der “Situation Total-Sanierung Uetlibergtunnel” analysiert, unter der Annahme, dass die oben genannten Massnahmen umgesetzt sind. Die Analysen der hier präsentierten Fallstudien werden dynamisch über den gesamten Tagesverlauf betrachtet. Zudem werden speziell diejenigen Bevölkerungsgruppen im Detail analysiert, die direkt von den Massnahmen betroffen und/oder in ihrem individuellen Mobilitätsverhalten beeinflusst werden. Die Resultate dienen den Projektverantwortlichen zur Beurteilung der flankierenden Massnahmen in der Stadt Zürich.

A new mode choice model for a multi-agent transport simulation
Paper presented at the 8th Swiss Transport Research Conference, Ascona, October 2008


This paper reports the development of a new mode choice model system, embedded in a multiagent transport simulation, aiming to obtain both a spatially and behaviourally fine representation of modal choice of the simulated individuals. The new model addresses the subtour level, substantially improving on previous work, which either addresses the trip or the tour. A subtour is any sequence of trips which starts and ends at the same location; a tour starts and ends at home. The subtour resolution allows us to consistently account for the differential availability of modes. Subtour mode choices are otherwise treated as independent events. The subtour level allows us to capture the noon peak behaviour appropriately. The model system has two stages both formulated as multinomial logit models. First, the mobility tool ownership is estimated for each agent; second, given this choice the mode choice is addressed at the sub-tour level of the individual daily activity chain. We considered car ownership and public transport season ticket ownership and their possible combinations. The subtour models for walk, bicycle, car, car passenger and public transport are estimated by subtour's main purpose to account for taste differences. The attributes of alternatives' utility functions are socio-demographic characteristics of individuals, such as age, income, mobility tool ownership, etc., and distance covered. The mode choice module is integrated in the simulation toolkit MATSim-T (Multi-Agent Transport Simulation Toolkit). The module has been tested with a large scale scenario,based on Swiss data, and proved able to reproduce real modal choices of the population with a fine spatial resolution. Results are presented for a scenario in which the whole Swiss population is simulated. They are compared to the most recent Swiss national travel diary survey.

Location choice modeling for shopping and leisure activities with MATSim: Combining micro-simulation and time geography
Arbeitsberichte Verkehrs- und Raumplanung, 527


This paper presents the concept and implementation of the MATSim (Multi-Agent Transport Simulation Toolkit) location choice module for shopping and leisure activities and shows first simulation results for the Zürich region of Switzerland using more than 60,000 people and 7,800 shop or leisure activity locations. MATSim is designed to handle large-scale scenarios. Thus, computational efficiency, that is fast convergence to a Nash Equilibrium, while maintaining behavioral precision is a fundamental objective. We show that to achieve this goal Hägerstrand’s time-geographic approach can be incorporated easily and consistently into MATSim and into disaggregated location choice simulations in general. Our novel time-geographic algorithm, tailored for the use in MATSim, is derived from a potential path area algorithm. It is extended to handle chains of multiple shop or leisure activities between two anchored activities (i.e., those with fixed start times, durations and locations) by using recursion. To improve both the behavioral precision and the stability of our model, we show how timedependent capacity restraints can be incorporated explicitly into iterative disaggregated location choice simulations, as capacity restraints for activity locations have an effect on people’s location choices, similar to the effect of road capacity restraints on people’s route choices. To our knowledge, to date, only static activity location attributes, such as opening hours and location size, are incorporated explicitly in disaggregated location choice simulations and thus our contribution is also meant to open up a discussion.

Concepts for a large scale car-sharing system: Modelling and evaluation with an agent-based approach
Arbeitsbericht Verkehrs und Raumplanung, 517


This paper is aimed to renew the debate on car-sharing and its future development. The recent worldwide success should not hide the fact that car-sharing is still a niche product. There is agreement that car-sharing produces benefits for the transport system, the environment and the society. However, the scale of such benefits is minimal. This is a reason to attempt the implementation of a system at a much larger scale. A car-sharing scheme of this type is sketched and some concepts on which the system would be based are suggested. They are the capillarity of the system, its flexibility and its integration with other urban mobility tools. For the future implementation of such a scheme it is crucial to find a methodology which would be able to realistically assess its potential. Reviewing the methodologies used so far to investigate car-sharing potential suggests that the adoption of an agent based approach might be the right answer. A framework to model a large scale carsharing scheme with such methodology is proposed in the context of MATSim-T, an existing agent based traffic micro-simulation tool.

A High-Performance Traffic Flow Microsimulation for Large Problems
Working paper, 509


Traffic flow microsimulations are interesting for transport planning problems due to their high temporal and spatial resolution. Unfortunately, most of them involve high computational costs making them impractical for running large scale scenarios. In this paper, we present how we extend our previous event-driven queue-based mircosimulation to run efficiently on parallel computers. Using appropriate load balancing and minimizing communication interfaces, we are able to simulate a test scenario involving 7 million simulated person days on a road network with 28k links in 87 seconds on 64 CPUs. Furthermore, we add support for signalled intersections that makes the model well suited for application to urban street networks. Finally, we show that our resulting model reproduces a reasonable relation between traffic flow and density similar to fundamental diagrams extracted from real world counts data.

Agent-based simulation of travel demand: Structure and computational performance of MATSim-T
Paper presented at the 2nd TRB Conference on Innovations in Travel Modeling, Portland, June 2008


Dieser Aufsatz gibt eine knappe Darstellung des Modellsystems MATSim-T und einer Anwendung in Grossraum Zürich. Die Betonung liegt auf den erreichbaren Rechenzeiten und damit auf der Anwendbarkeit für Planungsstudien. Der Ausblick zeigt auf, in welchen Teilen das Modell weiter verbessert werden kann und soll.

MATSim-T: Architektur und Rechenzeiten
Paper presented at the Heureka '08, Stuttgart, March 2008


Traditionell werden in der Verkehrsplanung Meso- und Makro-Simulationen angewandt. Dies hat verschiedene Gründe. Einerseits sind die Datengrundlagen typischerweise aggregierter Natur, wie zum Beispiel Strassenzählungen oder Pendlermatrizen, andererseits waren die Rechenleistungen ungenügend, um detaillierte und demzufolge speicher- und prozessorintensive Berechnungen durchzuführen. Diese Grenzen haben sich in den letzten Jahren auf imposante Weise verschoben. In der Informationstechnik ist kontinuierlich ein exponentielles Wachstum der Rechenkapazität zu einem beliebigen festen Preis zu beobachten. Hand in Hand mit dieser Entwicklung sind die Fragestellungen und Datengrundlagen in der Verkehrsplanung komplexer und detaillierter geworden. Es entstehen somit ganz neue Ansprüche an die Verkehrsplanungssoftware. Derzeit gewinnt die Mikrosimulation der Verkehrsnachfrage immer mehr an Bedeutung.
Die wichtigsten Gründe dafür sind:
- Verringerter Rechenaufwand und Speicherbedarf von grossen, mehrdimensionalen Wahrscheinlichkeitsmatrizen
- Variationsreichere Ausgabeoptionen, von aggregierten statistischen Analysen bis hin zu detaillierten Informationen über einzelne Individuen eines Szenarios
- Explizites Modellieren des Entscheidungsfindungsprozesses jedes einzelnen Individuums.
In dieser Arbeit wird eine solche Agenten-Simulation für die Verkehrsplanung vorgestellt. Die Arbeit ist Teil des Forschungsprojektes MATSim-T (Multi-Agent Transport Simulation Toolkit, http://www.matsim.org) und konzentriert sich auf Design- und Implementationsfragen des Systems, sowie auf die Rechenzeiten der einzelnen Teile des Toolkits. Anhand des Anwendungsfalls der gesamt-schweizerischen Nachfrage (ca. 2.3 Mio. Individuen für den motorisierten Individualverkehr mit gesamthaft etwa 7.1 Mio. Wege auf dem nationalen Netzmodell mit ca. 60’000 Kanten, vollständig zeit-dynamisch simuliert und optimiert für einen durchschnittlichen Werktag) wird gezeigt, dass MATSim-T den den Verkehr in ca. 36 Stunden modelliert.

Anwendung eines agentenbasierten Modells der Verkehrsnachfrage auf die Schweiz
Paper presented at the Heureka '08, Stuttgart, March 2008


Heutige Fragen an die Verkehrsplanung wecken ein Bedürfnis zur Abschätzung der tageszeitlichen Dynamik der Verkehrsnachfrage. Agentenbasierte Modelle des Verkehrsverhaltens simulieren individuelle Aktivitätenpläne und damit explizit die Tageszeit. In diesem Artikel wird die Anwendung des Simulationssystems MATSim-T auf den Strassenverkehr der ganzen Schweiz präsentiert. Die Dynamik der Verkehrsnachfrage kann gut abgebildet werden, was ein Vergleich mit repräsentativen Beobachtungen zeigt.

Generating Daily Activity Chains from Origin-Destination Matrices
A Case Study between VISUM and MATSIM based on Kanton Zurich Data, Switzerland


Microsimulation tools are becoming increasingly important in traffic demand modeling. The major advantage in comparison with traditional assignment models lies in the fact that each traveler is simulated individually. This means, for example, that decision making processes can be modeled for each individual. The traffic demand is a result of the decisions of each individual. Those decisions lead to plans which the individuals try to execute.

Because microsimulation includes not all relevant decision making processes, in particular not scheduling, individuals' plans have to be specified externally.

On the other hand, traditional assignment tools such as VISUM (1) or EMME/2 (2) use OD-matrices as inputs. Those matrices do not include any information about the chains of activities which define plans of the individuals. The question arises, if it is possible to impute (reconstruct) the plans from the given OD-matrices. The paper presents a first approach to achieve this goal. The results compared are those from a VISUM implementation—a traditional assignment model—and of MATSIM (3)—a microsimulation model. The Zurich area of Switzerland is employed for the tests.

Mobility tool ownership and mode choice decision processes in multi-agent transportation simulation
Paper presented at the 7th Swiss Transport Research Conference, Ascona, September 2007


This paper presents mode choice decision processes in the context of MATSimT (MultiAgent Transport Simulation Toolkit), a simulation toolkit which uses the concept of Evolutionary Algorithm (EA) in order to obtain individual, daily, travel demand. The module described in this paper implements logit models and is embedded in MATSimT as a planning module. At a first stage the mobility tool ownership among the population, based on the Swiss Mikrozenzus data of 2005, is estimated with a logit model. Then a mode choice is performed at the tour level of the daily activity chain. A multiple choice among four alternatives, walk, bicycle, car, transit, is modeled. The module is integrated in the preprocess stage of MATSimT and its goal is to produce a reasonable tour based mode choice for each individual. The attributes considered in the utility functions of alternatives are sociodemographic characteristics of the population, such as age, income, employment, car ownership, etc. Even in this simple form the module represents a substantial improvement of the toolkit, increasing its ability to simulate realistic traffic patterns. The module has been tested with different, large scale, scenarios based on Swiss data. The presented results show that the module is able to reproduce modal choices consistent with that data.

Fast shortest path computation in time-dependent traffic networks
Arbeitsbericht Verkehrs- und Raumplanung, 439


In agent based traffic simulations which use systematic relaxation to reach a steady state of the scenario, the performance of the routing algorithm used for finding a path from a start node to an end node in the network is crucial for the overall performance. For example, a systematic relaxation process for a large scale scenario with about 7.5 million inhabitants (roughly the population of Switzerland) performing approximately three trips per day on average requires about 2.25 million route calculations, assuming that 10% of the trips are adapted per iteration. Expecting about 100 iterations to reach a stable state, 225 million routes have to be delivered in total.

This paper focuses on routing algorithms and acceleration methods for point-to-point shortest path computations in directed graphs that are time-dependent, i.e. link weights vary during time. The work is done using MATSim-T (Multi-Agent Traffic Simulation Toolkit) which used for large-scale agent-based traffic simulations. The algorithms under investigation are both variations of Dijkstra’s algorithm and the A*-algorithm. Extensive performance tests are conducted on different traffic networks of Switzerland. The fastest algorithm is the A* algorithm with an enhanced heuristic estimate: While it is up to 400 times faster than Dijkstra’s original algorithm on short routes, the speed up compared to Dijkstra diminishes with the length of the route to be calculated.

Truly agent-oriented coupling of an activity-based demand generation with a multi-agent traffic simulation
Paper presented at the 86th Annual Meeting of the Transportation Research Board, Washington, D.C., Jan. 2007


The typical method to couple activity-based demand generation (ABDG) and dynamic traffic assignment (DTA) are time-dependent origin destination matrices. With that coupling method, the individual traveler's information gets lost. Delays at one trip do not affect later trips.

It is, however, possible to retain the full agent information from the ABDG by writing out all agents' “plans”, instead of the OD matrix. A plan is a sequence of activities, connected by trips. Since that information is typically already available inside the ABDG, this is fairly easy to achieve.

MATSim takes such plans as input. It iterates between the traffic flow simulation (sometimes called network loading) and the behavioral modules. The currently implemented behavioral modules are route finding, and time adjustment. Activity re-sequencing or activity dropping are conceptually clear but not yet implemented. Such a system will react to a time-dependent toll by possibly re-arranging the complete day; in consequence, it goes far beyond DTA (which just does route adaptation).

Our paper will report the status of our current Berlin implementation. The initial plans are taken from an ABDG, originally developed by Kutter; to our knowledge, this is the first time that traveler-based information (and not just OD matrices) is taken from an ABDG and used in a multi-agent simulation. The simulation results are compared against real world traffic counts from about 100 measurement stations.

Optimal route assignment in large scale micro-simulations
Arbeitsberichte Verkehrs- und Raumplanung, 409


Traffic management and route guidance are optimization problems by nature. In this article, we consider algorithms for centralized route guidance and discuss fairness aspects for the individual user resulting from optimal route guidance policies. The first part of this article deals with the mathematical aspects of these optimization problems from the viewpoint of network flow theory.

We present algorithms which solve the constrained multicommodity minimum cost flow problem (CMCF) to optimality. A feasible routing is given by a flow x, and the cost of flow x is the total travel time spent in the network. The corresponding optimum is a restricted system optimum with a globally controlled constrained or fairness factor . This approach implements a compromise between user equilibrium and system optimum. The goal is to find a route guidance strategy which minimizes global and community criteria with individual needs as constraints. The fairness factor L restricts the set of all feasible routes to the subset of acceptable routes. This might include the avoidance of routes which are much longer than shortest routes, the exclusion of certain streets, preferences for scenic paths, or restrictions on the number of turns to be taken. Most remarkably is that the subset of acceptable routes can also be interpreted as a mental map of routes. c(x) L > 1

In the second part we apply our CMCF algorithms in a large scale multi-agent transportation simulation toolkit, which is called MATSIM-T. We use as initial routes the ones computed by our CMCF algorithms. This choice of initial routes makes it possible to exploit the optimization potential within the simulation much better then it was done before. The result is a speed up of the iteration process in the simulation. We compare the existing simulation toolkit with the new integration of CMCF to proof our results.

planomat: A comprehensive scheduler for a large-scale multi-agent transportation simulation
Paper presented at the 11th International Conference on Travel Behaviour Research, Kyoto, August 2006


An external strategy module for an iterative multi-agent micro-simulation of travel demand is presented. This module called planomat currently optimizes the time allocation and route choice of activity plans, which are the agent-based representation of travel demand. The module combines broad search for alternative timing decisions with an optimization procedure for a scoring function that evaluates activity plans. As part of the existingMulti-Agent Transportation SIMulation Toolbox (MATSIM-T), regional traffic systems of several 100’000 agents can be simulated. The test scenario used here is the Canton of Zurich, the biggest metropolitan area of Switzerland, with 550’000 agents. The comprehensive optimization of activity plans leads to a system relaxation within an acceptable number of 60 iterations. The quality of the time allocation optimization is shown by departure time distributions.

planomat: a comprehensive scheduler for a large-scale multi-agent transportation simulation
Paper presented at the 6th Swiss Transport Research Conference, Ascona, March 2006


An external strategy module for an agent-based micro simulation of traffic systems is presented. This module called planomat modifies activity durations and departure times of activity plans, which are the agent-based representation of travel demand. The module combines broad search for alternative timing decisions with an optimization procedure for a scoring function that evaluates daily activity plans. The module is integrated into the existing framework MATSIM, which simulates traffic systems consisting ofseveral 100’000 agents entirely on activity level. In this paper, a test version of the Canton Zurich is simulated, the biggest metropolitan area of Switzerland. Main results are relaxation of the whole simulation system to a better stationary state than in previous versions of the simulation framework. This is shown by departure/arrival time distributions. The number of required iterations was significantly reduced to 100, which is one-two orders of magnitude better than before.

Shape morphing of roundabouts using curb side oriented driver simulation
Paper presented at the 6th Swiss Transport Research Conference, Ascona, March 2005


In a traffic network, capacities of parts of the network restrict the amount of transport units which can be handled by this network. The capacity of a given traffic network element is not fixed but influenced by parameters such as number of lanes, maximum speed, weather, view horizon and so on. These parameters also define the maximum capacity of roundabouts. Special shapes of roundabouts, particularly in urban regions, may further increase or decrease their capacity. This paper investigates how the capacity of such special roundabouts can be estimated with only the curbsides of a roundabout as an input. It is also of interest to see that changes to the shape decrease the amount of space “wasted” for the traffic roundabout while the capacity remains unchanged. In this case study one special roundabout is examined: “Central” in downtown Zurich, Switzerland. The particularity of this roundabout is that it partially behaves like a roundabout but also contains two uncontrolled intersections. Due to its central position in the city, the roundabout is very busy with both individual cars and public transport vehicles. In the first part of this paper, a simulation model is described which is able to produce realistically behaving vehicles only by using information about the curb side locations of the roundabout. In the second part of the paper, the simulation changes the topology of the scenario based on the observed behavior of the vehicles. Using a feedback loop allows one to optimize the capacity of the roundabout while its spatial extents are minimized.

An improved replanning module for agent-based micro simulations of travel behavior
Arbeitsberichte Verkehrs- und Raumplanung, 303, IVT, ETH Zürich


An external strategy module for an agent-based micro simulation of traffic systems is presented. It modifies activity durations and departure times of activity plans, which are the agent-based representation of travel demand. The module combines broad search for alternative timing decisions with an optimization procedure of a utility function. The idea is to replace a replanning module that changed timing decisions randomly. Main results are relaxation of the whole simulation system to a better stationary state, and much quicker convergence. The difference in overall performance compared to the previous implementation of the replanning module is one order of magnitude.

Matching geo-coded graphs
Paper presented at the 5th Swiss Transport Research Conference, Ascona, March 2005


In transportation planning and modelling feasible transportation networks are crucial. To be useful, the networks have to fulfil certain requirements: first, the geographical locations of network elements (typically nodes and links) have to be accurate; second, the given attributes (i.e. number of lanes, length, allowed speed, and so on) of the network elements should hold correct information; and—particularly for traffic path finding algorithms—any given network should be constructed such, that every node is accessible by any other node via at least one path.
Unfortunately, in practice there is no guarantee that these three requirements are fulfilled. At the same time often many different networks are available for the same geographical region. These networks often can differ in their emphasis, resulting in differences such as the resolution of the network, the correctness of the geographical locations and the correctness / completeness of the given attributes.

To deal with this problem, one is required to match different networks of the same region so that attributes can be easily shared between the given networks. In this paper some approaches for network matching are described and compared. Unlike other approaches attributes of the nodes and links are not used as part of the matching algorithm, since they are unreliable. The problem is thus reduced to two directed graph with the addition of spatial information–a geo-coded digraph.

Performance Improvements for Large Scale Traffic Simulation in MATSim



Multi-Agent transport simulation models, e.g. MATSim have proven to be suitable for modeling microscopic demand for large scale scenarios based on planning networks. In the recent years survey methods are using technologies which provides mobility information with a much higher spatial resolution (e.g. GPS tracking). Therefore, the need to model travel demand on detailed navigation networks rises, which slows down simulation speed significantly. This paper presents methods to increase the performance of the micro simulation model of MATSim using event driven concepts as well as a parallel implementation. The performance experiments with navigation networks of Switzerland containing up to one million roads and 7.3 million agents clearly show that large-scale, multi-agent micro-simulation can also be applied on high resolution networks.

Route Choice Sets for Very High-Resolution Data



With the increasing use of GPS in transport surveys, analysts can choose from numerous new ways to model transport behaviour – but also face 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 modelling. Out of many research issues, this paper focusses on generation of choice sets for car trips extracted from GPS data. The aim is to generate choice sets for about 36,000 car trips made by 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 chosen routes. However, it substantially increases the requirements for the choice set generation algorithm in regard to performance as well as choice set composition.
This paper presents a route set generation based on shortest path search with link elimination. The proposed procedure combines a Breadth First Search with a 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.

Location Choice Modeling for Shopping and Leisure Activities with MATSim: Utility Function Extension and Validation Results



This paper presents validation results for the activity-based multi-agent transport simulation MATSim (http://www.matsim.org), where the main focus lies on the location choice module for shopping and leisure activities. Validation results are produced by simulating a 10%sample of the Swiss motorized individual traffic.
For Switzerland detailed information about home, working and education locations together with the associated trip matrices are provided by the census. Naturally this level of detail is not available for shopping and leisure trips. In MATSim so far—to create feasible activity chains—location choice for these activities was done in a preprocessing step based on a simple nearest neighbor search, which clearly leads to a systematic underestimation of the traffic volume. In this paper a two-fold shopping and leisure location choice model is presented that produces substantially better results.
First and foremost, the to-date exclusively time-based utility function for shopping activities is extended to take into account further determinants of shopping location choice, such as the store size and the stores density in a given neighborhood. In activity-based models, shopping location choice is influenced by leisure location choice, which means that a meaningful shopping location choice model requires a sound leisure location choice model. The long-term goal of MATSim is to model leisure location choice by utility maximization and by including models of social interaction. But these models are far from being productive in agent-based transportationmodels in general. Hence, we introduce hollow space-time prisms that are derived from empirical data. This approach is—to our knowledge—a novel extension of Hägerstrand’s time geography that by construction produces statistically correct leisure location choice and improves the simulation results in general.
Furthermore, the potential of MATSim to also serve as a hypothesis testing tool—besides being a planning tool—is highlighted in this paper. It is shown that MATSim provides the possibility to test models, generated by utility maximizing approaches such as e. g., discrete choice models, in large scenarios, whereby use can be made of data (e. g., count data) that is potentially qualitatively distinct from the data that were used for estimating and validating the models in question in earlier stages.

New approaches to generating comprehensive all-day activity-travel schedules



Activity-based travel demand models derive travel demand from people’s desire to pursue activities in time and space. They generate activity-travel schedules for individual travelers or homogeneous groups of travelers. Comprehensive activity-travel schedules hold information on which activities are performed, in which order, where and for how long, and which travel modes are used between the activities including corresponding routes. This paper presents PlanomatX, a new scheduling algorithm based on Tabu Search that generates comprehensively optimized all-day schedules. The paper furthermore presents a new concept of schedule recycling that significantly reduces simulation run times by re-using schedules of optimized travelers for other non-optimized travelers. Both PlanomatX and schedule recycling are part of the agent-based microsimulation MATSim (Multi-Agent Transport Simulation, http://matsim.org). MATSim’s utility function has been adapted to cope with the enhanced functionality of PlanomatX and schedule recycling. First test results on the greater Zurich scenario with more than 170,000 agents show that PlanomatX achieves significantly better optimization results than MATSim’s existing scheduling algorithms. However, it also leads to disproportional simulation run times. Schedule recycling relieves this drawback and allows for generating comprehensively optimized all-day schedules for large-scale scenarios at affordable run times.

Generating comprehensive all-day schedules: Expanding activity-based travel demand modelling



Activity-based travel demand models generate activity-travel schedules for individual travellers or homogeneous groups of travellers. Travel demand is derived from these activitytravel schedules by the fact that most activities take place at different locations and people need to travel between these. The agent-based micro-simulation toolkit MATSim implements an activity-based approach to travel demand generation for large samples. A co-evolutionary learning process assigns every agent with an all-day activity-travel schedule. The schedule holds information on which activities the agent performs, in which order, where and for how long, and which travel modes the agent uses between the activities including corresponding routes. This paper proposes a new MATSim utility function for the performance of activities, based on an asymmetric S-shaped curve with an inflection point. The new utility function can cope with a flexible number of activities in a schedule as it formulates an optimal activity duration by its functional form. It has become necessary since a new algorithm was added to MATSim’s replanning step that comprehensively optimizes schedules, including their activity chain sequences. This paper further presents a methodology to empirically estimate the parameters of the new utility function through an enhanced Multinomial Logit (MNL) model. A similarity attribute in the systematic part of the utility function allows to overcome the MNL model’s IIA property. First estimates of a limited set of parameters are presented although the results are still preliminary and ambiguous.

Plug-in Hybrid Electric Vehicles and Smart Grid: Investigations Based on a Micro-Simulation



Introduction of Plug-in Hybrid Electric Vehicles (PHEVs) could potentially trigger a stepwise electrification of the whole transportation sector. But the impact on the electric grid by electrical vehicle charging is still not fully known. This paper investigates several PHEV charging schemes, including smart charging, using a novel iterative approach. An agent based traffic demand model is used for modeling the electrical demand of PHEVs over the day. For modeling the different parts of the electric grid, an approach based on interconnected multiple energy carrier systems is used. For a given charging scheme the power system simulation gives back a price signal indicating whether grid constraints, such as maximum power output at hub transformators, have been violated. This leads to a corrective step in the iterative process, until a charging pattern is found, which does not violate grid constraints. The proposed system allows to investigate existing electric grids, whether they are capable of meeting increased electricity demand by certain future PHEV penetration. Furthermore, in the future, different types of smart charging schemes can be added into the system for comparison.

Large scale use of collective taxis: a multi-agent approach



This paper reports on ongoing work aimed to estimate the potential use of collective taxis at large scale in urban areas as a mean to mitigate congestion and social exclusion. The methodology used to assess the potential of the system is agent based modeling. An existing open source software project, called MAT-Sim-T (Multi-Agent Transport Simulation Toolkit, http://matsim.org), has been enhanced within this project in order to allow the modeling of the taxi mode. Cur-rently the way in which the taxi mode has been added is quite simple. A cost structure reflecting the implementation scheme of the taxi system has been de-fined. The simulated individuals (agents) will have this additional option and will choose it, or not, according to the generalized cost it generate for their schedules (plans). Even in this simple form the model allows for a preliminary estimation of the collective taxi potential. The results of a test case for the city of Zurich, a sce-nario with about 160’000 agents, are reported and discussed.

Simulation of Information Oriented Knowledge Models



How does knowledge about the state of a traffic system influence the actions of people within this system?
Traffic planning and management is used to optimize traffic systems, particularly with respect to their efficiency. The individual preferences of road users have to be considered and – if possible - linked with the traffic management system as well as the global aspects of usage. The individual cognition of the current traffic situation exerts an essential influence on the decision-making of road users.
Therefore the key element of this study is to understand the impact of different personal knowledge levels regarding the local and global state of road traffic. To consider these consequences, patterns of different levels of knowledge are constructed and implemented using the simulation toolkit MATSim. Using the evaluations of experimental map based simulation runs, the implications of different levels of knowledge are analyzed and examined. Besides evaluating the quality of knowledge based routes an additional focus lies on the consideration of the simulation results from the viewpoint of traffic-planning.

Agenten-basierte Simulation für location based services
Schlussbericht KTI 8443.1 ESPP-ES


Mobilität bildet den Grundpfeiler auf dem ortsbasierte Dienstleistungen (Location Based Services) ruhen. Darum ist ein präzises Abbild der Mobilität der zu erreichenden Zielgruppen für diese Art von Diensten unerlässlich. Mikrosimulationsmodelle zur Generierung von Mobilität liefern hierzu eine vollständige Mobilitätsanalyse, inklusive der Entscheidungsfindungsprozesse mobiler Individuen und somit die Verknüpfung statistisch relevanter Kenngrössen für unterschiedliche demographische und sozio-demographische Gruppierungen mit detaillierter, dynamischer Mobilität. Diese Eigenschaften bilden den Kernaspekt zur Dimensionierung von ortsbasierte Dienstleistungen. Basierend auf dem Open-Source Forschungsprojekt MATSim ist das erste, vollständig integrierte Umlegungs- und Verhaltensmodell für Zeit-, Routen-, Verkehrsmittel- und Ortswahl für die gesamte Schweiz entstanden.

Estimating the Potential of a Large Scale Car-Sharing System with an Agent-Based Microsimulation Approach



This paper reports on ongoing work aimed to estimate the potential use of car sharing at large scale in urban areas as a mean to mitigate congestion and social exclusion. The methodology used to assess the potential of the system is agent based modelling. An existing open source software, called MATSim-T (Multi-Agent Transport Simulation Toolkit, http://matsim.org), has been enhanced within this project to allow the modelling of the car sharing mode. In order to add the car sharing mode to the simulation toolkit a cost structure reflecting the implementation scheme of the system has been defined. The simulated individuals (agents) will have this additional option and will choose it, or not, according to the generalized cost it generates for their schedules (plans). The travelling time for this mode, is analogue to that for car and it is calculated on a congested network, where all cars are simulated, adding realism to the model. The results of a test case for the city of Zurich, a scenario with about 160’000 agents, are reported and discussed.

Large-scale agent-based travel demand optimization applied to Switzerland, including mode choice



This paper presents the application of the agent-based transport simulation toolkit MATSim-T to a large-scale scenario of Switzerland. The scenario is called large-scale because ca. 6 million synthetic persons, “agents”, are simulated on a high-resolution network model with >1 million links. MATSim-T is able to compute a relaxed state of the simulation system within 60 iterations of the learning-based solution procedure with regard to mode choice, car route choice and choice of activity timing. This is achieved by applying improved optimization algorithms in the replanning stage. A genetic algorithm is used for times and and mode choice optimization of activity plans, together with an efficient implementation of time-dependent shortest path search for route choice.
The improvements of the behavioral model reported in this paper are focused on the scoring function which can process individualized parameters for measuring the quality of all-day activity plans. Combined with disaggregate input data for population and land use, it was possible to build a heterogeneous and thus more realistic scenario. Furthermore, four modes of transport (car, public transit, bike, walk) are considered in the presented application. The generalized cost of the car option is determined by a queue simulation of traffic flow. In order to prove the concept of mode choice optimization in a multi-agent microsimulation, the other modes are modelled as abstract alternatives with static travel costs constant throughout the modeled average workday. It is shown how the model is calibrated against observed modal split data. The results are validated with average workday count data. Despite the simple cost structure of the mode alternatives, and due to a mode choice concept based on subtours, the observed spatial distribution of the modal split can be reproduced within ±10 percentage points per mode.