Lecture: Agent Based Modeling in Transportation Lecturers: Dr. Francesco Ciari Dr. Rashid Waraich Assistant: Patrick Bösch Autumn Semester 2014

Lecture I September 16th 2014

Lecture Structure Theory Practice Paper Modeling Transport Agent Based Modeling Multi Agent Transport Simulation (MATSim) Practice Case studies (individual or in small groups) Paper The expected output is a case study report in the form of a proper scientific paper

Modeling transport(ation)

Modeling transportation Transportation: ??? Model: ???

Modeling transportation Transportation: is the movement of people, animals and goods from one location to another (Wikipedia) Model: ???

Modeling transportation Transportation: is the movement of people, animals and goods from one location to another (Wikipedia) Model: A simplified representation of a part of the real world which concentrates on certain elements considered important for its analysis from a particular point of view (Ortuzar and Wilumsen, 2006)

What for? Planning (i.e. infrastructure, systems) Policy making Type of model depends on: Decision making context Accuracy required Data Resources

Activity based paradigm

Transportation Transportation: is the movement of people, animals and goods from one location to another

Transportation Transportation: is the movement of people, animals and goods from one location to another

Transportation Transportation: is the movement of people, animals and goods from one location to another What are the reasons of this movement?

Activity approaches Activity approaches means «The consideration of revealed travel patterns in the context of a structure of activities, of the individual or household, with a framework emphasizing the importance of time and space constraints. (Goodwin, 1983)

Activity approaches Allow looking at important aspects of travel like: Activity Generation In home/out of home activities (patterns, substitution) Constraints Scheduling Social Networks (Kitamura, 1988)

Modeling with agents

What is an agent? An agent: Agents are: Has a set of attributes/characteristics Follows given behavioral rules Has decision making capability Is goal oriented Acts in an environment and interacts with other agents Is autonomous Can learn Agents are: Heterogeneous Attributes can change dynamically (Source: Macal and North, 2005)

Agent Attributes Behavioral rules Decision making Memory Agent-based modeling is a new approach to modeling systems comprised of autonomous agents. Each agent has a set of characteristics and rules governing its decisions-making capabilities. An agent is goal-oriented, having goals to achieve with respect to its behaviors. However, an agent is flexible, and has the ability to learn and adapt its behavior over times based on experience. This requires some form of knowledge.

Agent-based modeling Environment … Moreover, an agent is situated, living in an environment in which... Environment

Agent-based modeling … … … ...it interacts with other agents. Agents have some kind of protocols for interaction with other agents, and the capability to respond to the environment. Agents have the ability to recognize and distinguish the traits of other agents. …

Agent-based modeling ...it interacts with other agents. Agents have some kind of protocols for interaction with other agents, and the capability to respond to the environment. Agents have the ability to recognize and distinguish the traits of other agents.

Agent-based modeling Finally, there might be various types of agents in an agent-based modelling systems, which are interacting among each other.

Agent-based modeling

Agent-based modeling The actors of the (real) system modeled are represented at indivudual level and implement simple rules. The behavior of the system is not explictly modeled but emerges from the simulation

Agent-based modeling The actors of the (real) system that is modeled are represented at indivudual level and implement simple rules. The behavior of the system is not explictly modeled but emerges from the simulation Simple rules implemented at the micro-level (individual) allows modeling complex behavior at the macro-level (system)

Pros and cons Pros: Models Individuals Agents heterogeneity Emergent behavior Can deal with complexity Cons: Data hungry Skilled users

Why Agent-Based Modeling is becoming popular? Increasingly complex world Availability of high resolution level data Computer power

What about transportation?

Traditional Modeling Approach Four steps model

Four Step Process Trip generation Trip distribution Mode choice Define number of trips from and to each zone. Trip distribution Define for each zone where its trips are coming from and going to. Mode choice Define transport mode for each trip. Route assignment Assign a path to each route. 34

Four Step Process – Trip Generation 35

Four Step Process – Trip Distribution 36

Four Step Process – Mode Choice 37

Four Step Process – Route Assignment 38

Four Step Process – Facts Traditional approach in transport planning Simple, well known and understood Sequential execution Feedback not required, but desirable Aggregated Model No individual preferences of single travelers Only single trips, no trip chains Static, average flows for the selected hour, e.g. peak hour 39

Iterative Four Step Process Improvement of the traditional approach Iterations allow feedback to previous process steps Still an aggregated model 40

Modern Modeling Approaches Activity-based demand generation Dynamic traffic assignment

Activity-based demand generation Models the traffic demand on an individual level. Based on a synthetic population representing the original population. For each individual a detailed daily schedule is created, including descriptions of performed… …activities (location, start and end time, type) …trips (mode, departure and arrival time) Activity chains instead of unconnected activities and trips. Represents the first three steps of the 4 step process.

Activity-based demand generation Spatial resolution can be increased from zone to building/coordinate. High resolution input data is required such as… …the coordinates of all locations where an activity from type X can be performed. …the capacity of each of this locations. Examples of activity-based models ALBATROSS (A Learning-Based Transportation Oriented Simulation System) TASHA (Travel Activity Scheduler for Household agents)

Dynamic Traffic Assignment Supports detailed description of the demand (persons/households). Based on trip chains instead of single trips. Time dependent link volumes replace static traffic flows. Spatial and temporal dynamics are supported. Represents the fourth step of the 4 step process.

Dynamic Traffic Assignment Typical implementations are simulation based. Iterative simulation and optimization of traffic flows in a network on an individual level. Examples of DTA implementations DYNAMIT (Ben-Akiva et.al.) DYNASMART (Mahmassani et.al.) VISSIM (PTV; only small scenarios) TRANSIMS

State of the art Fully agent-based approach Combination of activity-based demand generation and dynamic traffic assignment

Fully Agent-based Approach Combines the benefits of activity-based demand generation and dynamic traffic assignment. Replaces all steps of the four step process. During the whole process, people from the synthetic population are maintained as individuals. Individual behavior can be modeled!

Macro-Simulation vs. Micro-Simulation Based on aggregated data Flows instead of individual movement Often planning networks Micro-Simulation Population is modeled as a set of individuals Traffic flows are based on the movement of single vehicles (or agents) and their interactions Various traffic flow models, e.g. cellular automata model, queue model or car following model Often high resolution networks (e.g. in navigation quality) 48

Introduction to MATSim

MATSim at a glance Implementation of a fully agent-based approach as part of a transport modeling tool Disaggregated Activity-based Dynamic Agent-based Open source framework written in java (GNU License) Started ~10 years ago, community is still growing Developed by Teams at ETH Zurich, TU Berlin and senozon AG www.matsim.org MATSim provides a toolbox to implement large-scale agent-based transport simulations. The toolbox consists of severel modules which can be combined or used stand-alone. Modules can be replaced by own implementations to test single aspects of your own work. Currently, MATSim offers a toolbox for demand-modeling, agent-based mobility-simulation (traffic flow simulation), re-planning, a controler to iteratively run simulations as well as methods to analyze the output generated by the modules. Multiagent traffic simulation toolkit, fast microscopic transport model. Each traveller of the real system is modelled as an individual agent in the simulation. Supply side is modelled as fixed constraints of the system. The agents are able to take decisions, according to given information and coherently with a predetermined goal. Agents are able to learn. Transport is a derived necessity in relation with the primary need of individuals to perform certain activities during the day.

Working with MATSim… Users Super-users Developers Black-box use Add new features Developers Add new fundamental features

Working with MATSim… Users Super-users Developers Black-box use Add new features Developers Add new fundamental features

MATSim Optimization Loop Optimization is based on a co-evoluationary algorithm Period-to-period replanning (typically day-to-day) Each agent has total information and acts like homo economicus Approach is valid for typical day situations

MATSim – Scenario Creation A MATSim scenario contains some mandatory as well as some supplementary data structures Mandatory Network Population Supplementary Facilities Transit (Schedule, Vehicles) Counts

55 Road network High resolution navigation network, including turning rules

56 Day-plan 7:30 7:40 7:50 7:56 17:03 17:09 17:13 17:25 17:45 17:55 19:24 19:31

Speed vs Resolution physical (VISSIM) Resolution CA (TRANSIMS) parallel Q event (MATSIM) Q (Cetin) Q event (MATSIM) meso (METROPOLIS) macro (VISUM) Speed

Facilities „Facilities“: Building location Activity options 58 Facilities „Facilities“: Building location Activity options Capacity, Opening time Source: Enterprise register, Building register

Performance - Scenario 59 Performance - Scenario Transportation system in Switzerland 24 h of an average Work-day 5.99 Mio Agents 1.6 Mio Facilities for 1.7 Mio Activities (5 Types) Navigation network with 1.0 Mio Links 4 Modes (others optional  i.e. shared modes) 22.2 Mio Trips Routes-, Time-, (Subtour-)Mode- und „Location“-Choice  One Iteration in ca. 4.5 hours 5.99 Mio Agenten = 88% der Gesamtbevölkerung Verkehrsmittel: MIV, ÖV, Fahrrad, zuFuss, Mitfahren 22.2 Mio Wege = 3.7 Wege/Agent Aktivitätentypen: zu Hause, Arbeit-Sektor 2, Arbeit-Sektor 3, Einkaufen, Freizeit, Schule-Kindergarten, Schule-Pflicht, Schule-Erweitert, Schule-Universität und Schule, und Übrige

Current research themes (I) Simulation of public transport Improved routing, multimodal simulation Replanning improvement Reduce the number of iterations, add other choice dimensions Simulation of traffic lights and lanes Focus on adaptive signal-control Queue simulation Parallelization Modeling of vehicle fleet Calculation of emissions Electric vehicles Simulation of the use of electric vehicles

Current research themes (II) Agents coordination Simulation of joint plans Parking Improvement of parking choice and search Introduction of land-use Integration with UrbanSim Location choice of retailers Addition of supply-side agents Car-sharing Car-sharing as an additional modal option Weather impacts Modeling of weather and climate change effects

Current scenarios Zurich and Switzerland Switzerland 7,6 Mio Agents Navigation road network with 1 Mio Links Berlin, Germany Singapore Gauteng, South-Africa Sioux Falls, USA Munich, Germany Germany/Europe – Main road network Padang, Indonesia Tel-Aviv, Israel Kyoto, Japan Toronto, Canada Caracas, Venezuela Tel Aviv, Israel Switzerland Berlin and Munich, Germany Toronto, Canada Gauteng, South Africa

MATSim Singapore 60FPS NEW TITLES.mkv (author: Pieter Fourie)

Possible Case Study Themes Carsharing Electric Vehicles Weather

Questions Laptop? Windows Mac

Additional Literature Bhat, C. R., J. Y. Guo, S. Srinivasan and A. Sivakumar (2004) A comprehensive econometric microsimulator for daily activity-travel patterns, Transportation Research Record, 1894, 57-66. Kitamura, R. (1988) An evaluation of activity-based travel analysis, Transportation, 15 (1) 9–34. Macal, C. M. and M. J. North (2005) Tutorial on agent-based modeling and simulation, Proceedings of the 37th Conference on Winter simulation, Orlando, December 2005. Mahmassani, H. S., T. Hu and R. Jayakrishnan (1992) Dynamic traffic assignment and simulation for advanced network informatics, in N. H. Gartner and G. Improta (eds.) Compendium of the Second International Seminar on Urban Traffic Networks. Ortuzar, J. D. D. and L. G. Willumsen (2006) Modelling Transport, John Wiley & Sons, Chichester.