Agent-Based Modeling

Agent-based modeling and multi-agent modeling are relatively new methods that have been successful in answering many biological, social and behavioral questions such as analysis of the spread of epidemics, workforce management, and modeling consumer behavior in recent years. They share overlapping roots with gaming theory and its concepts, and adopting this type of modeling into engineering systems decision-making can create solutions to our daily problems in ways that were not possible before.

Rothchild1 uses the Oxford English Dictionary definition for “to induce” and “to deduce” to define induction as “the process of inferring a general law or principle from observation of particular instances,” and deduction as “inference by reasoning from general to particular” or “the process of deducing from something known or assumed.”

By these definitions, a deductive (top-down) reasoning is a logical process in which a conclusion (e.g., x needs food to stay alive) is derived based on the existence of multiple premises (every human needs food to stay alive, and x is a human) that should generally be true.

On the other hand, inductive (bottom-up) reasoning is a logical process based on observation in which multiple premises (e.g., we have observed 1,000 swans and all of them are white), all believed true or that have a high probability of being true, are put together to predict a specific conclusion (e.g., all or the majority of the swans are white).

Agent-based modeling and multi-agent modeling processes have been compared to the method of deduction and induction in different literature. Axelrod2 indicates: “The purpose of induction is to find patterns in data and that of deduction is to find consequences of assumptions, the purpose of agent-based modeling is to aid intuition” and, Billari, et al.,3 states: “As with deduction, agent-based modeling starts with assumptions. However, unlike deduction, it does not prove theorems. The simulated data of agent-based models can be analyzed inductively, even though the data are not from the real world as in case of induction.”

The problems with controlled experiments in real-world situations (induction) are that it may not be possible to perform some experiments, performing some experiments requires a lot of time and money, and some ethical issues may exist with performing some experiments. The main advantage of agent-based modeling (simulation) is that to reach logical conclusions where it is impossible or very hard to perform proper experiments, proper simulation can work as the best (bottom-up) alternative.

Furthermore, Macal4 explained the conditions in which an agent-based modeling can have significant advantages over conventional simulation approaches, such as when the past is no prediction of the future, or when it is important that agents adapt and change their behaviors.

In general, the main elements of each agent-based model are a set of autonomous agents defined by the modeler, an environment that agents can move around in, and a framework for simulating the agents’ behaviors, links, and interactions with each other and with the environment.

To set up the agent-based model, the modeler develops a virtual world in which agents move and interact randomly inside one of the available platforms designed to run this type of modeling. A set of characteristics and behaviors also is assigned to the agents such as size, energy level, etc. Ticks are usually the representatives of time, and after each tick, the agents’ movements or interactions are updated.

The simulation continues for a period of time, and the final result shows the changes in each agent and environment after completion of the simulation. Of course, the software has the capacity of repeating the simulation multiple times and saving the data for each round of simulation in spreadsheet format for further study and analysis by the modeler.

One of the applications of agent-based modeling is predicting people’s escape flow during a stampede or panic situation. As an example, Bonabeau5 used an agent-based model that depicted a fire escape situation in a confined space such as a movie theater or a concert hall. In this model a few assumptions were made, such as there is only one exit available, and if somebody falls and gets injured, he cannot move and will impede the flow of people trying to escape. The model was run under different conditions to try to find under which condition the outflow of people from this confined space will increase.

To his surprise, adding a column (a pillar) about 1 m (3 ft) before the exit and slightly asymmetrically from the exit (e.g., to the left of the exit), helped significantly to regulate the people flow, decreased the amount of injury, and increased the speed of evacuation. As he further explained, this situation was later tested and was backed by real-world experimentation.

Treado and Delgoshaei6 in their paper presented a simple sample of an integrated combined heat and power (CHP), absorption chiller and thermal storage system that were modeled and examined with an agent-based modeling method. They showed how agent-based modeling can be used to improve the operation of building control systems and increase the level of building energy performance and comfort, and decrease the cost of building operation.

Liao et al.,7 proposed an agent-based model of occupancy in a building with an arbitrary number of zones and occupants. The model provides a simulation of occupancy dynamics in non-emergency situations for the building, and the results are compared to ones estimated from measurement in real commercial buildings. Their developed model was shown to be successful in predicting some variables such as mean occupancy, marginal distributions of the first arrival time, continuously occupied duration, and number of transitions between occupied and unoccupied states, while some variables, such as distribution of last departure time and cumulative occupied duration, were not predicted with high accuracy.

Agent-based modeling and multi-agent modeling are promising tools that not only can offer solutions to managerial aspects of running engineering firms, but also open doors to new venues in selecting the most efficient design options and best implementing strategies for operating the buildings.


  1. Rothchild, I. 2006. “Induction, Deduction, and the Scientific Method: An Eclectic Overview Of The Practice of Science.” Society for the Study of Reproduction.
  2. Axelrod, R. 1997. The Complexity of Cooperation. Agent-Based Models of Competition and Collaboration. Princeton: Princeton University Press.
  3. Billari, F.C., et al. 2006. Agent-Based Computational Modelling: Applications in Demography, Social, Economic and Environmental Sciences. Heidelberg, Germany: Physica-Verlag.
  4. Macal, C.M., M.J. North. 2008. “Agent-based modeling and simulation: ABM Examples.” Proceedings of the 2008 Winter Simulation Conference.
  5. Bonabeau, E. 2002. “Agent-based modeling: methods and techniques for simulating human systems.” Proc Natl Acad Sci USA.
  6. Treado, S.; P. Delgoshaei. 2010. “Agent-based approaches for adaptive building HVAC system control.” International High Performance Building Conference Paper 26.
  7. Liao, C., Y. Lin, P. Barooah. 2012. “Agent-based and graphical modeling of building occupancy.” Journal of Building Performance Simulation 5(1).

“Agent-Based Modeling” was published in the February 2016 edition of the ASHRAE Journal.
Author: Javad Khazaii, PhD, PE

Javad Khazaii, PhD, PE

Javad Khazaii, PhD, PE

Senior Associate, Mechanical Engineer
Javad has over 20 years of mechanical engineering design and project management experience with commercial and institutional projects and has significant experience with health care, industrial, academic and commercial clients, both nationally and abroad. Email Javad
Javad Khazaii, PhD, PE
Javad Khazaii, PhD, PE

Latest posts by Javad Khazaii, PhD, PE (see all)

Share This:

Comments ( 0 )

    Leave A Comment

    Your email address will not be published. Required fields are marked *