Multi-Agent Modeling for Complex Systems
Multi-Agent Modeling (MAM) or Agent-Based Modeling (ABM) is a form of computational modeling and artificial intelligence where agents operate in an environment based on a strictly defined set of rules. With ABM we can replicate the behavior that would normally require highly complicated equations by defining agents and a simple rule set. For example, modeling fluid dynamics would normally require differential equations, but computationally with ABM, you can set water molecules as agents with a velocity in a certain direction based on aggregating of the surrounding agents. Shown in this paper:
Here are examples of ABM frameworks where agents interact with an environment based on a rule set:
Learning agents are capable of changing behavior to optimize for a specific goal while
Learning Agent
Simple Agent
Applications: Chaos and Complexity
One of the most useful applications of agent-based models is the ability to replicate complex systems. Complex systems are systems with different components that interact in a way where predicting the system's behavior based on its properties ranges from extremely difficult to impossible. An example of a complex system is shown on the Chaos Theory Wikipedia page of a double rod pendulum. The swing of a single-rod pendulum is highly predictable by mathematical models. At a time t, we can predict with a high degree of certainty the position of a single rod pendulum given the starting position. However, as soon as another rod is added at the base of a single rod pendulum the position becomes close to impossible to predict even with all the components and initial environmental conditions known. The movement of the pendulum exhibits pure chaos.
Animation of a double rod pendulum
With Agent-Based Modeling, we can set each rod as an "Agent" and define an environment ruleset that causes the rods to follow basic physics rules, and we can model the double rod pendulum computationally with a few lines of code. While the modeling does not allow us to perfectly predict the behaviors of the double rod pendulum it allows us to run large quantities of simulations that give us useful insights into repeated behaviors over a large simulation sample. These common behaviors are called "emergent phenomena," which are unexpected behaviors that emerge out of the unpredictability of complex systems.
For more information visit this link as it describes the field of complex systems well.
Here are some examples of interesting research and applications of Agent-Based Modeling:
Southwest Airlines used an agent-based model to improve how it handled cargo (Seibel and Thomas, 2000).
Eli Lilly used an agent-based model for drug development (Bonabeau, 2003a).
Pacific Gas and Electric: Used an agent based model to see how energy flows through the power grid (Bonabeau, 2003a).
Procter and Gamble used an agent-based model to understand its consumer markets (North et al., 2010) while Hewlett-Packard used an agent-based model to understand how hiring strategies effect corporate culture (Bonabeau, 2003b).
Macy’s have used agent-based models for store design (Bonabeau, 2003b).
NASDAQ used and agent based model to explore changes to Stock Market's decimalization (Bonabeau, 2003b; Darley and Outkin, 2007).
Using a agent-based model to explore capacity and demand in theme parks (Bonabeau, 2000).
Traffic and pedestrian modeling (Helbing and Balietti, 2011).
Disease dynamics (e.g. Eubank et al., 2004).
Agent-based modeling has also been used for wild fire training, incident command and community outreach (Guerin and Carrera, 2010). For example SimTable was used in the 2016 Sand Fire in California.