An agent-based simulation model supports “complex decision systems where a system or network is modeled as a set of autonomous decision-making units called agents that individually evaluate their situation and make decisions on the basis of a set of predefined behavior and interaction rules.”[1] This modeling technique uses a bottom-up approach, targets dynamic systems, and employs adaptive learning versus optimization. A distinguishing characteristic is the capture of the results of system component interactions – something you might witness observing a flock of birds as it moves seemingly erratically across the sky.
This definition is, oddly enough, reminiscent of every James Bond movie. Our agent, James, exists in a tremendously chaotic global environment which continues to expand geographically (across the globe and in space) and politically (with the introduction of Spectre and other nefarious secret organizations), and in this environment he interacts with other agents and characters (variables), all very irregular (Goldfinger, Odd Job, Jaws, 006, etc.) while adhering to a flexible but fixed set of predefined behavior and interaction rules (from MI-6 and the rules of the gentleman).
The various factors that were fed into the agent-based simulation model included but are not limited to: the individual's’ age, residence, and level of general interaction with other members of the population (types of contact), number of individuals per household, household distribution, and behavioral aspects such as daily commutes, school attendance, and asymptomatic time period of the disease.[1] A generalizing assumption was also utilized, that being “every individual of working age used the nearest subway line to travel”.[1] This could have been more properly determined with a survey or other statistically sound method.
The benefits of using agent-based simulation models include their accounting for and positive exploitation of “real-world” considerations, the ability they provide to include the details and the interactions between those details that occur in every other than simple system. Granted, this technique could be considered inappropriate in a process where four Lego® blocks, each a blue cube with four studs on the top, are assembled together, one on top of the other, in a temperature- and humidity-controlled environment with consistent appropriate lighting, by personnel of the same age group and demographic, and the same amount of training, experience, and level of proficiency. Yes, this has an air of sarcasm about it and is intended to be somewhat humorous, but seriously, traditional linear models are seldom appropriate. But that is not typically the question at hand – it comes down to “what can we afford to do (data, time, personnel, intellect, and money)?” Unfortunately, the cost of not utilizing the best model leads to loss in various categories.
Agent-based simulation models could be employed in scenarios where populations are not made up of homogenous/identical individuals existing within a stable environment, and wherein segmentation or subpopulation modeling would be effective for mitigation strategy selection. Examples might include traffic patterns, social issues, and even operations management.
We are currently working with a major seafood provider, and agent-based modeling could be a tremendous asset in improving this client’s maintenance program. Different equipment technicians of different ages from different demographics with various technical capabilities working on machines of various makes, models, and ages, with unstable repair histories (the correct part was not always used – just the part that would keep the machine running), unpredictable production runs (beyond two weeks), etc. make the situation ripe for this type of simulation.
A similar effort was utilized in a study at Air Force Materiel Command (yes – it’s “materiel”) at Wright-Patterson Air Force Base (yes – where they supposedly keep the aliens), to “investigate the effects of differing levels of maintenance manning on sortie production capability, while examining those effects on the resulting Combat Mission Readiness (CMR) of a typical F-16 squadron.”[2] Roughly translated – the Air Force wanted to know how to maximize sorties (or aircraft deployments) with the fewest maintenance technicians possible. The application of this bottom-up approach in these situations might be considered revolutionary, as top-down methods have typically been used with grandiose assumptions (e.g., all technicians repair all equipment at the mean repair time every time). However, in dealing with the uncertainty and the numerous variables, levels, and interactions, the agent-based simulation model makes more sense.
References:
1. Sharda, R., Delen, D., and Turban, E. (2015). Business Intelligence and Analytics: Systems for Decision Support. Boston: Pearson.
2. Adam MacKenzie, et al. An Exploration of the Effects of Maintenance Manning on Combat Mission Readiness Utilizing Agent Based Modeling. Retrieved from: http://www.informs-sim.org/wsc10papers/127.pdf
Kommentare