Lean Manufacturing

MODELING FOOD SUPPLY CHAINS USING MULTI-AGENT SIMULATION

Caroline C. Krejci

Benita M. Beamon

University of Washington

Seattle, WA 98195, USA

ABSTRACT

In light of the pressures of increasing demands on earth’s resources, society faces serious challenges in

food production and distribution. Food supply chain (FSC) models are critically important, providing decision-

makers with tools that allow for the evaluation and design of FSCs, en route to ensuring sustainable

FSC productivity. Multi-agent simulation (MAS) is well-suited to modeling FSCs for this purpose,

enabling capture of decision-making, interactions, and adaptations of autonomous FSC actors. However,

certain characteristics of FSCs are particularly difficult to model in detail, as data requirements can be intensive.

In this paper we highlight some of the challenges modelers face in deciding the most appropriate

methods for representing the elements of an FSC in an MAS model. We provide examples from the literature

that show how other modelers have chosen to address these challenges. Finally, we discuss benefits

and limitations of each example’s approach, in terms of realism and data requirements.

1 MODELING FOOD SUPPLY CHAINS WITH MULTI-AGENT SIMULATION

Food supply chains (FSCs) range widely in size and complexity, from subsistence farmers growing their

own food to city-dwellers purchasing groceries from a supermarket. Because of food’s vital importance

to survival, and the multitude of pressures exerted on these systems, methods for producing food more efficiently

are an important area of study. One such method to improve food production efficiency is mathematical

modeling. FSC models are now potentially more useful than ever before, as human beings face

serious challenges with food production and distribution. Worldwide demand for food is growing, but issues

such as energy and water resource limitations, agricultural pollutants, and climate change constrain

our ability to increase food production. FSC models can help us face these challenges by improving our

ability to make decisions that support long-term human and environmental well-being. However, to be

useful, FSC models must balance tractability with the ability to realistically capture the essential elements

of FSCs.

Mathematical optimization is the most common method of modeling the food production stage of an

FSC. Many existing agricultural optimization models are static, deterministic linear programming (LP)

models with the single objective of maximizing farm income or profit, subject to constraints of farm input

costs and/or availability. However, very few of these models are able to capture stochastic or dynamic elements

of FSCs, and most of these models only analyze a single stage of the FSC – food production.

Food systems have also been modeled using discrete-event simulation. While discrete-event simulations

can explicitly model time dynamics and stochastic behavior, they are incapable of modeling the sociological

processes that influence decision-making by individual FSC actors (Higgins et al. 2010). To capture

the dynamic, stochastic, and multi-faceted elements of a FSC, recent research suggests that FSCs be modeled

as complex adaptive systems (Meter 2006, Higgins et al. 2010). A complex adaptive system (CAS)

is a system of interconnected autonomous entities that make choices to survive and, as a collective,

evolves and self-organizes over time (Pathak et al. 2007). Thus a CAS framework can be used to study

an FSC. Multi-agent simulation (MAS) is a modeling tool that can effectively model the heterogeneous,

997788--11--44667733--44778802--28//1122//$$3311..0000 ©22001122 IIEEEEEE 1167

Krejci and Beamon

autonomous, intelligent, and interacting actors that comprise a CAS, making MAS a particularly appropriate

tool for modeling an FSC.

This paper seeks to highlight some of the challenges that modelers face in deciding the most appropriate

methods for representing the

...