SimRisk: An Integrated Open-Source Tool for Agent-Based ...

2009b], and model-checking-based formal analysis technique[Tan and Xu, 2009a]. We also developed
a prototype of a modeling and simulation tool for supply-chain analysis [Tan and Xu, 2009a].
These accomplishments have prepared us to take on this project. This grant will provide crucial
resources we need to advance and complete the research we started in our preliminary study. If
funded, this project will leverage the benefits of recent advances in computer science, especially
in software engineering and formal methods and apply them to stochastic supply-chain analysis.
The project will also advance theories and methods crucial for modeling and analysis of largescale
stochastic supply chains. Finally the project will produce an open-source tool Simrisk as a
platform for delivering new analysis and optimization technologies to practitioners and researchers.
Simrisk will also serve as an open platform and technology testbed that allows other researchers to
experiment new technologies. If successful, this project will empower researchers and practitioners
with technologies and an open-source tool that are scalable for analyzing large-scale global supply
chains under uncertainty. With this new capability, next we will apply Simrisk to supply-chain
risk management, contracting, and performance evaluation. As the follow-up of this project, we
will team with industrial leaders such as Boeing, which we already have collaboration with on
supply chain consulting, and research institutions such as Pacific Northwest National Lab (PNNL),
which we are collaborating with on high-performance simulation technology. We will apply the
methods and the tool produced in this project to the analysis of intercontinental supply chains
using Peta-scale computing platforms.
5 Research plan
5.1 Specific aim 1: develop an agent-based stochastic supply-chain modeling
framework
The framework will model elements of a supply chain as automatous stochastic agents. It will also
formally define operational semantics of these agents and their interactions. This is the start point
of this project since other activities rely on this model language as the front end. Specifically we
will address the following issues,
1. Agent modeling. This activity will consider how to model supply chain elements as autonomous
agents. Since our emphasis is on stochastic behaviors of these elements and each
element has its own decision logic, we will model agents as Markov decision processes. In [Tan
and Xu, 2008] we proposed an extension of Markov decision process (EMDP) and modeled
each element as a restricted two-state EMDP. For example, every element in a 4-echelon supply
chain in Figure 1.(a) has only two states: working and failed. In this project, we plan to
lift such restriction. We will encode more complex decision logic for each element. Moreover,
we will make the following advances in agent modeling for supply chains:
(a) Extend E-EMDP (Element Extended Markov Decision Process) in [Tan and Xu, 2008] to
include logic that models the decision process of an element. For instance, for warehouse
w11 a in Figure 1.(a), the proposed extension will also support the encoding of its ordering
and distribution logic. The ordering logic of w11 a will decide when and where to place
orders based on factors including w11 a ’s inventory, the pricing structures of its suppliers
s a and s b , and requests from its customers w21 a and wa 22 .
7