Supporting Demand Supply Network Optimization with Petri Nets - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 368

net related methods (vad der Aalst, 1998a; van der Aalst, 1998b; van der Aalst & ter Hofstede, 2005;
Desel & Erwin, 2000). Each of these methods fits into one of four categories: deterministic analytical,
stochastic analytical, economic or simulation (Beamon, 1998). The six methods are briefly described
along with their position in Beamon's categorization.
Operations research has been used to analyze demand supply network problems at least since the early
1970's (Zeng & Rossetti, 2003; Thomas & Griffin, 1996). These methods have been used primarily in
plant and inventory location problems (Bowersox & Closs, 1989). Optimization methods were the first
technique to be used in demand supply network analysis, where the focus was on cost reduction given a
static customer service level (Bowersox & Closs, 1989; p. 139-40). The first mathematical programs had
drawbacks, such as the possibility of analyzing only single transport modes, but the mixed integer pro-
grams have been developed to include multiple transport modes, transfer pricing (Vidal & Goetschalckx,
2001), development and recycling costs (Fandel & Stammen, 2004). Currently, mixed integer programs
are the most powerful methods of finding the single best solution for very large problems. In finding
the optimum, they converge to the result rather than enumerating the entire solution space. operations
research methods are deterministic analytical, however, so they must be augmented with simulation to
understand the dynamic behaviour of solutions (Riddalls, Bennett, & Tipi, 2000).
Analytic hierarchy processes (Wang, Huang, & Dismukes, 2004) can be employed in situations where
each level in the product hierarchy has a large number of possible suppliers. Analytic hierarchy process
uses balanced scorecard approach with set criteria to determine the best supplier choices (without regard
to the complete product structure) for a single component. In the second stage of the analysis, the best
supplier choices for each component are fed into preemptive goal programming for the entire product
structure. This approach fits in situations where there are tens or hundreds of possible suppliers for each
component, and a short list is needed first. However, it also requires expert knowledge to faithfully com-
pare one supplier against another. Analytic Hierarchy Process is a deterministic analytical method.
Control theoretical methods for demand supply network analysis have recently surfaced as they
provide an analytical method for estimating bullwhip effects in demand supply networks (Ortega &
Lin, 2004). The demand supply networks are modeled in z-space, and z-transforms are used to arrive at
Bode plots for dynamic behaviour. This method can analyze only one demand supply network setup at
a time, but it has the benefit of estimating system dynamics without discrete simulation. However, the
modeling of supply networks in Z-space requires expert knowledge, which may be a drawback in the
business environment. Control theoretical methods of demand supply network analysis are stochastic
analytical in nature.
Discrete simulation methods allows for dynamic analysis of a single demand supply network (Persson
& Olhager, 2002). An arbitrary demand signal may be fed to the network, and simulation determines
possible stockouts and order fulfillment lead time violations. The advantage of discrete simulation over
control theoretical methods is the ease of specifying input signals, and the increasing processing power
of computers keeps simulation runtimes reasonable. Discrete simulation is currently viewed as a de
facto standard of analyzing dynamic behaviour of demand supply networks (Riddalls, Bennett, & Tipi,
2000). However, discrete simulation is incapable of network optimization - i.e. the user must specify
the network structure to be simulated herself.
Recently, simulation optimization emerged as a way of combining the static analysis power of opti-
mization and dynamic analysis power of simulation (Azadivar, 1999; Truong & Azadivar, 2003). This
method is a combination of “what-if” and “how-to” questions (Azadivar, 1999). In answering “what-
if” questions, a candidate demand supply network is simulated against a demand pattern to determine