Simulation Optimization in the CTDS Domain

In this chapter the optimization task is examined within the DEDS modelling and simulation context. Because of the inherent random behaviour within DEDS simulation models it is noted that the criterion function itself is a random variable and consequently its meaningful evaluation for any particular parameter vector selection requires multiple replications and an associated averaging operation. The number of such replications that are carried out can impact the outcome of the optimization procedure. A considerable variety of approaches have emerged for dealing with the optimization task when random effects are present. These can be separated into general categories and four such categories are identified and their special features are briefly outlined. These are heuristic methods inasmuch as access to the meaningful gradient information is, for the most part, not feasible. A case study is presented to illustrate some of the special challenges that can be encountered. The problem has seven parameters but each is constrained to a small integer range. The simulation model itself has steady state requirements and consequently a “warm-up” period is necessarily embedded in experimentation phase. Optimization results for the problem are generated using the Nelder-Mead method (outlined in Chap. 10) but with significant modifications to accommodate both the integer constraint on parameter values and their restricted range of permitted values. Because of these constraints on the parameter values, the search space is relatively small and the results of an exhaustive search over that space are also presented.