Applications of evolutionary algorithms and simulation to decision-making under uncertainty

Research on the integration of optimization techniques with computer simulation has provided important, theoretical advances in addressing stochastic optimization problems. However, this research has precluded the examination of large complex problems. Models that represent such systems are difficult to formulate analytically since they are characterized by large complex search spaces containing nonlinearities, stochastic parameters, and/or integer decision variables. Furthermore, many solution methods are not conducive to addressing such problem formulations. Those that are able to search complex spaces are unable to account for the stochastic nature of the problem while methods that are capable of dealing with stochastic parameters are only able to solve small scale problems. Consequently, there exists a significant disconnect in the existing literature concerning how to reconcile important theoretical issues with practical implementation of simulation optimization algorithms.; The purpose of this thesis is to demonstrate how simulation can be used to determine optimal expected performance for complex large-scale problems involving significant uncertainty. It consists of two application studies involving real world data. The first study is presented in chapters 3 and 4. In chapter 3 the municipal solid waste system of the Hamilton-Wentworth region in Ontario, Canada is examined using a simulation optimization algorithm. Specifically an evolutionary algorithm combined with simulation (EAS) is used for the first time to address a large-scale problem. The results show that the algorithm succeeds in identifying a more efficient allocation of resources than the one currently used by the municipality and improves upon those found in a previous deterministic study (Huang et al. 1998). Chapter 4 extends the analysis to demonstrate how EAS can be used to generate a set of solution alternatives that can assist in policy formulation and design. This study makes a contribution to the MGA (modeling to generate alternatives) literature by directly incorporating uncertainty into the solution generation process.; The second study is presented in chapter 5. This study models the uncertainty in the supply of raw materials resulting from the waste flow of durable goods. This waste flow is difficult to estimate primarily because of the stochastic nature of many of the relevant parameters of the problem. Consequently, previous studies have not satisfactorily accounted for this uncertainty in their models. In this chapter, a model of the stochastic nature of the durable good waste flow for the case of the United States television market is developed. The results provide annual estimates of the currently unknown and uncertain supply of waste available and the timing of this supply. This model demonstrates that even in the presence of technological obsolescence of a durable good, the problem of how to dispose of consumer durables may exist long after the product has become obsolete.