Abstract:
System Dynamics is a methodology to build valid models of systems in order to come up with policies to improve their dynamic behaviors. Search algorithms are increasingly being used in model building, validation, sensitivity analysis, and policy design phases of System Dynamics. This increase in application of search algorithms brings out the necessity to evaluate the efficacy of these algorithms with respect to the expectations of System Dynamics. This necessity forms the primary objective of this thesis. The thesis focuses on Genetic Algorithms (GA), which is a well-known and frequently used search algorithm. The main objective is to explore Genetic Algorithms’ adequacy in applications to System Dynamics methodology. We create eight variants of GA from the vast amount of options that the GA approach provides. The study focuses on six different dynamic models, three of them being from the area of control theory and the other three from socioeconomic System Dynamics literature. For each model, we determine an appropriate objective function based on some desired dynamic behavior and set of analysis parameters. The performance of eight GA configurations for all of the models indicates that Genetic Algorithms obtain good solutions, i.e. behaviors that match the desired behaviors. Specifically, explorative type of Genetic Algorithms is more suitable for System Dynamics methodology: as the search space grows, they are more likely to find good solutions fast. Additionally, we emphasize to the fact that the objective functions and results provided by algorithms should not be regarded as absolute or conclusive truths; rather they must be critically analyzed, objective function and parameters of the algorithm iteratively reformulated as necessary. We demonstrate this fact in some of the models by further analyzing the systems based on initial results provided by the algorithm.
