Date Awarded


Document Type


Degree Name

Doctor of Philosophy (Ph.D.)


Computer Science


Stefan Feyock


Procuring combat systems in the Department of Defense is a balancing act where many variables, only some under control of the department, shift simultaneously. Technology changes non-linearly, providing new opportunities and new challenges to the existing and potential force. Money available changes year over year to fit into the overall US Government budget. Numbers of employees change through political demands rather than by cost-effectiveness considerations. The intent is to provide the best mix of equipment to field the best force against an expected enemy while maintaining adequate capability against the unexpected. Confounding this desire is the inability of current simulations to dynamically model changing capabilities and the very large universe of potential combinations of equipment and tactics.;The problem can be characterized as a stochastic, mixed-integer, non-linear optimization problem. This dissertation proposes to combine an agent-based model developed to test solutions that constitute both equipment capabilities and tactics with a co-evolutionary genetic algorithm to search this hyper-dimensional solution space. In the process, the dissertation develops the theoretical underpinning for using agent-based simulations to model combat. It also provides the theoretical basis for improvement of search effectiveness by co-evolving multiple systems simultaneously, which increases exploitation of good schemata and widens exploration of new schemata. Further, it demonstrates the effectiveness of using agent-based models and co-evolution in this application confirming the theoretical results.;An open research issue is the value of increased information in a system. This dissertation uses the combination of an agent-based model with a co-evolutionary genetic algorithm to explore the value added by increasing information in a system. The result was an increased number of fit solutions, rather than an increase in the fitness of the best solutions. Formerly unfit solutions were improved by increasing the information available making them competitive with the most fit solutions whereas already fit solutions were not improved.



© The Author