Multi-Grid Genetic Algorithms For Optimal Radiation Shield Design
In 2004 I returned to graduate school at the University of Michigan. My PhD in the nuclear engineering department focused on using genetic algorithms to design radiation shields and a new method of genetic algorithms called multi-grid genetic algorithms which improves the performance of GA on geometrically scalable problems.
Genetic Algorithms (GA) are a powerful search and optimization technique that can be applied to numerous problems. Unfortunately, GA relies on large numbers of fitness evaluations to determine the relative merits of various solutions to a problem. For problems requiring computationally intensive fitness evaluations this can make GA too expensive to use. We describe a hierarchical technique that we have created called Multi-Grid Genetic Algorithms (MGGA). MGGA leverages the geometry of a problem space to build a hierarchy of increasingly smaller problem spaces. Optimizations over these smaller spaces are used to seed a population of solutions in a larger space. We explore how MGGA can be applied to several radiation shielding problems.