| The Shifting Balance Genetic Algorithm (SBGA) is an extension of the Genetic Algorithm (GA) that was created to improve performance in highly multimodal environments. However, in this thesis it was found that the original analysis and explanation of why the SBGA behaves as it does was flawed. Consequently a new behavioral model is presented and based on it, two modified version of the algorithm are given. Both keep the effective aspects from the original SBGA as well as incorporating some recent GA innovations. The first of the new systems introduces dynamic population sizes alongside a restart mechanism. The second system introduces to the SBGA a bi-directional ring communication topology. Finally these two systems are experimentally compared to the old SBGA, as well as various island model GAs and the canonic GA. The new systems demonstrate a greater ability to overcome local optima in multimodal environments and hence lessen premature convergence. |
