Evaluation of Simplified Individuals in an Evolutionary Computation Algorithm with Autonomous Merging and Splitting

Abstract

Traditional Genetic Algorithms (GA) face challenges with high computational costs and low diversity in large-scale problems. To address this, we developed a Multi-Agent GA (MA-GA) where autonomous agents build solutions through distributed cooperation. In this framework, applied to the polygon packing problem, agents repeatedly merge to form partial solutions and split them if they prove unpromising, which maintains search diversity. However, complex polygons with many vertices lead to prohibitive computational costs, hindering the search process. This study introduces a preprocessing step to mitigate this issue. We employ the Ramer-Douglas-Peucker (RDP) algorithm to create ”simplified individuals” by reducing the vertex count of polygons while preserving their essential shapes. Using these simplified polygons in the MA-GA reduces execution time without compromising the quality of the final packed solution, achieving a better trade-off between solution accuracy and efficiency.