Optimization of Initial Population Size for Enhancing Solution Quality and Reducing Premature Convergence in Genetic Algorithms

Abstract

 Genetic Algorithms (GAs) are robust optimisation techniques derived from natural evolution; nonetheless, their efficacy is contingent upon parameter configuration, particularly population size. An inadequately chosen population size may elevate computing demands or lead to early convergence resulting from diminished genetic diversity. This study optimises the initial population size to enhance solution quality and minimise premature convergence and convergence duration. This research analyses the impact of initial population sizes on genetic algorithm performance and suggests an optimised initialisation strategy that maintains diversity without excessive population growth. A quantitative experiment compares two genetic algorithm variants: BASIC_GA (standard random population initialisation) and NEW_GA (diversity-controlled initial population generated by a minimum distance parameter (δ). A systematic adjustment of population sizes is used to study convergence behaviour and solution quality. Diversity-preserving initialisation in NEW_GA precludes superfluous or clustered individuals. Standard deviation, average fitness, and best fitness value measure algorithm performance across generations. This comparison shows that diversity-aware population initialisation improves genetic algorithm performance and reduces premature convergence. Greater populations do not inherently yield superior solutions, and meticulously optimising population size can enhance genetic algorithm performance. This study suggests that the starting population size is a critical element in genetic algorithms that influences convergence and solution quality. Diversity-controlled initialization improves genetic algorithm convergence and robustness without added computational cost.