Convergence Analysis of Genetic Algorithms on Dynamic Production Scheduling

The paper analyses the convergence behavior of the open-ended relevant alleles preserving genetic algorithm (OERAPGA) in dynamic production scheduling. In a dynamic production environment, frequent changes to the scheduling problem influence the convergence behavior of the applied genetic algorithm. This study investigates the impact of the two types of changes on the optimization process: removing the first task in the current solution from the problem and randomly removing one material along with the sub-materials from the problem. The impact of the changes is tested for different intervals, affecting the optimizer problem update frequency. The research findings show that frequent and substantive changes significantly reduce the convergence rate and can potentially halt convergence. For less aggressive changes, withholding problem update information demonstrated mixed results regarding the convergence rate, but impacted the optimization quality negatively. Ultimately, it is concluded that frequent updating results in the best optimization results, even if the optimizer does not converge. This is counter-intuitive coming from static optimization.