Optimierung einer stochastischen MRP-Simulation unter Anwendung der Bayes’schen Optimierung

This paper explores whether the Bayesian optimization algorithms GPEI, TurBO and SAASBO are effective for stochastic material requirements planning simulations. It includes a comparison of other methods, with a focus on the convergence speed, a key factor in simulation-based optimization. The study uses a simple material requirement planning simulation model that is progressively expanded in complexity by adding products and levels to the bill of materials. This results in a high-dimensional optimization problem, which poses a significant challenge for simulation-based optimization. The Bayesian optimization methods are compared at each level of complexity to determine if they produce satisfactory results. Additionally, the convergence speed is analyzed in relation to method and complexity. A genetic algorithm, CMA-ES, and Sobol serve as benchmarks for the Bayesian optimization methods.