Metamodel-based Simulation Optimization Using Machine Learning for Solving Production Planning Problems in the Automotive Industry

Aug 3, 2026

Rising production-system complexity in the automotive industry makes simulation indispensable for analyzing dynamic behavior, yet exhaustive parameter search becomes impractical as combinations grow exponentially. This paper investigates simulation optimization accelerated through machine learning: metamodels integrate multiple ML algorithms with metaheuristics to optimize production planning cases with several parameter classes. The proposed approach targets faster evaluation of complex systems while preserving decision-relevant accuracy for industrial use.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Simulation and optimization

Background and related work

Simulation and optimization

Methodology and metamodel construction

Algorithms and optimization integration

Case study and transferability

Automotive production planning use case

Transferability

Conclusion and outlook

Proceedings of the 58th Hawaii International Conference on System Sciences | 2025 Metamodel-based Simulation Optimization Using Machine Learning for Solving Production Planning Problems in the Automotive Industry Felicia Schweitzer Mercedes-Benz AG Stuttgart felicia.schweitzer@mercedes -[benz.com](benz.com) Lars Habel Mercedes-Benz AG Stuttgart lars-christian.habel@mercedes[benz.com](benz.com) Oscar Jhonny Canaviri Vilca Mercedes-Benz AG Sindelfingen oscar jhonnyy.canaviri vilca@ [mercedes-benz.com](mercedes-benz.com) Tobias Kulzer Mercedes-Benz AG Sindelfingen [tobias.t.kulzer@mercedes-benz.com](tobias.t.kulzer@mercedes-benz.com) Abstract Due to the rising complexity of production systems in the automotive industry, simulation has become an established tool for analyzing dynamic systems. However, once the number of parameter combinations rises exponentially, the generation and evaluation of all possible solutions gets impractical. While the combination of simulation and optimization has a long tradition in academic research, its adoption in the automotive industry remains limited, often due to the high execution time associated with optimization experiments. To enable more efficient decisionmaking, this paper explores the integration of machine learning and optimization for simulation optimization. Specifically, it focuses on the use of metamodels incorporating various machine learning algorithmsand metaheuristics to optimize two production planning problems with multiple parameter classes. The presented approach enables decision-makers to conduct a rapid assessment of complex production systems. Keywords: Material flow simulation, optimization, machine learning, metaheuristics, metamodeling 1. Introduction The utilization of analysis tools to support decision-making has become essential for companies to evaluate dynamic systems in a production environment. Thus, simulation is an established tool for analyzing complex relationships in production and logistics. Especially in the automotive industry, simulation is a commonly used tool due to the growing complexity in their production and logistics systems (Gutenschwager et al., 2017) . For analyzing such systems, different parameters, constraints, and multiple objective functions are taken into Sigrid Wenzel University of Kassel [s.wenzel@uni-kassel.de](s.wenzel@uni-kassel.de) consideration. When the number of parameter combinations rises exponentially, evaluating all possible solutions becomes impractical. In these cases, it is beneficial to employ smart experiment design to effectively explore the parameter space. One concept for exploring large parameter spaces and gaining insights is Data Farming (Feldkamp et al., 2017) . After reducing the number of necessary simulation experiments, simulation combined with optimization can be applied for refining and identifying optimal solutions (Barton, 2009; März et al., 2011) . The objective of optimization is the determination of the best possible solution from a range of possibilities by using algorithmic methods (VDI, 2020) . In particular, the combination of optimization and simulation is often used as a decision support system for solving industrial problems such as resource allocation or buffer allocation (Soares do Amaral et al., 2022; Yelkenci Kose & Kilincci, 2020) . However, existing simulation optimization methods, especially when integrating simulation into the optimization, are often time-consuming and inefficient for large and complex problems (Liu et al., 2018) . This is because even hybrid-simulations still face the challenge of computational effort as optimization algorithms such as metaheuristics require many simulation runs for evaluation if they are combined with simulation directly (Sobottka et al., 2019) . Therefore, the utilization of metamodeling approaches representing the input and output relations of particular production planning problems has received increasing attention in

FAQs

Why does production planning simulation optimization become difficult in automotive manufacturing?

As parameter combinations increase exponentially, evaluating all possible solutions requires too many simulation runs and becomes impractical, especially when optimization algorithms need many evaluations.

What is the main idea of the proposed approach?

The approach combines machine learning with optimization by using metamodels that represent relationships between inputs and outputs, then applies metaheuristics to efficiently search for better production planning solutions.

How does the paper aim to support decision-making?

By enabling rapid assessment of complex production systems, decision-makers can evaluate promising configurations more quickly than with time-intensive direct simulation-optimization.

Aug 3, 2026

Upload by Theodore