Real-World Cardboard Production Optimization Using Multi-Objective Evolutionary Algorithms

This paper proposes a real-world cardboard production planning optimization system based on multi-objective evolutionary algorithms. The cardboard production planning is a kind of optimization problem finding a base paper combination for an order list of many different cardboards. The proposed system aims to minimize production costs and carryover orders using evolutionary computation while reducing the reliance on skilled human experts in the field. To optimize the problem with evolutionary algorithms, this paper proposes a variable design that assigns a base paper number specifying a paper material and a paper width to each cardboard in the order. Additionally, this paper proposes a two-step crossover aligned with the proposed variable design to generate new solutions. Specifically, it consists of (1) flip crossover for the paper material and (2) simulated binary crossover (SBX) for the paper width. Experiments were conducted using simulated data to compare the production costs and carryover orders generated by the proposed system against those created by a skilled human expert. The results demonstrate that among various multi-objective evolutionary algorithms, NSGA-II achieves the highest optimization performance. Furthermore, NSGA-II consistently outperforms manual planning by the human expert in terms of both production costs and carryover orders across five days with different order lists. These findings suggest that the proposed system can reduce the workload of skilled experts in daily production planning and mitigate reliance on human-dependent expertise.