Tuning Metaheuristics with Tree-Structured Parzen Estimator: A Case Study on Scheduling

Optimizing production schedules with sequence-dependent setup times (SDSTs) represents a critical challenge in manufacturing operations. This study proposes an intelligent scheduling framework that combines a Genetic Algorithm (GA) with automated parameter tuning using the Tree-Structured Parzen Estimator (TPE). The TPE learns from previous optimization runs to systematically adjust GA parameters, eliminating manual trial-and-error approaches. We evaluate the method on parallel machine scheduling problems with SDSTs, comparing performance against exact algorithms under equivalent computational time constraints. Experimental results demonstrate significant total completion time reductions of up to 32.12% in best cases and 14.87% on average compared to exact methods. When TPE is applied to Simulated Annealing instead of GA, performance consistently degrades by 1.10% to 13.23%, confirming the superiority of the GA-based approach. The proposed framework offers a scalable, adaptive solution for manufacturing scheduling optimization with direct implications for production efficiency and cost reduction.