摘要
Vertically integrated manufacturers increasingly couple high-throughput flow lines with a flexible job-shop stage. Scheduling such hybrid systems is challenging when semi-finished products (SFPs) are perishable and processing times are stochastic, two constraints that seldom appear together in prior work. We formulate a stochastic hybrid-shop scheduling problem that jointly minimizes order tardiness, SFP holding cost, and expiry cost. To solve this problem efficiently, we propose the Metamodel-Guided L & eacute;vy Optimization (MGLO). MGLO combines L & eacute;vy-flight search with an XGBoost classifier that is continuously refreshed through an active-incremental learning loop. A bidirectional coordination scheme alternates optimization of the flow line and job shop subsystems while adaptively switching between global and local L & eacute;vy parameters. Monte Carlo experiments on both small-and large-order scenarios show that MGLO attains lower median and mean costs than both a genetic algorithm and a stand-alone L & eacute;vy heuristic across all processing-time distributions, with statistically significant gains in the large-order setting. In addition, the dynamic metamodel update achieves about 87% classification accuracy versus about 82% for a periodic-update baseline. Sensitivity and order-selection studies demonstrate the framework's robustness and yield actionable managerial guidance. The proposed method advances hybrid-shop scheduling by embedding perishability and uncertainty in a single optimization-learning loop and offers a reusable template for smart, adaptive manufacturing.