Integrated Co-optimization of Lot Sizing and Scheduling in the Agri-Food In dustry: a Data-Driven Hybrid VNS–ALNS Approach
Abstract:
This paper addresses the problem of integrated co-optimization of lot-sizing and scheduling in multi-level agro-industrial production chains, characterized by product perishability, sequence-dependent setup times (SDST), and significant demand uncertainty. We propose a hybrid metaheuristic combining Variable Neighbor hood Search (VNS) and Adaptive Large Neighborhood Search (ALNS), enhanced with a data-driven guidance mechanism based on machine learning. This mechanism aims to predict the potential impact of neighborhood moves in order to filter out unpromising explorations and accelerate convergence. The performance of the approach is evaluated on a set of realistic instances inspired by the agro-industrial sector, including 20 to 100 products, multi-level structures, and up to 70 stochastic demand scenarios. Experimental results demon strate significant reductions in total cost (up to 20%), notable improvements in solution robustness (measured through CVaR and worst-case scenario cost), as well as increased stability of production plans compared to classical and unguided approaches. These findings highlight the benefits of integrating data-driven mecha nisms into hybrid metaheuristics for production planning under uncertainty in the agro-industrial context, effectively balancing solution quality, decision robustness, and computational efficiency.
