Research on supply chain resilience-cost optimization joint decision making in multinational enterprise digital transformation based on multi-task deep reinforcement learning
TL;DR
The results suggest that MT-DRL can improve the simulated resilience-cost trade-off, while further validation with transparent datasets and real-world deployment tests remains necessary.
Abstract
This study develops a multi-task deep reinforcement learning (MT-DRL) framework for joint resilience-cost decision-making in multinational enterprise supply chains under digital transformation. The framework addresses the computational problem of learning a joint policy under dynamic disruptions rather than merely applying a managerial simulation. Its main novelties include: (1) a decoupled network architecture that combines shared representation learning with task-specific cost and resilience heads; (2) an adaptive weighting mechanism that updates objective priorities according to environmental uncertainty; and (3) a reproducible digital supply chain simulation environment integrating state, action, reward, and disruption-generation modules. Experiments on a 12-node supply chain network and simulated disruption scenarios show that, compared with selected single-objective and fixed-weight baselines, the proposed model increases the resilience index by 18.7%, reduces total operational cost by 14.3%, improves the order fulfillment rate by 12.5%, and shortens recovery time by 32%. The results suggest that MT-DRL can improve the simulated resilience-cost trade-off, while further validation with transparent datasets and real-world deployment tests remains necessary.
