Management strategies of multinational corporations' green supply chains by deep reinforcement learning

Multinational corporations (MNCs) face the dual challenge of optimizing sustainability performance along with operational efficiency within complex global supply chains, particularly under emerging trade regulations such as the Carbon Border Adjustment Mechanism. This study aims to develop an intelligent decision-making framework to support both strategic and real-time optimization in green supply chain management (G-SCM). The proposed approach integrates an improved Transformer architecture with deep reinforcement learning (DRL) by modeling the dynamic supply chain environment as a Markov Decision Process. Environmental indicators, such as carbon emissions and energy consumption, are coupled with operational metrics, including inventory turnover and order fulfillment rates, into a unified reward function. A hierarchical attention mechanism is employed to capture the high-dimensional dependencies among supplier relationships, logistics pathways, and green practices. Experimental results demonstrate that the proposed model significantly outperforms traditional DRL baselines in reducing carbon emissions, increasing renewable energy usage, and maintaining high order fulfillment rates. The inclusion of a supply chain complexity index further enhances robustness against disruptions. Additional analysis reveals that the model's attention weights reflect the relative influence of key decision variables on overall optimization, providing interpretable pathways for multi-objective coordination in supply chains. Overall, this study presents an intelligent optimization method capable of real-time adaptation to complex supply chain dynamics while balancing environmental and operational performance, providing valuable insights for both practical G-SCM operations and policy formulation.