Hierarchical deep reinforcement learning for hydrogen supply chain management

For an effective transition from fossil fuel-based energy sources to renewable energy sources, it is crucial to accompany research on the design and optimization of supply chain for new energy sources. The traditional tool for optimizing supply chain management (SCM), the mathematical programming (MP) method, has limitations in terms of computation time and cost as the scale and complexity of the supply chain increase. Therefore, we need a new powerful optimization methodology that enables real-time decision making, considers the interaction among various components within the supply chain, and accommodates the uncertainties in demand and energy supply. In this study, we proposed deep reinforcement learning (DRL) as a new tool to overcome the limitations of MP and satisfy the conditions required for optimizing SCM, as mentioned earlier. Furthermore, we aim to compare a single-agent reinforcement learning (SARL) system with a multi-agent reinforcement learning (MARL) system. Our model achieves successful performance by converging to a value like the optimum of the MP.