Scholarlink

Supply Chain Management (SCM) is crucial for balancing customer satisfaction and cost control, especially under global risks such as natural disasters, geopolitical tensions, and demand fluctuations which can lead to extended procurement or transportation delays in customer delivery. Traditional inventory optimization methods often rely on statistical models assuming predictable probability distributions for demand and lead times. However, real-world supply chains frequently encounter unpredictable disruptions like sudden demand surges or prolonged delays, which complicate these approaches. To address these issues, we propose a Recurrent Neural Network (RNN) based Deep Reinforcement Learning (DRN) method with an Agent-based SC Simulator (ASCS) as a Cyber-Physical Systems (CPS) that create a virtual representation of information, material flow including inventory, business processes, and cash flows. First, we introduce inventory optimization problem to balance order fill rate and inventory cost with a multi-stage SC structure. Next, an Agent-based SC simulator (ASCS) is introduced as a CPS tool. Then, we propose a RNN based DRN framework with a ASCS for an inventory optimization problem. Finally, we verify its effectiveness through computational experiments. Experiment results show that a proposed RNN based DRN framework leans optimal policy to get parameters for a ASCS minimizing inventory cost while satisfying constraints.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Reinforcement Learning for Multi-stage Inventory Optimization with Agent-Based Supply Chain Simulator

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Atsuki Kiuchi,
  • Hitohiro Eto

摘要

Supply Chain Management (SCM) is crucial for balancing customer satisfaction and cost control, especially under global risks such as natural disasters, geopolitical tensions, and demand fluctuations which can lead to extended procurement or transportation delays in customer delivery. Traditional inventory optimization methods often rely on statistical models assuming predictable probability distributions for demand and lead times. However, real-world supply chains frequently encounter unpredictable disruptions like sudden demand surges or prolonged delays, which complicate these approaches. To address these issues, we propose a Recurrent Neural Network (RNN) based Deep Reinforcement Learning (DRN) method with an Agent-based SC Simulator (ASCS) as a Cyber-Physical Systems (CPS) that create a virtual representation of information, material flow including inventory, business processes, and cash flows. First, we introduce inventory optimization problem to balance order fill rate and inventory cost with a multi-stage SC structure. Next, an Agent-based SC simulator (ASCS) is introduced as a CPS tool. Then, we propose a RNN based DRN framework with a ASCS for an inventory optimization problem. Finally, we verify its effectiveness through computational experiments. Experiment results show that a proposed RNN based DRN framework leans optimal policy to get parameters for a ASCS minimizing inventory cost while satisfying constraints.