Sign in Subscribe
Inventory control

Enhanced Multi-Task Deep Reinforcement Learning for Integrated Inventory-Routing Problem Under VMI Mode

Strategic Insights

— 1 min read

Researchers at Wuhan University of Technology have developed a novel approach to optimize inventory replenishment and routing problems in Vendor Managed Inventory (VMI) mode. The study, published in the Management System Engineering journal, proposes an enhanced Multi-Task Proximal Policy Optimization (MTPPO) with deep reinforcement learning to refine inventory replenishment strategies and optimize routing strategies. The model demonstrates significant reductions in inventory costs and total costs, outperforming heuristic algorithms.

Key Takeaways:

  • The research focuses on the integrated optimization of inventory replenishment and routing problems in VMI mode, highlighting the importance of joint optimization for minimizing supply chain costs and enhancing operational efficiency.
  • The proposed MTPPO model refines inventory replenishment strategies by learning from inventory status and retailer location data, reducing inventory costs by 8.58%.
  • Routing strategies are optimized by utilizing a Graph Isomorphism Network (GIN) to analyze the network data of retailers and formulate routing strategies based on delivery requirements and retailer network information.
  • Experimental results demonstrate that the MTPPO outperforms heuristic algorithms, reducing total costs by 6.18%.
  • The MTPPO model is an enhanced version of the Multi-Task Proximal Policy Optimization (MTPPO) algorithm, incorporating deep reinforcement learning to optimize inventory replenishment and routing strategies.
  • The research team, led by Gang Lu, includes Junmin Wan, Lijing Du, and Xiaofang Chen as co-authors.
  • The study has significant implications for supply chain management, particularly in the context of Vendor Managed Inventory (VMI) mode.

Statistics:

  • Inventory costs are reduced by 8.58% using the proposed MTPPO model.
  • Total costs are reduced by 6.18% compared to heuristic algorithms.
  • The MTPPO model demonstrates better performance than heuristic algorithms in Experimental results.
  • The research was financially supported by the National Natural Science Foundation of China and the Humanities and Social Science Fund of the Ministry of Education of China.

Sources:

  • Enhanced multi-task deep reinforcement learning for the integrated inventory-routing problem under VMI mode. Management System Engineering, 2025, 4(1):1-19. (https://doi.org/10.1007/s44176-025-00053-2)
  • Journal of Engineering. October 20, 2025; p 3347.
  • Springer.
  • Wuhan University of Technology.
  • NewsRx LLC.

Read more