Scholarlink

Distribution networks help to move goods from where they are made to where they are needed. However, figuring out the best way to do this is not easy such as how to reduce the cost of moving goods over a long time. This involves a lot of different goods and not knowing for sure how much people will want to buy. This problem is called the Shipping Point Assignment (SPA) problem. This decides which warehouse should get deliveries from suppliers. The goal is to do this in a way that makes it easier to manage the goods and not have to move them around too much. Therefore, in this study, we used recent improvements in solving tough problems like Reinforcement Learning (RL) and a Graph Convolutional Network (GCN). These helped us to make better decisions about which warehouse should get which deliveries. However, when tested our new method against some older ones using computer simulations—surprisingly—our new method did not always do better than the old ones—but we found some interesting things to explore more in future research. Therefore, this study shows that there is still a lot to learn about making distribution networks better. Even though the new method did not always win—it gives new ideas for solving these kinds of problems.

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

Optimizing Distribution Networks: A Study on Reinforcement Learning and Graph Convolutional Networks for Shipping Point Assignment

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Anoop Kumar,
  • Arpita Soni,
  • Rajeev Arora

摘要

Distribution networks help to move goods from where they are made to where they are needed. However, figuring out the best way to do this is not easy such as how to reduce the cost of moving goods over a long time. This involves a lot of different goods and not knowing for sure how much people will want to buy. This problem is called the Shipping Point Assignment (SPA) problem. This decides which warehouse should get deliveries from suppliers. The goal is to do this in a way that makes it easier to manage the goods and not have to move them around too much. Therefore, in this study, we used recent improvements in solving tough problems like Reinforcement Learning (RL) and a Graph Convolutional Network (GCN). These helped us to make better decisions about which warehouse should get which deliveries. However, when tested our new method against some older ones using computer simulations—surprisingly—our new method did not always do better than the old ones—but we found some interesting things to explore more in future research. Therefore, this study shows that there is still a lot to learn about making distribution networks better. Even though the new method did not always win—it gives new ideas for solving these kinds of problems.