[5M1-GS-2b] Machine learning

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)

[5M1-GS-2b-03] Reinforcement Learning Model for Cash-in-Transit Rebalancing Problem Considering Reuse of Collected Cash

〇Ryoga Miyajima¹, Ai Kondoh¹, Hideaki Tamai¹ (1. Oki Electric Industry Co., Ltd.)

Keywords:

Combinatorial Optimization,Operations Research,Reinforcement Learning

Cash-in-Transit (CIT) refers to the secure transportation of cash by security companies using specialized vehicles. This paper focuses on CIT operations that replenish or collect cash from devices such as ATMs and cash dispensers to prevent shortages or overflows. Traditionally, CIT bases must prepare cash in advance for loading and return collected cash for inspection, resulting in significant cash handling costs at the base. To address this issue, a CIT-Rebalancing approach has been proposed, in which cash collected from devices is inspected on the vehicle and reused for loading into other devices. However, optimizing daily CIT-Rebalancing plans to minimize cash handling costs at the base is an NP-hard problem, making exact solutions impractical for real-world applications. In this paper, we formulate the CIT-Rebalancing problem as a Markov Decision Process and propose a deep reinforcement learning method to efficiently obtain approximate solutions. Numerical experiments show our method outperforms conventional heuristic approaches.