Reward shaping to improve the performance of deep reinforcement learning in perishable inventory management

Bram J. De Moor, Joren Gijsbrechts, Robert N. BouteEur. J. Oper. Res 2022 · DOI · OpenReview

Abstract

Highlights•We use known policies to improve deep reinforcement learning for inventory control.•We adopt potential-based reward shaping to accelerate training and learning.•We implement our approach on a deep Q-network to manage perishable inventories.•Shaping can improve policy performance, while it stabilizes and accelerates learning.
BibTeX
@article{moor2022reward,
  author = {Bram J. De Moor and Joren Gijsbrechts and Robert N. Boute},
  title = {{Reward shaping to improve the performance of deep reinforcement learning in perishable inventory management}},
  journal = {Eur. J. Oper. Res.},
  year = {2022},
  doi = {10.1016/j.ejor.2021.10.045},
}

Loading related content...

Figure