Deep reinforcement learning approach for dynamic capacity planning in decentralised regenerative medicine supply chains

Abstract

Decentralized manufacturing has the benefits of fast fulfillment, reducing risks of distant delivery, and improving patient access to personalised regenerative medicine (PRM). Implementing the decentralised PRM manufacturing system successfully needs a capacity planning strategy involving inventory replenishment, capacity allocation, and demand sharing to mitigate the impacts of supplier disruption and satisfy demand with a high service level. However, existing methods for generating optimal capacity planning policies for such PRM systems require knowing the distributions of the supplier disruption and demand uncertainty, which is usually unknown for PRM supply chains. This study proposes a data-driven approach that can learn effective capacity planning policy under various manufacturing circumstances without knowing the exact distributions. The proposed approach utilises a production simulation model and a deep reinforcement learning method. Case study results demonstrate that the proposed method can outperform existing methods when ground-truth demand forecasts differ from priori estimations. The results also support that the proposed method not only can be applied in regenerative medicine but also in many other sectors.