Optimal Order Acceptance and Scheduling via Deep Reinforcement Learning

Abstract

Order acceptance and scheduling (OAS) in a vital problem for the management of production, storage, transportation processes. In an OAS problem, a decision maker should determine whether or not to accept a newly arrived order, and upon an order accomplished, it should further determine which order will be produced subsequently. This paper proposes an MDP for modeling the OAS problem, where a decision is made by jointly considering multiple features including price, quantity and latest delivery date, delayed delivery cost, and others. Furthermore, a deep reinforcement learning algorithm is developed for solving the optimal OAS policy, which outperforms existing schemes in numerical simulations.