Agent-Based Dynamic Order Acceptance Policy in Make-to-Order Manufacturing

Order acceptance is a key success factor in make-to-order (MTO) manufacturing firms. In this work, in order to maximize average revenue in an infinite planning horizon, we use dynamic programming to model the order acceptance problem, and solve it with reinforcement learning approach. A novel approach for simulation-based development for dynamic order acceptance using average-reward reinforcement learning is proposed. Through the simulation, an intelligent decision policy to dynamically control the coming orders is learned by the agent. Comparisons made with First-Come-First-Serve (FCFS) highlight the effectiveness of the proposed novel approach to maximize the average revenue.