Intelligent Order Acceptance in Make-to-Order Manufacturing using Reinforcement Learning

FACUNDO ARREDONDO; ERNESTO MARTÍNEZ

Santa Fe, Argentina

Conferencia; 34th Conferencia Latinoamericana de Informática, CLEI 2008; 2008

SADIO

Order acceptance (OA) is a critical decision-making problem at the interface between customer relationship management and production planning in make-to-order manufacturing systems. The main issue in this work is the development of an intelligent OA framework based on reinforcement learning. The attention is focused on iterative development of an order acceptance policy based on evaluative feedback of profits obtained. The OA policy is first obtained using simulations and later updated on-line based on revenues per unit cost of proccesing capacity to deal with problem uncertainty. Locally weighted regression is used to generalize cumulative rewards of accepting or rejecting similar orders regarding attributes such as product mix, size and due date. The balance exploration-exploitation of the learned policy is based on classifying an arriving order as belonging either to the acceptance set or to the rejection set. Results highlight the importance of automated learning of an adaptative OA policy using actual rewards when demands exceed capacity