|
Reinforcement Learning approaches for the Parallel Machines Job Shop Scheduling Problem Host Publication: Proceedings of the Cuba-Flanders Workshop on Machine Learning and Knowledge Discovery Authors: Y. Martinez Jimenez, W. Tony, A. Nowé, K. Verbeeck, D. Causmaecker Patrick, J. Suarez and R. Bello Publication Date: Feb. 2010
Abstract: This paper presents two Reinforcement Learning approaches for the Parallel Machines Job Shop Scheduling Problem. This NP-hard optimization problem where operations have to be assigned to machines of different workcenters, each workcenter having a number of identical parallel machines, is hard to solve and is therefore tackled with learning based methods. The objective used is the minimization of the schedule makespan. Two viewpoints are taken, one where resources are intelligent agents and have to choose what operation to process next, and another where operations themselves are seen as the agents that have to choose their mutual scheduling order. As Reinforcement Learning methods we use a value iteration method (Q-Learning) and a policy iteration method
(Learning Automata). The results of both approaches improve on recent published results from the literature.
|
