Masakazu WATABE, Keiki TAKADAMA and Katsunori SHIMOHARA
Scheduling with Organizational Learning Agent
Abstract:The new possible application of our organizational learning model is revealed in this study. The model consists of several concepts including Reinforcement Learning (RL), Rule-based system (including Generation/Exchange of rules), Multiagent system. This model, called Organizational-Learning Oriented Classifier System, is applied to the scheduling of space crews' tasks. As well as the normal scheduling, we conduct the scheduling with anomalies aimed at measuring the effectiveness of this model in unexpected situations. Furthermore, we discuss the effective way to design the agents' actions in the domain. The series of experiments shows the acceptable performances of the model; it provides practically feasible schedules at low computational cost and with completion times of all tasks in both expected and unexpected situations.
Keywords: organization-learning oriented classifier system, learning classifier system, task scheduling/rescheduling, multiple learning agents, rule design