Agent-based project scheduling: computational study of large problems

Abstract

We present in this paper the results of a computational study for project scheduling based on new ideas for project representation taken from digital circuit technology (Knotts et al. , 1998a) and a solution approach based on the artificial intelligence notion of agent technology. We experimented with projects with up to 10 000 stochastic duration activities which can be executed in a number of modes requiring renewable, nonrenewable, and periodically renewable resources. This study is about agent implementation in a project scheduling domain. It compares agent types and priority rules with respect to their impact on project schedule duration and computational performance. This work demonstrates: (i) that artificial intelligence concepts of agent technology can be successfully implemented for project scheduling; and (ii) in conducting project scheduling studies we can experiment successfully with large project networks. Both points made in this research are new.