An Analysis of the Interacting Roles of Population Size and Crossover in Genetic Algorithms
PPSN I Proceedings of the 1st Workshop on Parallel Problem Solving from Nature
Nash q-learning for general-sum stochastic games
The Journal of Machine Learning Research
| Hi-index | 0.00 |
This paper addresses decentralized multi-project scheduling under uncertainty. The problem instance we study is the scheduling of airport ground handling services, where aircraft turnarounds can be seen as multiple projects, ground handling services as activities, and service providers as resources. In this environment aircraft requiring ground handling services and the corresponding service providers are self-interested autonomous parties. Moreover, the environment is well-known for its large number of disturbances. We employ a heterogeneous multiagent scheduling framework with two types of autonomous agents representing aircraft and ground service providers respectively. We use online scheduling to cope with uncertainty in the release time of project: the uncertainty in aircraft arrival time at an airport. To balance the interests of the two types of agents in this heterogeneous multiagent system, we propose a market-based mechanism to assign time slots to aircraft turnaround activities. We study the use of this mechanism in a cooperative and a non-cooperative setting. In a dynamic environment such as airport ground handling, the execution of project schedules may be invalidated by various disruptions. As a result project agents may incur high costs if they have to reschedule some of their activities. The insertion of slack time between activities is a well known solution. The delay cost incurred by inserting slack should balance the expected costs of rescheduling some activities. Since in a dynamic multiagent system it is hard to analytically calculate optimal slack time between activities, we propose that agents determine these slack time using a co-evolutionary learning approach. Experiment show that our decentralized scheduling approach scores on average as high as well-established OR-based heuristics, and that slack times to keep a schedule stable can be learned.