Scheduling in Queueing Systems and Networks Using ANFIS

This paper is concerned with a scheduling problem in many real-world systems where the customers must be waiting for a service known as queueing system. Classical queueing systems are handled using probabilistic theories, mostly based on asymptotic theory and/or samples analysis. We address a situation where neither enough statistical data exists, nor asymptotic behavior can be applied to. This way, we propose to use an Adaptive Neuro-Fuzzy Inference System (ANFIS) method to infer scheduling rules of a queueing problem, based on uncertain data. We use the utilization ratio and the work in process (WIP) of a queue to train an ANFIS network to finally obtain the estimated cycle time of all tasks. Multiple tasks and rework are considered into the problem, so it cannot be easily modeled using classical probability theory. The experiment results through simulation analysis show an improvement of our ANFIS method in the performance measures compared with traditional scheduling policies.