A robust scheduling rule using a Neural Network in dynamically changing job-shop environments

Scheduling plays a critical role in job-shops that produce a wide variety of different jobs. This paper presents a robust and effective scheduling rule using a Neural Network (NN) trained as a priority rule for these complex and dynamic job-shops. The training is efficiently carried out through two stages: the first for effective scheduling under specific scheduling conditions, and the second for robust scheduling under various scheduling conditions. Numerical experiments under various scheduling conditions in which the level of machine utilisation and due-date tightness dynamically changes show that a trained NN outperforms the best dispatching rules available in the literature.