Learning Priority Rule using Neural Network for Job Shop Scheduling with regard to Weighted Tardiness

抄録

This paper deals with job shop scheduling using priority rules. Because of its computational efficiency, applying priority rules is the most practical approach for real-world practical scheduling problems. However, effective priority rules have been developed by trial and error, and scheduling performance generated by simple priority rules is not always good. This paper examines learning priority rules for job shop scheduling with regard to weighed tardiness using the data obtained from optimal or near-optimal schedules. The optimal or near-optimal schedules are generated using an IP solver or a genetic algorithm. Numerical experiments show that it is possible to learn effective priority rules if training data of enough quality are used for learning.