Exploring Reward-based Hyper-heuristics for the Job-shop Scheduling Problem

Abstract

The Job-Shop Scheduling Problem represents a challenging field of study due to its NP-Hard nature. Its many industrial and practical, real-world applications skyrocket its importance. Particularly, hyper-heuristics have attracted the attention of researchers on this topic due to their promising results in this, and other optimization problems. A hyper-heuristic is a method that determines which heuristic to apply at each step while solving a problem. This investigation aims at rendering hyper-heuristics by combining unsupervised and reinforcement learning techniques. The proposed solution applies a clustering approach over the feature space, and then, it generates knowledge about heuristic selection through a reward-based system. Results show that our hyper-heuristics surmount competent heuristics, such as SPT and MRT, in various test instances. Besides, some of these hyper-heuristics outperformed the best result obtained among all the heuristics in more than 33% of the instances. Hence, we believe that the proposed approach is promising and that more knowledge about its benefits and limitations should be derived through its application on different problems.