Hyperheuristic Method Based on Deep Reinforcement Learning

Abstract

For solving combinatorial optimization problems, methods in which a candidate solution is iteratively updated are often used. One of the problems in the iterative methods is how to update the candidate solution, and it is not easy to design an appropriate update method. To solve the problem, hyperheuristics have been proposed. They can find a near-optimal solution by using multiple update methods and automatically selecting an appropriate update method, often combined with an existing optimization algorithm such as evolutionary computation. On the other hand, deep reinforcement learning has recently attracted attention due to its ability to solve large-scale and complicated problems. This paper proposes a hyperheuristic method introducing deep reinforcement learning to automatically find the appropriate update method. As a case study, we apply the proposed method to a drone delivery problem and evaluate its performance.