흐름 생산 공정문제를 위한 dueling double Q-러닝 기반 강화학습 기법 연구

Abstract

Flow shop scheduling is an important optimization problem for operating actual manufacturing facilities. In this paper, we propose a novel flow shop scheduling scheme based on the double Q-learning and the dueling architecture. We designed double Q-learning process using two estimation functions rather than a single estimation function to avoid over-estimation problem. In addition, the adaptation of the dueling architecture, which can provide robust policy estimation performance using both the state function and the advantage function, leading to less-variance and efficient learning process. We conducted extensive simulations of flow shop scheduling using multiple datasets with various scheduling scales. From the simulation result, we observed that the proposed scheme outperforms the existing heuristic and reinforcement learning-based scheduling schemes in terms of the final manufacturing time consumption for various flow shop scales.