Deep Reinforcement Learning を用いたマシニングセンタのジョブ選択基準のモデル化

抄録

Multi-agent reinforcement learning has been applied to the autonomous distributed scheduling method based on utility values in order to improve the sum of earliness and tardiness of all jobs, in the previous researches. New distributed scheduling method is proposed by using deep reinforcement learning in this research. Firstly, job selection criteria of machining centers are proposed by using Q-learning based on status s of manufacturing system and value Q(s, a) of action a to determine utility values. Deep Q-network is one of deep reinforcement learning method. The Q-network is neural network model whose input data are status s and output data are value Q(s, a). Deep Q-network algorithm is applied to the proposed distributed scheduling method by using Q-learning in order to estimate the value Q(s, a) by machining center to select the suitable job for the distributed scheduling. Training of the Q-network are also carried out efficiently in the Deep Q-network algorithm.