Adaptive Deep Reinforcement Learning for Dynamic Parallel Machines Scheduling with Maintenance Activities

In this paper, the parallel machines scheduling with jobs arrive over time and maintenance activities is investigated. The difficulty of the problem is how to achieve the trade-off between the scheduling and maintenance on parallel machines to optimize the cost and reliability of machines under the dynamic environment. For this challenge, this paper proposes an adaptive deep reinforcement learning method to achieve rapid response and optimization. Under the framework, a deep reinforcement learning is applied to learn the optimal scheduling and maintenance policy. However, it is difficult to adapt to dynamic environment by observing static state features. Therefore, a feature selection method based on deep reinforcement learning was raised to adaptively select the key state features of the production process so as to improve the adaptive ability and generalization of the proposed method. The experimental results of the instances with different scales indicate the proposed method can select state features adaptively and effectively solve the dynamical scheduling on parallel machines with maintenance activities.