A deep Q network algorithm for a car resequencing problem in automobile factories

In this paper a car resequencing problem between the paint shop and the assembly shop in an automobile manufacturing factory is studied. Key characteristics of each car body include color, power type and drive type. Since each shop has different production preferences and constraints, they cannot work according to the same sequence, which requires the establishment of a painted body store (PBS). PBS is used to adjust the outgoing sequence of the paint shop to the incoming sequence that meets the constraints of the assembly shop (maximizing the reward value according to certain rules). An MDP model incorporating the objective function into the definition of the states is constructed and a deep Q network algorithm (DQN \(_0\) $_0^{}$ ) is developed to solve this problem. Greedy algorithms and another deep Q network algorithm based on an alternative MDP model (DQN \(_1\) $_1^{}$ ) are also designed. Numerical experiments are carried out and the results show that (1) DQN \(_0\) $_0^{}$ algorithm can obtain a solution very fast; (2) it always outperforms the greedy algorithms; (3) DQN \(_0\) $_0^{}$ can obtain a solution as good as DQN \(_1\) $_1^{}$ , but it runs much faster.