The steel plate marking process in shipbuilding largely consists of two main tasks: marking and cutting. This study focuses on the sequencing problem of the marking task. Unlike cutting, marking does not involve significant heat, resulting in fewer industrial constraints. Therefore, the primary objective becomes minimizing the movement path of the marking machine. From the perspective of minimizing movement, the marking task can be generalized as a problem of passing through all marking lines while minimizing travel distance, which can be interpreted as a Travelling Salesman Problem (TSP). However, since marking targets lines rather than points, there is an additional constraint in which the machine must enter from one endpoint of a line and exit through the other. To solve this TSP variant with such constraints, this study proposes an encoder-decoder-based reinforcement learning approach. Experimental results confirm that the proposed algorithm outperforms existing methods.