A multi-stage machine state model was proposed to address the single-machine scheduling problem under machine degradation and maintenance strategies, with the objective of minimizing total production cost. A state transition mechanism was designed to incorporate both degradation evolution and maintenance effects. Job tardiness cost, machine operating cost, and maintenance cost were jointly considered to improve economic efficiency in production of the entire production process. An integrated decision-making framework for scheduling and maintenance based on deep reinforcement learning was developed, in which the Agent was trained through interaction with the environment to learn optimized scheduling and maintenance strategies. Joint decisions on job sequencing and maintenance timing were realized in complex dynamic systems. Benchmark instances of various scales were designed, and the effectiveness of the proposed model and framework was validated through computational experiments. The results indicate that the proposed approach achieves better performance in minimizing total scheduling and maintenance costs compared with several integrated optimization strategies. The conflict between production scheduling and machine maintenance is effectively balanced, and a more advantageous integrated optimization strategy for scheduling and maintenance is realized in dynamic and uncertain environments.