With the continuous promotion of new generation network technology, the sky-ground integrated network integrates satellite communication and ground network, which brings brand new possibilities and challenges to the communication field. Aiming at the existing network’s inability to solve the problem of matching multi-service demand and resource scheduling under different conditions, this paper proposes a deep reinforcement learning-based intelligent scheduling algorithm (DRLIS), which optimises network performance by dynamically scheduling flexible transmission intervals through the DRLIS algorithm at the time slot level. The DRLIS algorithm dynamically schedules flexible transmission intervals to optimise network performance, dynamically and intelligently allocates resources to URLLC traffic through environmental interactions, as well as makes real-time decisions to resolve uncertainties, and considers enhanced mobile broadband (eMBB) and ultra-reliable low latency communications through a well-designed reward function. Reliable low latency communications (URLLC) service requirements. Simulation proves that the proposed DRLIS algorithm has better performance compared to the traditional TP algorithm.