Abstract:This paper aims to address the intelligent and highly reliable operation requirements of distribution communication networks in the context of emerging power systems by proposing an integrated sensing, communication, and computing architecture together with an optimized computing resource scheduling method. To overcome the inefficiencies, latency instability, and lack of robustness inherent in traditional architectures, a three-layer collaborative framework consisting of sensing, communication, and computing layers is constructed. This framework enables efficient multi-source information acquisition, dynamic heterogeneous network transmission, and unified multi-level computing resource scheduling, providing a foundation for intelligent decision-making. Building upon this framework, a TS-PPO algorithm is designed by incorporating Knowledge Distillation (KD) and Imitation Learning (IL), combining expert prior strategies with reinforcement learning to accelerate policy convergence and enhance decision-making accuracy. To validate the proposed method, a heterogeneous simulation platform based on a typical medium-voltage distribution network scenario is established, and comprehensive performance evaluations are conducted under various service types, computing resource scales, and network state fluctuations. Experimental results show that the proposed algorithm outperforms random scheduling, greedy strategies, and conventional PPO in terms of task completion latency, energy consumption, success rate, and computing resource utilization. Specifically, average latency is reduced by 35.4%, energy consumption decreases by 32.8%, the task success rate reaches 92.4%, and resource utilization improves to 81.7%. Moreover, the algorithm maintains stable performance under resource scaling and network uncertainty, demonstrating strong robustness and adaptability. This research provides a new technical pathway and algorithmic foundation for building efficient, intelligent, and scalable distribution communication networks, offering theoretical and engineering value for future smart grid scheduling.