As a new network paradigm, Computing Power Network (CPN) interconnect the scattered computing resources through networks to form a collaborative computing power sharing pool, so as to achieve flexible scheduling and global optimization of computing tasks. Existing scheduling research mainly relies on centralized or heuristic-based algorithms, which struggle with scalability, adaptability, and robustness in dynamic environments. Moreover, most existing studies ignore the collaborative optimization of computing tasks and resource allocation in distributed systems, limiting resource utilization. This paper proposed a distributed computing network task scheduling algorithm based on Multi-Agent Deep Deterministic Policy Gradient (MADDPG). The algorithm treats each routing node as an independent agent, enabling distributed collaboration through local state perception and inter-agent strategy interaction. With latency and resource utilization as optimization objectives, it dynamically schedules tasks in response to changing network conditions. Experiments show that this algorithm can achieve flexible scheduling of computing tasks on different topologies and its performance is better than the other two baseline algorithms.