Heterogeneous Resource Scheduling in Computing Power Networks: A Method Based on Improved D3QN

With the continuous development of new network services and the increasing demand for computing power, large-scale Compute Power Networks (CPN) has become a research hotspot. When a large number of task requests are uploaded to CPN, the power nodes may face a high load, resulting in these tasks experiencing significant delays or even being discarded at the deadline. In order to solve the problem of efficiently scheduling heterogeneous resources in CPN, a computing power model is proposed to realize the unified abstraction of different computing resources. Then, an Improved Dueling Double Deep Q Network for Heterogeneous Resource Scheduling (ID3QN-HRS) is proposed. By introducing priority experience playback into the D3QN framework, it solves the joint optimization problem of resource scheduling under multiple constraints. The experimental results show that compared with the benchmark experiments, our algorithm significantly reduces the number of dropped tasks, the total delay, and the power consumption.