Achieving Intelligent and Trusted Task Scheduling in Wireless Computing Power Networks

In the intelligent era of the internet of everything, the explosive growth of data volume makes distributed computing power encounter the bottleneck of insufficient resources. Breaking the isolation of computing power and efficiently using it is a huge challenge. The emergence of Computing Power Networks (CPNs) integrates the three-tier architecture, connecting cloud, edge and end computing power resources through the network. Task scheduling is an important method of resource allocation in CPNs. In this paper, we establish an intelligent and trusted task scheduling framework in the wireless CPNs for the resource allocation problem in the Industrial Internet of Things (IIoT) environment. At the same time, we build a reputation model to evaluate the reputation of each task publisher and CPN nodes, and combine blockchain to ensure the accessibility and security of reputation. The task scheduling problem is formulated as a Markov Decision Process (MDP), targeting the joint minimization of network latency and operational cost while sustaining a high task completion rate. A Multi-Agent Deep Reinforcement Learning (MADRL) scheme is introduced to derive an adaptive stationary scheduling policy. The proposed framework is validated through extensive simulations, demonstrating its effectiveness and performance advantages.