PMWQ: A Priority-Based Multi-Objective Task Offloading Optimization Algorithm in Computing Power Network

The computing power network (CPN) overcomes the limitations of single-point computing performance by integrating computing and network resources. Latency and energy consumption during the task offloading process are key performance indicators for evaluating the CPN. However, most existing algorithms execute tasks based on their order of arrival, which fails to meet the requirements of latency-sensitive tasks. Additionally, they lack adjustment of offloading preferences when the number of network devices dynamically increases, leading to uneven task distribution and increased energy consumption. To address these issues, this paper proposes a priority-based multiobjective weight-adjusted Q-value task offloading optimization algorithm (PMWQ). First, tasks are prioritized before offloading to ensure that latency-sensitive tasks are processed first, thereby reducing latency. Second, an adaptive dynamic weight Q-value adjustment strategy is employed to balance the load among devices at different levels and reduce energy consumption. Moreover, the algorithm optimizes the feedback mechanism through a dual Q-function table to improve decision accuracy. Simulations show that, with 210 devices, the PMWQ algorithm outperforms baseline methods by increasing task success rates by 3.72%-17.47%, reducing average waiting delays by 58.81%-75.93%, lowering average energy consumption by 4.32%-10.24%, and enhancing average CPU utilization by 14.82%-27.92%.