A deep reinforcement learning optimization method of edge task offloading combined with evolutionary algorithm

By combining the SAC and PSO algorithms for edge task offloading, the dynamic and uncertain issues of task offloading and resource allocation in mobile edge computing environments are solved, achieving low latency, efficient resource utilization and energy management, and is suitable for various MEC application scenarios.

CN122240202APending Publication Date: 2026-06-19ZHONGSHU (XIAMEN) INFORMATION TECH CO LTD
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
ZHONGSHU (XIAMEN) INFORMATION TECH CO LTD
Filing Date
2026-03-23
Publication Date
2026-06-19

AI Technical Summary

⚠Technical Problem

Task offloading and resource allocation in mobile edge computing environments face dynamic and uncertain challenges, resulting in high latency and energy consumption, which traditional methods struggle to address effectively.

⚗Method used

A deep reinforcement learning optimization method combining edge task offloading and evolutionary algorithms is adopted. The SAC algorithm is used to learn the dynamics of the environment and the PSO algorithm is used to optimize resource allocation, so as to realize the automatic adjustment of task offloading decision and hyperparameters.

🎯Benefits of technology

It significantly reduces device power consumption and computing latency, improves system throughput and resource utilization, adapts to different MEC application scenarios, eliminates the need for retraining for specific scenarios, and reduces training costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

This invention discloses a deep reinforcement learning optimization method combining edge task offloading and evolutionary algorithms. Addressing the high dynamism and uncertainty of MEC environments, as well as the problems of high training costs, weak adaptability, and insufficient generalization in existing deep reinforcement learning methods for task offloading, this invention proposes a two-stage optimization scheme integrating deep reinforcement learning and evolutionary algorithms. The first stage uses the deep reinforcement learning SAC algorithm for task offloading decision-making, comprehensively considering factors such as edge node computing power, load status, latency, and energy consumption to select the optimal offloading node. The second stage optimizes the hyperparameters of the SAC algorithm and implements resource allocation through the particle swarm optimization (PSO) algorithm. This invention can quickly adapt to dynamic changes in nodes, task requirements, and fluctuations in resource availability in MEC environments, improving resource utilization and system throughput, reducing device energy consumption and computational latency, and exhibiting good dynamic adaptability, exploratory ability, and generalization performance.
Need to check novelty before this filing date? Find Prior Art