A Distributed Spatial Node Task Offloading and Resource Allocation Method Based on DQN

CN116669105BActive Publication Date: 2026-05-26NORTHEASTERN UNIV CHINA
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310421894.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-04-19
Publication Date
2026-05-26
Estimated Expiration
2043-04-19

AI Technical Summary

Technical Problem

Existing two-stage solution methods ignore the interplay between factors such as channel state, transmission power, and computing resources in aviation edge computing scenarios. This results in the inability to traverse the entire solution space and the acquisition of only approximate optimal solutions, which cannot meet the requirements of rapid mission response in time-varying aviation scenarios.

Method used

A distributed airborne node task offloading and resource allocation method based on DQN is adopted. By decomposing the target problem into continuous variable computational resource subproblems and transmission subproblems, and combining DQN network and artificial hummingbird algorithm for iterative optimization, the decision variables such as offloading mode, channel selection, computational resources and transmission power are optimized to achieve global optimal decision.

Benefits of technology

Under the constraints of maximum latency and energy consumption, it can quickly respond to the needs of intelligent tasks, reduce offloading costs, improve service completion efficiency, adapt to the time-varying characteristics of aviation scenarios, avoid getting trapped in local optima, and make full use of communication and computing resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116669105B_ABST
    Figure CN116669105B_ABST
Patent Text Reader

Abstract

This invention discloses a distributed space-based node task offloading and resource allocation method based on DQN, comprising the following steps: S1: Characterizing the offloading cost using the objective problem formula P1, and decomposing it into a computational resource sub-problem P2 and a transmission sub-problem P3 concerning continuous variables; S2: Using the inverse value of the objective problem P1 as the activation function of the upper-layer DQN network, and making decisions on the offloading mode, offloading destination, and channel selection under given initial transmit power and computational resource allocation conditions, thereby optimizing discrete variables; S3: Inputting the upper-layer optimal discrete offloading decision as a given condition to the lower-layer decision, and using an iterative algorithm based on the Cook-Taun condition and an artificial hummingbird algorithm to solve the convex optimization problem P2 and the non-convex optimization problem P3 respectively, thereby optimizing continuous variables; S4: If |V UP -V DOWN If |V is greater than the threshold δ, return to S2, and use the optimal solution of computational resources and transmission power obtained from the lower-level decision as a given condition to return to the upper level, repeating the iteration until |V UP -V DOWN |≤δ, enables task unloading and resource allocation.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Multi-unmanned aerial vehicle air charging and task scheduling method based on deep reinforcement learning

    CN114048689A

  • Distributed computing unloading method and device based on deep reinforcement learning

    CN114449584A