A Dual-Branch Deep Deterministic Policy Gradient-Based Approach for Joint Task Orchestration and Channel Allocation
Year: 2025
Type: conference-paper
Abstract: In the application scenarios of Internet of Things and Non-Terrestrial Networks convergence, the task orchestration problem faces challenges such as a large number of tasks, limited channel resources, and a highly dynamic and complex environment. Traditional rule-based methods exhibit insufficient adaptability when dealing with multidimensional factors such as task priority, channel state, and minimal task transmission time. To address this, this paper proposes an intelligent task orchestration method based on Deep Deterministic Policy Gradient (DDPG), which models the task scheduling process as a reinforcement learning problem. By constructing a state space that incorporates task data volume, priority, timestamp, as well as channel bandwidth and signal-to-noise ratio (SNR), a dual-branch Actor network is designed to output task priority weights and channel suitability scores respectively, enabling dynamic mapping between tasks and channels. The Critic network is employed to evaluate action values, while the experience replay and soft update mechanisms are integrated to ensure training stability. The designed reward function comprehensively considers priority fulfillment, channel utilization, and transmission delay, guiding the model to learn the optimal policy in dynamic environments. The proposed method can achieve efficient task scheduling and channel allocation in complex communication environments, demonstrating strong self-adaptability and robustness. It provides a novel solution for resource optimization in intelligent edge gateways within NTN systems.
Authors: Ji Li, Rongke Liu, Ling Zhao, Shaopeng Zhu, Haiyao Xie
Institution: Beihang University
Language: English
Topic: IoT and Edge/Fog Computing
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7127046149
