Multi-Agent Deep Reinforcement Learning for Joint Task Offloading and Resource Allocation in IIoT with Dynamic...

Area of Science:

• Computer Science
• Artificial Intelligence
• Distributed Systems

Background:

• The proliferation of Industrial Internet of Things (IIoT) terminals increases task concurrency and resource demands in edge computing.
• Existing task scheduling methods struggle to balance offloading, resource allocation, and priority adaptation in competitive edge environments.
• Cloud-edge-end collaborative computing offers potential but requires joint optimization for efficiency.

Purpose of the Study:

• To propose a novel two-stage dynamic-priority-aware joint task offloading and resource allocation method (DPTORA) for edge computing.
• To address the limitations of existing methods in balancing task execution and resource utilization under dynamic priorities and resource constraints.
• To enhance the efficiency and performance of task scheduling in complex IIoT environments.

Main Methods:

• Developed a two-stage approach: Stage 1 uses an improved Multi-Agent Proximal Policy Optimization (MAPPO) with a Priority-Gated Attention Module (PGAM) for dynamic priority-aware offloading.
• Stage 2 formulates resource allocation as a single-objective convex optimization problem, solved via the Lagrangian dual method.
• Integrated cloud-edge-end collaborative computing principles for cross-layer task offloading.

Main Results:

• The proposed DPTORA method significantly reduces task latency and energy consumption.
• DPTORA demonstrates a higher task completion rate compared to existing multi-agent reinforcement learning baselines.
• The MAPPO-PGAM integration improved the robustness and accuracy of offloading strategies under dynamic priorities.

Conclusions:

• DPTORA effectively balances task execution and resource utilization in resource-constrained edge computing scenarios.
• The joint optimization of task offloading, resource allocation, and priority adaptation is crucial for efficient IIoT task scheduling.
• The proposed method offers a significant advancement for intelligent task scheduling in edge computing environments.