A Dual-Agent Deep Reinforcement Learning-Based Algorithm for Cloud-Edge Collaborative Task Offloading in Industrial IoT (IIoT)

This paper focuses on energy-efficient task offloading in Industrial Internet of Things (IIoT) environments, a key requirement for enabling sustainable smart manufacturing. Existing approaches often struggle to simultaneously address the dual challenges of dynamic task prioritization and computational resource allocation. To tackle these issues, we propose DaDRLA, a dual-agent deep reinforcement learning framework enhanced with attention mechanisms. Each agent employs a Transformer-based encoder to cooperatively learn task offloading strategies under centralized training with decentralized execution. Experimental results show that DaDRLA achieves energy savings of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$7.4-10.3 \%$</tex> over baseline methods while ensuring faster convergence and greater robustness across various industrial scenarios. This work contributes to more intelligent and energy-aware decision-making in real-world IIoT systems.