Diffusion-Based Latency-Sensitive Task Scheduling for Generative Digital Twins in Industrial Internet of Things

Generative Digital Twins (GDT) integrates Generative Artificial Intelligence (AI) with Digital Twin (DT) technology, providing high-fidelity modeling and real-time decision-making capabilities for smart manufacturing in Industrial Internet of Things (IIoT). However, the complex scheduling of computationally intensive and latency-sensitive generative twin tasks in dynamic scenarios poses significant challenges to this process. To address this issue, this paper proposes a Diffusion-based Q-network generative twin Task Scheduling scheme (DQTS). Specifically, the DQTS scheme utilizes a diffusion model for probabilistic modeling and sample augmentation on historical scheduling data, generating diverse virtual actions to enhance the scheme's exploration efficiency and generalization capability. Meanwhile, by combining collaborative policy learning in the cloud with distributed execution at the edge, it achieves joint optimization of global and local scheduling decisions. Simulation results demonstrate that the proposed DQTS scheme outperforms comparison algorithms in terms of synchronization latency and fidelity. It also exhibits good robustness and stability under environmental disturbances and changes in equipment numbers, effectively satisfying the real-time synchronization and high-fidelity requirements of GDT systems in IIoT.