A Graph Attention Diffusion Approach for Multi‐Node Flexible Response Optimization in <scp>IIoT</scp> Load Networks

ABSTRACT The rapid expansion of the Industrial Internet of Things (IIoT) enables fine‐grained sensing and distributed actuation for demand‐side flexibility, but also introduces large‐scale heterogeneous load networks that must coordinate in real time under time‐varying connectivity and coupled constraints. Under the Industry 5.0 paradigm, such coordination requires not only efficiency but also resilience and transparency under unreliable communications. To address these challenges, this paper proposes a Graph Attention Diffusion–based Flexible Response Optimization (GAD‐FRO) algorithm, which integrates attention‐guided diffusion operators into a distributed primal–dual framework. GAD‐FRO achieves topology‐aware and communication‐robust coordination without centralized control by adaptively prioritizing reliable neighbors and smoothing information propagation over dynamic graphs. Experiments on four benchmark datasets (Pecan Street, Building Data Genome 2, UK‐DALE, and ACN‐Data) demonstrate that GAD‐FRO achieves faster convergence, lower objective cost, and stronger robustness to packet loss than representative distributed baselines, while approaching centralized oracle performance with low runtime and communication overhead.