Dynamic Resource Scheduling for Deterministic Communication, Computation, and Control Integration in Industrial Cyber–Physical Systems

Due to insufficient resource synergy and uncontrollable transmission delay in existing industrial networks, the construction of artificial intelligence generated content (AIGC) services for industrial cyber-physical systems (ICPS) faces challenges. To this end, this paper presents a Deterministic Communication, Computation, and Control (Det3C) integration network architecture, which is composed of four layers, i.e., end, edge, core, and cloud layers. Specifically, the distributed ICPS domains of end layers are responsible for local training of AIGC models. The updated model parameters are aggregated in edge severs and then transmitted to a cloud sever for global aggregation through wide-area core networks. Besides, we adopt a cyclic queuing and forwarding (CQF) mechanism and design an enhanced cycle specified queuing and forwarding (E-CSQF) mechanism to ensure the deterministic transmission of parameter flows. To obtain optimal ICPS domain selection and computing and temporal (i.e., queue) resource allocation decisions for flow processing and transmission, we formulate a joint optimization problem with the objective to minimize the overall delay of AIGC models. Due to complicated coupled constraints among decisions, a particle swarm optimization (PSO)-based cross-domain computing orchestration scheme is proposed to reduce computing delay and a proximal policy optimization (PPO)-based deterministic flow scheduling scheme is designed to optimize transmission delay while improving scheduling success ratio. Simulation results demonstrate that the proposed Det3C can significantly reduce the overall delay compared with benchmarks.