Cloud-Edge Cooperative MPC With Event-Triggered Strategy for Large-Scale Complex Systems

TY - JOUR

T1 - Cloud-Edge Cooperative MPC With Event-Triggered Strategy for Large-Scale Complex Systems

AU - Ma, Yaling

AU - Yang, Huan

AU - Zhao, Junxiao

AU - Xie, Huahui

AU - Dai, Li

AU - Xia, Yuanqing

N1 - Publisher Copyright: © 2014 IEEE.

PY - 2025

Y1 - 2025

N2 - This article addresses the challenges of slow computational efficiency and the complexity inherent in nonlinear model predictive control (NMPC) for large-scale systems (LSSs) by leveraging cloud computing. While cloud-based NMPC has the potential to manage numerous variables, nonlinearities, and constraints, existing studies often neglect critical factors, such as reliability and resource efficiency, which can hinder its practical application in LSSs. To overcome these limitations, we propose a cloud-edge cooperative MPC architecture that incorporates mode-dependent dwell time (MDT) with an event-triggered strategy. The proposed architecture features high-fidelity NMPC in the cloud layer and tube-based linear model predictive control (LMPC) in the edge layer, integrated with a Switch Module that employs MDT and an event-triggered strategy. This design effectively combines the abundant resources of cloud computing with the reliability of edge computing, where the tube-based LMPC assumes control in the event of Cloud NMPC failure. A derived feasibility MDT ensures the recursive feasibility of the cloud and edge MPC optimization problems, while a stability MDT, established offline using Lyapunov functions, guarantees exponential stability for the nominal trajectories. Additionally, the actual state trajectories converge to a robust positive invariant set exponentially fast by ensuring that the switching between the two controllers adheres to the feasibility MDT and the stability MDT. To balance control performance and resource efficiency, we devise an event-triggered strategy considering the MDTs for Cloud NMPC, reducing unnecessary communication and computation. Simulations on plug-in hybrid electric vehicles (PHEVs) demonstrate the effectiveness of our theoretical results. A comparative analysis against four other MPC schemes highlights the superiority of the proposed approach in computational and resource efficiency, making it highly suitable for real-time applications and large-scale control tasks while maintaining high control performance.

AB - This article addresses the challenges of slow computational efficiency and the complexity inherent in nonlinear model predictive control (NMPC) for large-scale systems (LSSs) by leveraging cloud computing. While cloud-based NMPC has the potential to manage numerous variables, nonlinearities, and constraints, existing studies often neglect critical factors, such as reliability and resource efficiency, which can hinder its practical application in LSSs. To overcome these limitations, we propose a cloud-edge cooperative MPC architecture that incorporates mode-dependent dwell time (MDT) with an event-triggered strategy. The proposed architecture features high-fidelity NMPC in the cloud layer and tube-based linear model predictive control (LMPC) in the edge layer, integrated with a Switch Module that employs MDT and an event-triggered strategy. This design effectively combines the abundant resources of cloud computing with the reliability of edge computing, where the tube-based LMPC assumes control in the event of Cloud NMPC failure. A derived feasibility MDT ensures the recursive feasibility of the cloud and edge MPC optimization problems, while a stability MDT, established offline using Lyapunov functions, guarantees exponential stability for the nominal trajectories. Additionally, the actual state trajectories converge to a robust positive invariant set exponentially fast by ensuring that the switching between the two controllers adheres to the feasibility MDT and the stability MDT. To balance control performance and resource efficiency, we devise an event-triggered strategy considering the MDTs for Cloud NMPC, reducing unnecessary communication and computation. Simulations on plug-in hybrid electric vehicles (PHEVs) demonstrate the effectiveness of our theoretical results. A comparative analysis against four other MPC schemes highlights the superiority of the proposed approach in computational and resource efficiency, making it highly suitable for real-time applications and large-scale control tasks while maintaining high control performance.

KW - Cloud computing

KW - event-triggered

KW - mode-dependent dwell time (MDT)

KW - nonlinear model predictive control (NMPC)

KW - stability

UR - https://www.scopus.com/pages/publications/105006896875

U2 - 10.1109/JIOT.2025.3573692

DO - 10.1109/JIOT.2025.3573692

M3 - Article

AN - SCOPUS:105006896875

SN - 2327-4662

VL - 12

SP - 31095

EP - 31111

JO - IEEE Internet of Things Journal

JF - IEEE Internet of Things Journal

IS - 15

ER -