Cloud-Edge Cooperative MPC for Large-Scale Complex Systems With Input Nonlinearity

Nonlinear model predictive control (NMPC) is a promising approach for controlling large-scale complex systems (LSS) that exhibit nonlinearity and constraints. However, its computational and real-time limitations hinder its widespread adoption. To address this challenge, we propose a cloud-edge cooperative model predictive control (MPC) scheme that overcomes these limitations while ensuring the desired control performance. Specifically, our proposed approach involves designing a cloud-based NMPC with a high-fidelity nonlinear model for the cloud layer. Meanwhile, the edge layer is equipped with a simplified backup linear model predictive control (LMPC) that uses a linearized model based on constraint tightening to mitigate model mismatch errors. Additionally, we develop an automatic strategy that employs a sliding weighted average method to switch between the cloud and edge controllers, enhancing the system's reliability under non-ideal networking conditions. We provide a thorough analysis of the recursive feasibility and asymptotic average performance of the control scheme with different prediction models in the cloud and edge layers. To validate our approach, we apply it to a charging system for plug-in hybrid electric vehicles (PHEVs). Furthermore, we compare the performance and computation efficiency of our proposed cloud-edge cooperative MPC scheme with four other MPC schemes.