Xinrui Li, DeQuan Kong
To address the energy efficiency degradation of magnetic levitation chillers caused by model mismatch and communication latency under variable operating conditions, this paper proposes a Koopman-operator-enabled adaptive model predictive control (MPC) method supported by cooperative edge computing. First, a distributed edge computing architecture encompassing the compressor, condenser, evaporator, and electronic expansion valve is established, in which graph theory is employed to characterize the inter-node topology and information exchange mechanism, thereby overcoming the data congestion and single-point-of-failure limitations inherent in conventional centralized control schemes. Second, Koopman operator theory combined with the online Extended Dynamic Mode Decomposition (EDMD) algorithm is adopted to lift the nonlinear thermodynamic dynamics of the chiller into a high-dimensional linear subspace, achieving global linearization across the full operating envelope together with real-time adaptive identification of the model parameters. On this basis, a distributed cooperative optimization strategy based on the Alternating Direction Method of Multipliers (ADMM) is developed to solve the system-level energy-efficiency maximization problem while strictly maintaining the magnetic-bearing safety margins and satisfying the surge boundary constraints. Simulation and experimental results show that, compared with conventional PID control and fixed-parameter MPC, the proposed method improves the Coefficient of Performance (COP) by approximately 12.4% under part-load conditions, accelerates the dynamic response by 25%, and enhances the model prediction accuracy by 18.6%, effectively realizing edge-intelligent cooperative control and energy-efficient operation of magnetic levitation chillers.