A Data-Driven Koopman–MPC Framework for Multi-Actuator Control of Air-Handling Units with Edge-Deployment Considerations

Abstract: Air-handling units (AHUs) are major energy consumers in HVAC systems due to their coupled actuators and constrained multi-input multi-output operation. This paper proposes a data-driven Koopman–MPC framework that identifies lifted linear dynamics from building management system (BMS) data using EDMDc, enabling constrained MPC design for nonlinear AHU behavior. The approach is validated on a residential AHU in Stockholm, Sweden. Results demonstrate the feasibility of deploying Koopman-based predictive control using operational data, and a BMS-integrated edge implementation workflow is presented for practical building applications.