To improve the control precision of a control system under large and frequent perturbations

the model-based predictive control (MPC) has been developed for industrial heating

ventilation

and air conditioning (HVAC) systems. Because the thermal dynamic characteristics of HVAC systems are time-variant

nonlinear

and contain uncertainties during the control process

conventional controller methods face several challenges. In this study

a nonlinear MPC for variable air volume (VAV) systems is developed and investigated. A nonlinear autoregressive network with exogenous inputs (NARX) and particle swarm optimization (PSO) are employed for the nonlinear MPC. The NARX aims to predict the controlled parameter (room temperature) of the VAV system and the PSO serves as an optimizer to obtain the optimal control variables of the VAV system. By assigning different weight values to the objectives of the cost function

the proposed nonlinear MPC can generate different control solutions considering both the control precision and energy saving of a VAV system. Two scenarios of NARX-based MPC were investigated in an experimental VAV system. Scenario 1 only considers the control accuracy while scenario 2 considers both energy saving and control precision. The experimental results show that the NARX-based MPC under scenario 1 can achieve much higher control precision (±0.5 ℃) at room temperature than that of the constant static pressure (CSP) method with a PI controller. The NARX-based MPC under scenario 2 shows a better energy-saving characteristic

saving 23.7% of energy consumption

as compared to that of the CSP method with a PI controller. This work will contribute to the development of nonlinear MPC and its applications in HVAC systems.