Global optimization control based on dynamic constraint conditions of a dedicated outdoor air-conditioning system using machine learning and model predictive control

The advantages of a dedicated outdoor air-conditioning system (DOAS) include lower operating energy consumption and better indoor air quality. The global operation optimization of DOASs has rarely been a focus of previous research, resulting in the inapparent energy-saving effect. In this paper, the machine learning-based model predictive control (MLB-MPC) with dynamic constraint conditions was proposed for optimizing the DOAS to obtain the highest operation efficiency. The support vector regression (SVR) was used to predict the total energy consumption of the DOAS and its disturbing factors. The particle swarm optimization was utilized to search for optimal setpoints of the crucial operating parameters. The dynamic constraint conditions were applied to ensure the cooling capacity of the DOAS. The case study based on Modelica simulation demonstrated that: 1) The proposed MLB-MPC with dynamic constraint conditions can ensure robustness for the DOAS, e.g., the indoor temperature and humidity of different air-conditioning areas were stably controlled in ranges 25.4°C to 26.1°C and 57% to 65%, respectively, achieving a good indoor heat comfort. 2) The root mean squared errors using SVR for the outdoor wet-bulb temperature and the outdoor dew-point temperature were 0.21°C and 0.36°C, respectively; the mean absolute percentage errors using SVR for the cooling load and the DOAS energy consumption were 2.73% and 3.12%, respectively, obtaining high prediction accuracies. 3) The system performances were significantly improved after optimization, e.g., during the cooling season, the DOAS’s COP enhanced from 3.14 to 3.91, with an improvement of 24.5%, demonstrating a significant energy-saving effect. The proposed global optimization control would provide a valuable reference for optimizing the operation efficiency of DOASs. • The energy consumption of a dedicated outdoor air-conditioning system (DOAS) was optimized using the machine learning-based model predictive control (MLB-MPC). • The dynamic constraint conditions based on the prediction of outdoor dew-point temperature were proposed. • The DOAS energy consumption and its disturbing factors were predicted accurately by the support vector regression (SVR) model. • The Modelica-based platform for simulation and optimization of the DOAS was established.