A dynamic simulation model of a transcritical CO2 parallel compression system was established using GT-SUITE simulation software to explore an efficient control method for the transcritical CO2 parallel compression system. Based on the system performance dataset obtained by the simulation

the second-order polynomial model and the neural network model were established and compared as the system performance prediction models. Based on the neural network model

a model predictive controller for the transcritical CO2 parallel compression system was developed

and the performance of the controller in terms of the stability

high efficiency

and real-time control of the system was studied. The results show that under the action of the model predictive controller

the system can reach a stable operating state within 150 s for different cooling conditions. The performance of the system using model predictive control is 13.3% higher than that using fixed value control. The simulation verifies that the proposed model predictive control strategy is feasible and optimizes the real-time control performance of the CO2 parallel compression system; the overall performance is improved by 7.3% compared with the fixed value control under the given working conditions. The control strategy proposed in this study is significant for the use of machine learning methods in designing system controllers to improve the performance of heat pump air-conditioning systems.