Global Optimization Of Dedicated Outdoor Air System With Double Heat Recovery Based On Machine Learning And Model Predictive Control

The energy-saving operation of air-conditioning systems is a crucial pathway for achieving the strategic target of "carbon peak and carbon neutrality" in the China building sector.Among the promising air conditioning systems,the Dedicated Outdoor Air System(DOAS)has great potential to reduce building operating energy consumption as well as improve indoor air quality.However,current DOAS optimization methods mostly focused on local control while lacking consideration of global optimization.This makes it difficult for DOAS to ensure the optimal operating performance of the system.Moreover,existing DOAS simulation studies were more simplified.Most of them were based on the single-zone(single-room)simulation or the idealized local control,thus ignoring the impact of the equipment response delay and the pipeline resistance characteristics on the system performance,therefore,limiting their application.In order to deeply explore the energy-saving potential of DOAS,this paper proposed a comprehensive DOAS global optimization framework based on machine learning,model predictive control(MPC),and efficient Modelica modeling language.The main research results of this paper are as follows:(1)The methods to enhance the performance of prediction models were developed.After data preprocessing,key input variables of the prediction model were selected based on sensitivity analysis to reduce irrelevant dimensions in the training data and improve the quality of the model training.And then,the structure parameters of the prediction model were optimized using the grid search and cross-validation methods to improve the prediction accuracy.(2)A optimal matching method of the prediction model and the optimization algorithm was proposed to improve the performance of MPC.Firstly,the performance of different combinations of the prediction model and the optimization algorithm was compared and analyzed.Secondly,the optimal combination of the prediction model and the optimization algorithm was obtained based on the trade-off among the prediction accuracy,the optimization accuracy,and the optimization time.Finally,this method was validated through a case study,and the optimal combination was the support vector regression(SVR)and the particle swarm optimization(PSO).This optimal combination would be applied to the global optimization framework.(3)The prediction methods of weather parameters,cooling load,and system energy consumption were developed.The key elements of the global optimization framework were identified,and the energy consumption optimization method based on the disturbance prediction was developed.The prediction errors of the outdoor wet-bulb temperature and the outdoor dew-point temperature in Guangzhou using the optimized SVR model were as low as0.21℃ and 0.36℃(root mean square error),respectively.The prediction errors of the cooling load and the system total energy consumption of the DOAS using the optimized SVR model were as low as 2.73% and 3.12%(mean absolute percentage error),respectively.These predictions all obtained high accuracy.(4)The DOAS with double heat recovery and its robust control methods were proposed,and its efficient simulation platform was built.Firstly,to address the issue of unreasonable energy consumption in conventional DOASs,a novel DOAS with double heat recovery was proposed.Secondly,the control characteristics of the novel system were analyzed,and the decoupling control method of heat and humidity was proposed,i.e.,the indoor temperature is controlled by the terminal equipment and the indoor humidity is controlled by the fresh air system.Finally,based on the form and the control strategies of the novel system,the flexible and efficient simulation platform for the novel system was established by coupling the physical system models,the control system models,and the data post-processing modules using the efficient Modelica modeling method.(5)The energy-saving effect and global optimization control performance of the DOAS with double heat recovery were verified.In the cooling season(May to September)of the Guangzhou office building,the DOAS with double heat recovery can save 60.5% of the total energy consumption compared with the conventional DOAS.After applying the global optimization framework,the DOAS with double heat recovery can further save 20.3% of total energy consumption.The method of variable optimization boundaries based on the outdoor dew-point temperature prediction was proposed to enhance the robustness of the indoor air parameters control.During the cooling season,the indoor temperature in each air-conditioning zone was stably controlled within the range of 25.4℃-26.1℃,and the indoor humidity was stably controlled within the range of 57%-65%,which ensured good indoor thermal comfort.The optimal setting regularities of the chilled water supply temperature,the approximation of the cooling water supply temperature,the mass flow rate of chilled water,and the mass flow rate of cooling water were analyzed.These optimal setting regularities indicated that the specific system forms should be analyzed on a case-by-case basis rather than directly adopting experience or common sense.The global optimization framework and research methodologies proposed in this paper can provide important guidance for the operating optimization of air-conditioning systems.