Research On Hybrid Modeling Of High Temperature Cascade Heat Pump System Based On Neural Network

With the approaching of carbon peak and carbon neutralization period,technologies of energy saving and emission reduction have attracted much attention in all industrial sectors.Heat pump is one of the important technologies to improve the quality of low-grade residual heat and increase energy efficiency.The high-temperature cascade heat pump system should be worthy of attention due to its large temperature lift to broaden the scope of heat users.Compared with experimental research,simulation is an efficient,convenient and low-cost research method.In the current era of big data,the application of machine learning in simulation has developed rapidly.Considering the complexity of two-stage cycle thermodynamic process,hybrid modeling of the high temperature cascade heat pump system based on neural network is studied by combining experiment and simulation.A water source high temperature cascade heat pump system test bench was built.Taking R245 fa and R134 a as the working fluid in high temperature cycle and low temperature cycle the performance test was carried out at different outlet temperature of condenser under the condition that the heat source temperature was 20℃,30℃ and 40℃,The experimental results showed that the variation of working condition affected on system performance.A large number of historical experimental data are obtained,which provide data support for system modeling and guarantee for model performance verification.BP artificial neural network is a widely used data-driven modeling method.In this paper,according to the characteristics of BP artificial neural network modeling,the corresponding optimization of neural network is carried out in the process of modeling.Using correlation analysis to the optimization of input and output parameter selection,trial and error method was applied to BP artificial neural network among the number of hidden layer nodes,BP artificial neural network for initial weights and threshold of random assignment problem.Longhorn beard search algorithm a global optimization search algorithm,is employed to obtain the optimal initial weights of BP artificial neural network threshold,Which can improve the performance of artificial neural network.Mathematical models of individual components are established according to their characteristics.Neural network models are employed in compressors,electronic expansion valves and cascade heat exchanger because mechanism modeling in these components is difficult and inaccurate.Mechanism models are employed in plate heat exchangers due to their accuracy.The results show that the optimized BAS-BP artificial neural network has better prediction accuracy and better extrapolation.Based on models of individual component,a systematic hybrid simulation model of is formed by connecting each according to the operation principle of cascade heat pump system.The experimental data were used to verify the systematich ybrid model.Based on the experimental data of the heat source temperature at 20℃,30℃ and 40℃,the performance parameters of the overlapping high temperature heat pump system were simulated when the outlet temperature of the high temperature condenser was 112℃,114℃,116℃,118℃ and120℃.The simulation results of the input power of the system show that when the heat source temperature is 20℃,30℃ and 40℃,the average relative errors of the simulated input power of the low-temperature compressor are-1.16%,1.37% and 2.86%,and the maximum relative errors are 2.85%,2.27% and3.27%.The average relative errors of the input power simulation values of the high temperature compressor are 1.68%,106% and 2.83%,and the maximum relative errors are 4.72%,1.76% and4.84%.When simulating the outlet temperature of different high-temperature condensers,the average relative errors of the simulated heat production value of high-temperature condensers are-2.52%,-1.36% and-1.56%,and the maximum relative errors are-5.16%,-2.53% and 2.81%.When the heat source temperature is 20℃,30℃ and 40℃,the average relative error of COP simulation value of the system is-3.26%,-2.06% and-2.71%,and the maximum relative error is-5.21%,-3.18% and-4.58%when the outlet temperature of different high temperature condenser is simulated.In general,the simulation errors of the hybrid model of the overlapping high temperature heat pump system under different working conditions are within ±6%,which indicates that the hybrid model has good accuracy and good extrapolation,so the hybrid modeling in this paper has a certain guiding role in the development and popularization of the application of the overlapping heat pump system.