| Ground source heat pump is a kind of HVAC technology which directly uses the low-grade heat energy of underground rock and soil.It uses underground rock and soil as heat source(heat sink)to input a small amount of high-grade electric energy into the system,so that the heat energy is transferred between underground rock and ground buildings,and the heating(heating condition)and cooling(refrigeration condition)of building and indoor air are realized.In the ground source heat pump system,the buried pipe heat exchanger is the core energy exchange equipment.Through this equipment,the ground source heat pump can extract heat from the rock and soil in winter to heat the building,and the heat in the building is discharged to the underground rock and soil in summer to realize refrigeration.In engineering practice,how to establish the heat transfer analysis model of ground heat exchanger with high calculation accuracy and fast calculation speed has great practical significance to the design,operation monitoring and optimization management of ground source heat pump system.Artificial neural network(ANN)is a kind of technology which can automatically obtain the relationship between data by self-learning a large amount of existing data.This technology has the advantages of non-linear,non-convexity,non-convex characterization,strong generalization ability and has been widely used in the indusitrial field.The purpose of this paper is to combine the heat transfer process of the ground heat exchanger with the technology of artificial neural network,and apply the technology of artificial neural network to the calculation of borehole thermal resistance of the ground heat exchanger and the heat transfer analysis model of the ground heat exchanger,making full use of the advantages of the artificial neural network,such as strong self adaptability,non-linear and strong generalization ability,to establish a high calculation accuracy and high feasibility of heat transfer model of ground heat exchange base on artificial neural network technology,which can meet the needs of engineering practice.The prediction model of dimensionless borehole thermal resistance(2πK_g·R_b)is established by using the BP(Back-Propagation)neural network in the calculation of the borehole thermal resistance in the ground heat exchanger.On this basis,the optimal network topology can be found by changing the neural network structure.The results show that the optimal topological structure of BP neural network is 3＿5＿7＿1,the correlation coefficients of the training set and the validation set are the highest,which are0.9992 and 0.9999,respectively.At this time,the correlation coefficient of the test set are0.99992.The minimum relative error between the dimensionless borehole thermal resistance and the reference value predicted by the BP neural network of this structure is-1.68%,the maximum relative error is 2.30%,and the average absolute error is 0.19%.Compared with other borehole thermal resistance experiences or calculation fitting formulas,the BP neural network calculation model constructed in this paper has better prediction calculation accuracy.In this paper,LSTM(Long Short-Term Memory)neural network is used to establish the heat transfer analysis model of ground heat exchange to predict the fluid temperature at the outlet of the buried tube.On this basis,the effects of different sample input mode and data window size and other factors on the performance of LSTM network are investigated,and the network is compared with the CFD numerical analysis model of ground heat exchanger.The results show that the trained LSTM neural network has the best performance on the test set under the conditions of input training samples in 4)mode,penalty function of 0.001,learning rate attenuation,data window size of 30,number of LSTM neural units in the hidden layer of 10 and so on.The average absolute error and the maximum relative error are 0.00563%and 3.6%,respectively.The errors are mainly concentrated in the initial time of each model,and the predicted value at the later time is very consistent with the calculated value of CFD.Moreover,the trained LSTM neural network requires only 6s to predict the outlet temperature of a single ground heat exchanger in 24h operation time.Compared CFD numerical simulation,the calculation time under the same conditions is reduced by 4 orders of magnitude.Hence,the neural network model of ground heat exchanger LSTM established in this paper can well predict the variation of current carrier temperature with time at the outlet of buried tube.Compared with other calculation models,the model has the advantages of high calculation accuracy and short calculation time,which can basically meet the engineering practice needs of ground source heat pump system. |
