Performance Prediction Of Ground Source Heat Pump System Based On Random Forest Model And BPNN Model

The Ground Source Heat Pump(GSHP)technology has been more and more widely applied in China due to its advantages of energy saving and environmental protection.The performance of GSHP system is evaluated mainly through the analysis of measured data,building simulation model and data-driven technology prediction.Firstly,it is very difficult to obtain measured data because many GSHP systems are not installed with monitoring systems.Secondly,the establishment process of simulation model is very complex,so more and more data mining technologies are used in the research on the performance of GSHP systems.Among all of data mining technologies,Random Forest(RF)model has the advantages of strong stability and high accuracy,and has been widely used in many fields,but it is rarely used in the performance prediction of ground source heat pump system.This paper measured the operating data in summer of eight GSHP systems,and then the RF model and Back Propagation Neural Network(BPNN)model are established to predict the COP of the GSHP systems.Measured data include the outdoor air temperature,the inlet/outlet temperature of chilled/heated water on the user side,the inlet/outlet temperature of chilled/heated water of GHE,the volume flow rate of unit in user side,the volume flow rate of GHE,heating load/cooling load,power consumption of GSHP system and load ratio.The correlation analysis indicated that the correlation coefficient of load ratio and heating load/cooling load is 1,so load ratio is determined as a redundant variable and was removed form the input parameters of the model.The input parameters of models include COP historical data,the outdoor air temperature,the inlet/outlet temperature of chilled/heated water on the user side,the inlet/outlet temperature of chilled/heated water of GHE,the volume flow rate of unit in user side,the volume flow rate of GHE,heating load/cooling load and power consumption of GSHP systems,and the output parameters is system COP.In this paper,a prediction model is established for each case,and the data set is divided into training set and prediction set,which are respectively used to evaluate the magic precision and generalization ability The mean absolute error(MAE),the root mean squared error(RMSE)and mean absolute percentage error(MAPE)and the coefficient of multiple determinations(R~2)are used to assess the prediction performance of the models.The prediction result shows that both RF and BPNN models have high accuracy in predicting the performance of the GSHP systems.By comparing the two models,it is found that RF model is superior to BPNN model in model prediction accuracy,and the generalization ability of both models is equal.In the process of model prediction,it is found that BPNN model is better than RF model in predicting the GSHP system with small sample size,but the RF model is less sensitive to the change of sample size than the BPNN model.The distribution characteristics and data quality of samples have a certain influence on the prediction performance of the model.In the prediction of the performance of the GSHP systems,the result shows that the higher the COP value is,the better the prediction performance will be.Moreover,the outliers need to be eliminated to ensure the prediction performance of the model.The final study proves that the prediction accuracy of the two models in the performance prediction of GSHP system is relatively high,both of them have good generalization ability,and the stability of RF prediction model is superior to the BPNN model.