Prediction of the ground temperature variations caused by the operation of GSHP system with ANN

Abstract

Ground temperature is an important factor that affects the operation efficiency of ground source heat pump (GSHP) systems. Therefore, it is very important to acquire the ground temperature variations, which can be realized by means of sensor measurements or software simulations. But the monitor system sometimes may be ineffective or inaccurate because of the faulty operation of sensors. And also, the conventional physical models have some limitations in obtaining the detailed ground temperature variations caused by the long term operation of GSHP systems. However, an advantage of the measurement system is that a large amount of monitored data of the ground temperature could be obtained, which provide good solutions for the prediction by black-box methods. In this study, artificial neural network (ANN) was adopted for the prediction of the disturbed ground temperature caused by a vertical GSHP system. Two predictions were conducted with this method. The first is that future variations of the average ground temperature within 100 m deep during the long term operation of GSHP system were predicted with enough accuracy using ANN by 'UseParallel' and 'UseGPU' methods. It has been found that, for the average ground temperature prediction, 'UseParallel' is superior to 'UseGPU' in prediction accuracy, but inferior in calculation speed. The second is that the missing data of the monitored ground temperature variations were supplemented with the ANN model. Mean relative error (MRE), max relative error (MAE), mean square error (MSE) and absolute fraction of variance (R2) of the ANN model are 0.3223%, 9.8999%, 0.5036% and more than 0.9994, respectively. The second ANN model could predict the missing data, but failed in predicting the future detailed ground temperature, which should be further improved in the future.