Modeling of a direct expansion geothermal heat pump using artificial neural networks

Abstract The real potential for energy savings exists in heating, ventilation, and air conditioning systems in general, and especially in geothermal heat pumps systems. Recent studies indicate that a mere 1% improvement in the efficiency of such systems generates millions of dollars in savings at the national level. This efficiency can be optimized when better control strategies are implemented. A first step in the control and optimization process is to establish a model that describes the system's behavior. In this study, artificial neural networks were selected for modeling a particular type of heat pump called direct expansion geothermal heat pump because the ground heat exchanger is a component of the heat pump, and thus directly plays the role of condenser or evaporator according to the operation mode. The data collection methodology and the algorithms used for training are presented. Of the four algorithms tested in this study with variable numbers of neurons in the hidden layers, the model obtained using the Levenberg–Marquardt (LM) algorithm with 28 neurons in the hidden layer appears to be the best, with an average coefficient of multiple determinations of about 0.9991, an average RMS of 0.16330, and an average COV of 2.9319.