Study on high-precision identification method of ground thermal properties based on neural network model

Accurately estimating ground thermal properties from thermal response tests (TRTs) is critical to design ground source heat pump system (GSHPS). The traditional method may lead to large errors due to the difference between the heat transfer process described by identification models and actual situation. To avoid it, this paper proposes a high-precision identification method based on artificial neural network (ANN), which can directly establish the nonlinear mapping relationship between thermal response parameters (TRPs) and ground thermal properties. Through the inversed orthogonal method, the training and validation samples are obtained from a large number of TRTs on a full-scale simulation platform that is verified by experiments. The estimation accuracy of traditional method and ANN under different ground thermal properties is studied. The results indicate that the estimation accuracy of traditional method varies greatly under different ground thermal properties, and the relative errors of identifying thermal conductivity and volumetric heat capacity vary from −3.61% to 60.14% and −52.06%–110.20% respectively. The estimation accuracy of ANN is almost not affected by the ground thermal properties, and the corresponding errors range from −7.78% to 0.28% and −1.75%–15.6% respectively. This paper provides a new perspective to reduce error caused by identification model.