Accurate identification of soil thermal parameters and groundwater flow from thermal response tests

The soil heat transfer parameters constitute the key information required for designing ground source heat pump (GSHP). Due to the simplification of the common analytical models, traditional methods are difficult to achieve accurate identification of soil thermal parameters and seepage velocity simultaneously. In this work, a high-precision method to simultaneously identify soil thermal conductivity, volumetric heat capacity, and seepage velocity based on deep neural network is proposed. Through the inversed orthogonal method, the training and validation samples are obtained from a large number of thermal response tests (TRTs) on a full-scale simulation platform. The accuracy of this method was verified by comparing identification results with the true values. Meanwhile, the uncertainty of identification results under different noise conditions was quantified, and the impact of test duration was discussed. The results showed that when the maximum random noise is 0.1 degrees C, the identification errors of the thermal conductivity, volumetric heat capacity, and seepage velocity are only 1.14 %, -4.34 %, and -3.08 %, respectively. The identification reliability can be improved by obtaining the average value of the results under multiple tests and extending the test duration. When the test duration increased from 50 to 100 h, the uncertainty of the identified parameters reduced by 54.57 %, 48.41 %, and 65.70 %, respectively.