Estimation of Hydraulic and Thermal Parameters Using Convolutional Neural Network and Hydraulic Tomography

The ground-source heat pump (GSHP) is an efficient thermal exchange system that utilizes natural environmental heat for heating and cooling. Heat exchange efficiency depends not only on factors such as pipe material and diameter but also on groundwater's flow field and soil's thermal parameters. This study aims to estimate hydraulic and geothermal parameters by utilizing convolutional encoder-decoder architecture neural networks and hydraulic tomography, a data collection strategy. The proposed method is named THT-NN. To examine the capability of the THT-NN on parameter estimation, we developed numerical experiments to test THT-NN. Further, to produce the training and validation data pairs, we create a two-dimensional heterogeneous groundwater and heat transport model by TOUGH2 with constant injection patterns and 10000+ realizations of parameter fields. The groundwater heads and temperature collected from the monitoring well groups are used to develop two channels of the input layers, and four parameters' fields (hydraulic conductivity, porosity, heat conductivity, and specific heat) are used to develop four channels of the output layers. Subsequently, the estimated parameters results are examined by R2 and root mean squared error. The performance of the proposed THT-NN is discussed in this study.