Ground Thermal Properties By Neural Networks on Well-log Datasets

This research focuses on using well-log data from abandoned oil and gas wells to predict ground thermal properties and assess geothermal potential. The study utilizes an Artificial Neural Network (ANN) approach to predict thermal conductivity in three oil fields. Initially, thermal conductivity is calculated using recommended values for each mineral from the UNIPD Cheap-GSHPs thermal database and a geometric average (Clauser, 2006) in the first oil field. It is then validated by laboratory measurements, showing a strong correlation. The ANN model incorporates seven training datasets, including Gamma Ray (GR), Neutron Porosity (NPHI), Sonic Travel Time (DT), Bulk Density (RHOB), Total Porosity (PHIT), and Resistivity logs, with thermal conductivity as the associated log. The model is trained in the first oil field and validated, achieving a correlation coefficient of over 95%. Performance metrics (MAE, MSE, and R-squared) for thermal conductivity predictions are reported for all oil fields. The study also explores the relationship between predicted thermal conductivity and other properties, such as porosity and density. The study identified three oil fields with different properties for geothermal applications. The first oil field, with a thermal conductivity of about 2.7 W/m · K and a temperature of 115◦C, is suitable for medium geothermal purposes. The second oil field, made up of limestone, has a thermal conductivity of 2.8 W/m · K and a temperature of138◦C, making it ideal for high geothermal applications. The third oil field, composed of dolomite, exhibits thermal conductivity ranging between 2.9W/m · K and 3.0W/m · K and temperatures between 85◦C and 105◦C, making it suitable for shallow geothermal purposes