Ensemble Machine Learning for Real-Time Sonic Log Prediction in Geothermal Exploration

Abstract Geothermal energy represents a sustainable and renewable resource with the potential to significantly contribute to the global energy demand Manzella, 2017. Accurate subsurface characterization is essential for successful geothermal exploration and reservoir development. Sonic logs, providing information about subsurface rock properties and formation mechanics, play a crucial role in this characterization process. However, in real-time geothermal exploration, the traditional method involving the use of sensors is most times affected by the high subsurface temperature conditions, thereby returning missing sonic log values for some important depth intervals Joshi et al., 2023. Hence, in this study, an ensemble machine-learning approach was adopted for the real-time prediction of Sonic compressional (ΔTc) and Shear (ΔTs) logs. The ensemble method involved specifically, the Random Forest (RF), K-neighbors (KN), and support vector (SVM), to predict missing sonic log values Dietterich, 2000. By leveraging data from the Utah Forge project, input parameters such as Depth, Cartridge temperature, Caliper, Gamma-ray, Generalized Borehole temperature, Intermittent transit time, and Poison's Ratio Ma, 2019 were used to build the prediction model. The R2 and Root Mean Squared error (RMSE) were used as evaluation metrics between the predicted values and actual values. The results indicated that the developed sonic log models had good generalization on the test data set, as each individual models showed high R2 score above 0.9 and lower RMSE below 0.1, further optimization using ensemble voting regression indicated improved accuracy in model performance as overall R2 score and RMSE values rose above 0.9 and further below 0.01 respectively. In general, the SVM showed better accuracy in model performance on test data. Hence, the result of this study provides insights into the application of machine learning in the prediction of sonic logs and helps provide other associated inferences that can guide the decision-making process for geothermal drilling operations.