Abstract

This research addresses the challenge of missing data in the energy industry by applying machine learning techniques to predict bottom hole temperature for wells in the United States and Canada. The availability of complete and accurate datasets is critical for assessing geothermal resources, especially as the global energy landscape shifts towards increased reliance on renewable sources. Several variations of the XGBoost algorithm were explored to develop machine learning models for imputing bottom hole temperature independently for the US and Canada. The resulting dataset formed the basis for a comprehensive data analysis workflow, encompassing assessment, cleaning, and preparation for geothermal exploration. The study revealed that the simple XGBoost algorithm outperformed other variations, achieving an average mean absolute percentage error (MAPE) score of 7% for the whole dataset evaluated. These results underscore the efficacy of predictive modeling in handling missing data, providing a reliable tool for assessing subsurface temperature.

Furthermore, we present a workflow that applies the imputed bottom hole temperature bridging data gaps in the Texas and Louisiana Gulf Coast Basin. The workflow integrates cloud and subsurface applications, ensuring a scalable approach to basins throughout North America. The developed workflow offers a preliminary solution for geothermal resource assessment, with implications for sustainable energy exploration. This study provides an overview of the technical aspects of the project, from algorithm selection and model development to the practical assessment of bottom hole temperature datasets for geothermal energy exploration. The research contributes to the understanding of how machine learning can be applied to address data challenges in the geosciences and contributes to the broader objective of advancing geothermal resource assessments.