Application of machine learning and magnetotellurics to aid in subsurface characterization of petroleum and geothermal reservoirs.


Energy is the foundation of society and with future energy demand expected to increase significantly over the next few decades, solutions contributing to future energy resources are of high interest scientifically, geopolitically, and economically. Data analytics and machine learning provide useful tools to more efficiently and cost-effectively produce petroleum and geothermal resources vital for our energy future. Supervised and unsupervised machine learning can aid in the prediction of sedimentological and reservoir attributes in wells lacking core control to better and more efficiently characterize subsurface petroleum reservoirs. Using tree-based machine learning models, core-observed depositional attributes from the Late Devonian Duvernay Formation in Alberta, Canada may be predicted in wells lacking core control when class proportion and thickness conditions are met. Unsupervised machine learning technique, non-negative matrix factorization with k-means clustering (NMFk), automatically identifies reservoir significance, undetected through the traditional deterministic modelling, within the Duvernay Formation without calibration to core observations. The application of NMFk with petrophysical data may assist in highlighting intervals of interest in advance of core descriptions reducing observer inconsistency and bias and enhancing the quality and relevance of core description for reservoir correlation and mapping. Machine learning methods provide precise, consistent, and objective petrophysical interpretations and reservoir characterization, and increases the consistency and accuracy of resource assessment for petroleum exploration and production. Unsupervised machine learning and magnetotellurics are useful analytical tools to assess prospective geothermal resources in the Tularosa Basin of south-central New Mexico based on heat flow, temperature, porosity, and permeability. The unsupervised machine learning method, NMFk, identifies locations with the highest likelihood of geothermal success, and the passive geophysical method, magnetotellurics can detect subsurface geothermal prospects. The integration of NMFk and MT can provide a 3D assessment of heat flow, temperature, and permeability for geothermal exploration. This research provides innovative methods to aid in the development of efficient and cost-effective approaches for future energy exploration and production.