5:15 PM - 7:15 PM
[SCG60-P12] Reservoir Characterization by Machine Learning: Using Core and Logging Data
Keywords:machine learning, core, logging data, permeability
Sandstone formations play an important role as oil and natural gas reservoirs and are also expected to be promising reservoirs for Carbon Capture and Storage (CCS), which has recently been attracting attention as a global warming countermeasure. In evaluating the production capacity of oil and natural gas reservoirs and the storage capacity of CCS reservoirs, it is important to understand the lithology and properties (porosity, permeability, etc.) of the reservoirs, and core tests and geophysical logging analysis are generally used for these evaluations. However, the number of wells from which core samples can be obtained is extremely limited, and while geophysical logging data can provide continuous information in the depth direction, some wells do not necessarily have all the logging items necessary for analysis. In addition, several methods have been proposed to calculate porosity and permeability, each with its own uncertainties. For these reasons, more advanced well data analysis methods have recently been developed by combining AI techniques such as machine learning (artificial neural networks, random forests, etc.) and deep learning. Research using machine learning often discusses methods that utilize all available log data in pursuit of higher prediction accuracy, and the impact of combining different geophysical log data on prediction accuracy has not been fully explored. In the oil exploration field, geological interpretation and logging analysis are often performed independently of each other, which may hinder the integrated understanding and utilization of both. In view of this situation, this study addresses the following two points
(1) To compare and evaluate the effects of different combinations of geophysical logging data on the prediction accuracy of reservoir properties (especially permeability), and to propose an optimal data selection method.
(2) To incorporate geological interpretation (lithological classification based on core observations) into machine learning models using geophysical logging data, and to propose a prediction approach that integrates core and logging data.
In this study, we used data from the TR-1 well acquired by the former Japan National Oil Corporation (now JOGMEC) in Niigata Prefecture in 1992. The wellbore was drilled in a deep marine sandstone-mudstone alternation of the Pliocene Kawaguchi Formation, and the core and its analytical data (permeability, porosity, etc.) and various geophysical logging data (pore size log, natural potential log, gamma ray log, resistivity log, density log, neutron porosity log, sonic log, etc.) were obtained in the section 126 m below the ground surface. The data sets are now available. In addition to these data sets, we have analyzed the lithology and properties of the sandstone reservoir section using newly acquired core sequence photographs and X-ray CT images.
Core samples and geophysical logging data provided by Japan Organization for Metals and Energy Security (JOGMEC) were used in this study. The authors would like to express their deepest gratitude to JOGMEC.
(1) To compare and evaluate the effects of different combinations of geophysical logging data on the prediction accuracy of reservoir properties (especially permeability), and to propose an optimal data selection method.
(2) To incorporate geological interpretation (lithological classification based on core observations) into machine learning models using geophysical logging data, and to propose a prediction approach that integrates core and logging data.
In this study, we used data from the TR-1 well acquired by the former Japan National Oil Corporation (now JOGMEC) in Niigata Prefecture in 1992. The wellbore was drilled in a deep marine sandstone-mudstone alternation of the Pliocene Kawaguchi Formation, and the core and its analytical data (permeability, porosity, etc.) and various geophysical logging data (pore size log, natural potential log, gamma ray log, resistivity log, density log, neutron porosity log, sonic log, etc.) were obtained in the section 126 m below the ground surface. The data sets are now available. In addition to these data sets, we have analyzed the lithology and properties of the sandstone reservoir section using newly acquired core sequence photographs and X-ray CT images.
Core samples and geophysical logging data provided by Japan Organization for Metals and Energy Security (JOGMEC) were used in this study. The authors would like to express their deepest gratitude to JOGMEC.