A New Supervised AI Logging Lithofacies Identification Technique and its Application in K Oilfield in the Middle East

Abstract Lithology identification is a very important task in reservoir characterization. Through the identification of lithofacies, we can know the state and distribution of oil and gas in the underground reservoir, and provide the basis for oil and gas exploitation and transportation, which is helpful to improve oil recovery and production efficiency. With the development of oilfield, the exploration and development of lithological oil reservoirs become more and more important, and well log lithofacies identification has gained greater significance in reservoir characterization. Conventional well log lithofacies identification faces challenges, due to limited core data and low-quality log curves such as missing local well log curves, wellbore collapse, and anomalous log responses in fault zones. To improve the accuracy of well log lithofacies identification, a new supervised AI technique for well log lithofacies identification was proposed in this study. The workflow mainly includes the following 5 steps: 1) Integrated lithofacies identification based on core description, thin section, high-quality well log, FMI, and CMR; 2) AI well log conditioning based on the high-quality well log curve intervals are selected as learning and training samples; 3) Sensitive log curves were chosen and cutoff were determined based on core-based lithofacies analysis and cross-plot analysis; 4) Log lithofacies identification for no core intervals based on sensitive well logs deep-learning; 5) Two method for lithofacies identification result QC, one method is using cored wells as a blind test well, the other method is using multiple wells to do comparative analysis. This supervised AI logging lithofacies identification technique has been successfully applied in the K oil field in the Middle East. Blind cored well validation indicates that supervised AI well logging facies identification results have a high matching accuracy of 97%, enabling high-accurate and efficient well log facies prediction. Multi-well comparisons demonstrate that the well logging facies identification results in consistent with each other, and match with the results of sequence stratigraphic analysis in the study area.