Abstract
Abstract This study aims to enhance the efficiency of interpreting well log data for reserve calculations at Kazakhstan's Kalamkas field by integrating automation and machine learning techniques. The research focuses on developing and implementing automated workflows and machine learning models to expedite data processing, improve accuracy, and support more informed decision-making in reserve estimation. Additionally, the results are compared to manual interpretations conducted by petrophysicists to evaluate the effectiveness and reliability of the automated approach. Our methodology encompassed several key steps, starting with automatic selection of dataset based on depth, zones, and log availability to ensure its relevance to the study. The resulting database, comprising 3573 wells with structured files, undergoes a preprocessing stage for machine learning suitability. Subsequently, machine learning algorithms were applied to detect and remove outliers, enhancing the overall quality and reliability of the data. Following this, thresholds for distinguishing shale and clean formations were established on automatic bases supporting more accurate lithological interpretation. Finally, we conducted additional analyses to define formation properties. Utilization of machine learning algorithm facilitated the automatic identification of parameters for calculating shale content, thereby reducing the influence of human error and enabling more precise calculations of reservoir properties. The use of a specialized AI platform and the configuration of a connector for analyzing results with the petrophysical software allowed reducing the workflow runtime to one hour for all wells in the field. The results provide a solid foundation for future geological modeling, leading to improved facies and stratigraphic modeling as well as a reduction in uncertainty. The adoption of an automated method for petrophysical interpretation made it possible to efficiently finalize the reservoir reserve calculations for the Kalamkas field and successfully defend them for State Commission on Reserves in an exceptionally short 1-year time. This paper introduces a reliable approach for log processing through machine learning, confirmed by actual field data. It highlights the significance of combining domain knowledge with data science methods to provide value to the petroleum industry in the era of artificial intelligence