Maximizing Information Value Through Machine Learning: AI-Powered Petrophysical Solution for the Mature Uzen Field

Abstract

Abstract In response to declining oil production in Kazakhstan's Uzen oilfield (over 9000 wells), a renovation program was launched. The program's objective is to review the geological model, FDP, and address petrophysical challenges associated with a significant volume of data, time constraints, resource limitations, and poor log data quality. The conventional approaches proved inadequate to handle these challenges efficiently. To overcome these obstacles, an AI/ML-based solution was employed to perform logs quality check, data reconstruction, and address petrophysical complexities. Our study utilized an Automated Log Editing solution for the entire process of petrophysical data processing. We started with log data preparation stage which include harmonization of mnemonics and units of measurements. Subsequently, we employed a machine learning (ML) approach, encompassing further exploratory data analysis to remove non-physical data points, followed by outlier detection to identify anomalous intervals. To establish well-level proximity, we utilized a similarity analysis algorithm based on petrophysical properties, and the resulting relationships were leveraged for log reconstruction. The analysis of the project results revealed that approximately 15% of the wells required significant log corrections. These corrections primarily focused on washed-out intervals, where inaccuracies were observed in the logs. Additionally, intervals with incorrect neutron porosity calculations were identified and promptly corrected. The reliability and accuracy of the project results were validated by comparing them with the high-quality NMR logs from the three wells. The NMR logs served as a valuable reference for formation evaluation, ensuring the integrity of the project outcomes. The utilization of the automated log editing solution proved instrumental in reducing data processing time and minimizing bias during log reconstruction. By swiftly identifying and loading missing data, the project enhanced the efficiency of the data auditing process. The automated approach demonstrated its effectiveness in identifying and correcting log discrepancies. The successful implementation of the project highlights the importance of utilizing automated log editing solutions for accurate data auditing and log corrections in well logging. By incorporating high-quality NMR logs as a validation tool, the project outcomes were verified and ensured to meet the required standards of reliability. This approach contributes to improving the overall quality and reliability of log data in formation evaluation. This paper will present novel information by demonstrating the successful utilization of historically available Soviet Union data, including logs like SP, acquired over a 60-year. The use of custom-built and adapted algorithms in the project sets it apart from previous research in the field. The findings of this study will provide valuable insights to professionals in the industry, offering a new approach to leveraging old data, reducing the requirement for additional wellsite operations, and minimizing associated CO2 emissions.