Enhancing Reservoir Potential Assessment with Machine Learning Techniques: A Systematic Approach in The X Field

Abstract

Abstract Due to the limited resolution of well logging instruments, particularly older generations, and the presence of complex multiple hydrocarbon systems in a basin, pay zones may be overlooked during measurement and analysis of reservoir characterization. Moreover, while proven and standing the test of time, the conventional approach to stratigraphic correlation and the trial-and-error method for reservoir zone identification are tedious, time-consuming, and expensive. This study proposes a systematic, automated, and accurate reservoir prospect assessment, or reassessment, methodology augmented by machine learning algorithms. The case study investigates S Formation in the X Field, which is marked by a high water cut and the presence of discontinuous parasequence sets. Two machine learning models will be modeled and tuned to X field characteristics. The first machine learning model will assess reservoir stratigraphic relationships, where an unsupervised machine learning model will analyze patterns from log data to geologically describe the reservoir. This can help detect rock formations such as tight shales that conventional interpretation may miss. Third-party stratigraphic data will validate the model results. The second supervised machine learning model will identify hydrocarbon-producing zones. This machine learning model will use resistivity log data and potential zone information from the first model to detect hydrocarbon zones. Swab test data and stratigraphic analysis will confirm machine learning model predictions. The first model successfully captures the stratigraphic patterns of the formation, which aligns with geological validation. The best-performing first machine learning model was able to classify three different clusters of facies, namely shale, sandstone, and shaly sands. The sequence of rock strata and stratigraphic correlations between wells were shown to be consistent with the conventional analysis, with only a minor deviation (0.5-2ft) in the well-to-well correlation sand markers. The second model accurately identifies potential zones based on the given features. The best-performing machine learning model was able to identify the initial condition of the reservoir interval with an accuracy of 84%. Moreover, it could identify a total of 346 potential zone intervals under the same condition. Moreover, as a result of the reservoir re-evaluation using the proposed workflow that has been carried out, a total of 35 candidates for primary perforation and 70 candidates for secondary perforation are proposed in the X-Field. These proposed candidates highlight the results of the workflow implemented to maximize production in the X-Field.