Pay Zone Determination Using Enhanced Workflow and Neural Network

Abstract

Amplitude versus offset (AVO) analysis and attributes are frequently utilized during the early stages of exploration when no well has been drilled. However, there are still some drawbacks to this method, including the fact that it involves a substantial amount of time and experience, as well as the subjectivity of manual analysis. By utilizing unsupervised learning, this process can be done more objectively and faster. Unsupervised learning can detect anomalies and identify patterns to understand more about the datasets since, at this early stage of exploration, there is still a lack of information and labelling. A type of unsupervised learning referred to as self-organizing maps (SOM) is applied in this study to delineate hydrocarbons from given AVO properties that were used to detect hydrocarbons. SOM is also used to eliminate redundancy in the selection of attributes prior to the delineation procedure. The investigation began with well log data and progressed ahead into multiple fluid conditions to evaluate the model’s ability to identify hydrocarbons. The analysis can then be extended to the seismic dataset. By combining SOM, correlation coefficient, and mean–median, a method is devised for filtering features to remove redundancy. On the hydrocarbon delineation process, the model managed to detect hydrocarbons using well log simulations and was confirmed using water saturation logs. Additionally, the model is validated using real seismic data, demonstrating a promising performance in defining probable hydrocarbons. The proposed method enables early detection of hydrocarbon content during the preliminary stage of exploration when no well is accessible.

Keywords

• hydrocarbon delineation
• unsupervised learning
• self-organizing maps