Potential Hydrocarbon Zone Classification Using Machine Learning, S1 Oil Field

Abstract The S1 Oil Field, recognized as a brown field production site in Thailand, faces the significant challenges in maintaining production levels and maximizing reserves in its late stage of development. One effective technique to enhance production involves adding perforation zones in the existing wells by targeting the remaining potential hydrocarbon zones, also known as Additional Zones of Interest (AZI). However, a critical challenge arises when these AZI contain water, as their addition could inadvertently increase water production, thereby complicating the production management plan. This study aims to classify Potential AZI and Non-potential AZI using a combination of petrophysical data and advanced machine learning techniques. The project workflow involves using perforated AZI data along with a geological parameter dataset, and labeling the type of AZI based on production tracking to train a supervised model. After training, the model is used to classify un-perforated AZI data, enabling the formulation of perforation plans for Potential AZI and the cancellation of perforation plans for Non-potential AZI. the dataset comprises 8 key features: volume of shale, porosity, water saturation, thickness, pressure, mobility, permeability are derived from a comprehensive petrophysical database and trap type. A total of 564 perforated AZI samples from 5 areas covering the production zone of the S1 asset. These samples were categorized as either "Potential AZI" (indicating oil production) or "Non-potential AZI" (indicating water production) based on production tracking data. The classification was determined by observing the gross increase in production after adding perforation zones. If there was a net increase in oil production, the AZI was classified as Potential AZI. Conversely, if there was an increase in water production, the AZI was classified as Non-potential AZI. The next steps in the study involved cleaning the dataset by removing outliers that did not align with geological concept or guideline and filling in any missing data. The Random Forest (RF) model was selected for the classification task due to its accuracy. Hyperparameter tuning was employed to identify the optimal model parameters that yielded the highest accuracy. The final step involved evaluating the model using the F1 score, which is calculated based on the precision and accuracy of the model's predictions. The model achieved an impressive F1 score of 84%, demonstrating its high accuracy in classifying Additional Zones of Interest (AZI). To ensure the reliability of these results, they were validated through detailed geological correlations and meticulous production tracking. The Potential AZI zones showed a strong consistency with the hydrocarbon column height, as determined through well log correlations. On the other hand, the Non-potential AZI zones were characterized by low resistivity and low gas show responses, indicating less favorable conditions for hydrocarbon presence. Additionally, production tracking from wells that had undergone perforation AZI activities showed an increase in gross water volume, which matched the model's predictions for Non-potential AZI. This thorough validation process confirms the model's effectiveness in accurately identifying and classifying AZI, providing valuable insights for optimizing production strategies. The approach taken in this study incorporates digitalization to enhance the screening of perforation zones in existing wells, thereby optimizing production strategies during the brown field phase of the S1 Oil Field.