Utilizing Machine Learning Classification Models for Acid-Oil Emulsion Prediction in Laboratory Acidizing Static Tests by Using a Hybrid Databank

Message:
Article Type:
Research/Original Article (دارای رتبه معتبر)
Abstract:

During the lifetime of an oil well, the near wellbore areas are usually exposed to formation damage due to factors such as fines migration, clay swelling, etc., significantly reducing the oil well›s productivity and injectivity rates. One of the widely used well-stimulation methods to remove formation damage is acidizing in which the acid and chemicals (additives) are injected into the formation to increase the permeability of the formation by dissolving carbonate rocks. However, the lack of laboratory examination of the compatibility of injection fluids with formation fluids at the design stage results in induced damage such as acid emulsion in oil in formation. The conduction of laboratory tests in order to execute compatibility between fluids is time-consuming, expensive, and has issues related to safety. This research aims to predict the primary results of anti-emulsion tests using data-driven models in a short time. For this purpose, the most influential data on the results of these tests, including type and concentration of acid, and additives like anti-emulsion, anti-sludge, surface tension reducer, and iron ion reducer, as well as properties of 13 different types of oil from various reservoirs, such as viscosity, density, and ferric ion concentration, were collected and recorded as inputs to a data set. Then, some supervised classification models including random forest, support vector machine, multi-layer perceptron, and extreme gradient boosting algorithms have been implemented to predict the output of anti-emulsion tests. Additionally, the statistical technique SMOTE (Synthetic Minority Oversampling Technique) was employed to generate artificial data samples and enhance AI models’ performance. Ultimately, results indicate that the extreme gradient boosting with five estimators achieved the best performance with Cohen›s kappa values of 0.79 and 0.523 for training and testing datasets, respectively.

Language:
Persian
Published:
Petroleum Research, Volume:34 Issue: 139, 2025
Pages:
73 to 91
https://www.magiran.com/p2847856  
سامانه نویسندگان
  • Hossein Kheirollahi
    Author (2)
    MSc Graduated Petroleum and Natural Gas, Sahand University Of Technology, Tabriz, Iran
    Kheirollahi، Hossein
اطلاعات نویسنده(گان) توسط ایشان ثبت و تکمیل شده‌است. برای مشاهده مشخصات و فهرست همه مطالب، صفحه رزومه را ببینید.
مقالات دیگری از این نویسنده (گان)