Loading...

Utilizing Artificial Intelligence Technique in Acidizing Process of Asphaltenic Oil Wells

Sepideh Atrbar Mohammadi | 2023

227 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56015 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Ayatollahi, Shahaboddin; Pishvaie, Mahmoud Reza
  7. Abstract:
  8. The Oil wells are usually damaged because of the drilling process and production scenarios or fluid injection during EOR processes. These damages would critically affect the rate of production and injectivity of the well in the form of plugging damage. Different methods are used to fix these damages and increase the production flow from the oil wells. One of the most widely used well-stimulation methods to remediate this challenge is well acidizing. Although this method has very high efficiency in improving the ability of wells, if it is not designed and implemented correctly and optimally, it can cause induced damage and even lead to the well shutting. This challenge is especially reported in asphaltic oil reservoirs and it is one of the main challenges of the acidizing process, which should be given sufficient attention. If the reservoir oil and the injecting acid are not compatible, this can lead to the formation and precipitation of sludge in the wellbore area. Besides, this incompatibility would cause oil and acid emulsion formation, which critically reduces the efficiency of acidizing operations. Therefore, static tests are performed before the acid operation to check the possible incompatibility of oil and acid and to determine the appropriate inhibiting additives to solve these problems. These tests are costly and time-consuming which affects the whole acidizing process. Artificial intelligence techniques can be used to predict the results of these tests hence reducing the costs and time. Using these techniques would also decrease the risks associated with acid-based tests in the labs. This research is aimed to introduce machine learning methods predicting the outcomes of sludge and emulsion formation tests. These techniques would require collecting different test parameters and results related to the static tests to be used in machine learning tools to reach a model with the highest level of accuracy. It is important to note that the number of data, the number and type of inputs as well as the type of data set output are very vital for training machine learning models. This research work showed that the use of SMOTE method to balance the unbalanced data set and also increase the number of data in the data set would improve the performance of classification models. Checking the accuracy of the function by calculating Cohen's Kappa values (values of 0.59 and 0.46 as the highest and most suitable values for the test data, for the sludge and emulsion respectively) validates the model. These datasets include synthetic data prepared by SMOTE method as well as main inputs of viscosity, oil API and concentration of all additives. This study showed that the use of data sets with continuous numerical outputs compared to discrete and classified outputs leads to better performance in data prediction. This is concluded by comparing the value of 0.45 for the R2 of the test data and the negative values of Cohen's Kappa for the test data of the emulsion data set, respectively, with continuous numerical outputs and class outputs without synthetic data. According to the results, the use of accurate values for oil haracteristics such as viscosity, API, and composition (SARA fraction) is known to be very important and effective to build accurate machine-learning models
  9. Keywords:
  10. Reservoirs Acidizing ; Acid-Oil Emulsion ; Machine Learning ; Acidizing Treatment ; Acidizing Static Tests ; Sludge Precipitation ; Asphaltenic Oil

 Digital Object List

 Bookmark

No TOC