Insights to Surfactant Huff-Puff Design in Carbonate Reservoirs based on Machine Learning Modeling

Abstract

Surfactants could react with adsorbates on carbonate rock surface to alter wettability from oil-wetness to water-wetness, which is effective to enhance oil recovery. Surfactant huff-puff treatment is mostly applied for this purpose and the resulting surfactant performance is the outcome of complex interfacial processes. Currently, the effect of important parameters on surfactant performance is not completely reported and the contribution of each parameter to surfactant performance is hard to be quantified. Traditional methods to optimize surfactant performance are time-consuming and show strong dependency on extensive experiments. In this paper, we address these problems from machine learning (ML) perspectives. Several ML models are established to predict surfactant performance and Random Forest (RF) model presents better accuracy. Based on RF model, we apply Shapley additive explanations (SHAP) approach to interpret modeling results to obtain new insights and provide solutions to unsolved problems. Results show that when interfacial tension is lower than a critical value or oil API gravity is higher than a critical value, surfactant performance shows obvious improvements. In general, porosity, permeability, and surfactant concentration are positively associated with surfactant performance. The trend becomes less obvious when parameter value exceeds a certain level. In addition, SHAP value could effectively decompose surfactant performance into individual effect of each parameter. This analysis is valuable to indicate certain parameters to be optimized. It is shown that surfactant concentration in a proportion of samples is kept in a low level and complete wettability alteration is not achieved. To optimize surfactant concentration, an innovative procedure integrating RF prediction model with Powell's method is proposed. This procedure is effective to avoid deficient and superabundant surfactant concentration. With optimization, the average surfactant concentration is increased from 0.37 wt% to 0.93 wt%. The average probability of high-oil-recovery class is improved from 0.38 to 0.50. The incremental oil recovery is improved from low-oil-recovery class to high-oil-recovery class. Our work promotes the understanding of wettability alteration by surfactants and provides a fast framework to predict, analyze, and optimize surfactant performance. This framework could greatly save experiment time and cost.

Department(s)

Geosciences and Geological and Petroleum Engineering

Comments

This work was supported in part by the National Science Foundation under Grant No. OAC-1919789.

Keywords and Phrases

Carbonate Reservoirs; Machine Learning; Spontaneous Imbibition Test; Surfactant Huff-Puff Design; Wettability Alteration

International Standard Serial Number (ISSN)

1385-8947

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2022 Elsevier, All rights reserved.

Publication Date

01 Jan 2023

Link to Full Text

Share

 
COinS