Maghsoudian Ali, Izadpanahi Amin, Bahmani Zahra, Avvali Amir Hossein, Esfandiarian Ali
Department of Petroleum Engineering, Ahvaz Faculty of Petroleum, Petroleum University of Technology, Ahvaz, Iran.
Escola Politécnica, Universidade de São Paulo, Sao Paulo, Brazil.
Sci Rep. 2025 Jan 2;15(1):537. doi: 10.1038/s41598-024-84402-3.
Smart water injection (SWI) is a practical enhanced oil recovery (EOR) technique that improves displacement efficiency on micro and macro scales by different physiochemical mechanisms. However, the development of a reliable smart tool to predict oil recovery factors is necessary to reduce the challenges related to experimental procedures. These challenges include the cost and complexity of experimental equipment and time-consuming experimental methods for obtaining the recovery factor (RF). In this paper, three predictive algorithms including adaptive neuro-fuzzy inference system (ANFIS), artificial neural network (ANN), and multigene genetic programming (MGGP) are developed to predict the RF of smart water flooding in carbonate reservoirs. Accordingly, 205 data points from coreflooding tests and 122 from Amott-cell tests were collected from previous studies. Porosity, permeability, oil viscosity, and oil density at reservoir temperature, injection rate, total dissolved solids (TDS), temperature, injection time, and initial water saturation (S) were selected as the input parameters. Results show the great performance of ANN, compared to other employed algorithms. Coefficients of determination (R) of ANN obtained from Amott-cell data for training, testing, validation, and overall data are 0.9748, 0.9021, 0.9765, and 0.9646, respectively. The corresponding values from coreflooding data are 0.9502, 0.9582, 0.9837, and 0.9523, respectively. Moreover, parametric sensitivity analysis was performed for the input parameters. Based on this analysis, time and injection rate have the most positive impact on the Amott-cell and coreflooding, respectively. Sensitivity analysis from Amott-cell data introduces TDS and oil viscosity have the most negative effects on RF performance. Furthermore, the most negative effects belong to porosity and permeability for coreflooding experiments.
智能注水（SWI）是一种实用的提高采收率（EOR）技术，它通过不同的物理化学机制在微观和宏观尺度上提高驱替效率。然而，开发一种可靠的智能工具来预测采收率因子对于减少与实验程序相关的挑战是必要的。这些挑战包括实验设备的成本和复杂性以及获取采收率（RF）的耗时实验方法。本文开发了三种预测算法，包括自适应神经模糊推理系统（ANFIS）、人工神经网络（ANN）和多基因遗传编程（MGGP），以预测碳酸盐岩油藏智能水驱的采收率。相应地，从先前的研究中收集了来自岩心驱替试验的205个数据点和来自阿莫特岩心试验的122个数据点。选择孔隙度、渗透率、油藏温度下的油粘度和油密度、注入速率、总溶解固体（TDS）、温度、注入时间和初始含水饱和度（S）作为输入参数。结果表明，与其他采用的算法相比，ANN具有优异的性能。从阿莫特岩心数据获得的用于训练、测试、验证和整体数据的ANN决定系数（R）分别为0.9748、0.9021、0.9765和0.9646。来自岩心驱替数据的相应值分别为0.9502、0.9582、0.9837和0.9523。此外，对输入参数进行了参数敏感性分析。基于该分析，时间和注入速率分别对阿莫特岩心和岩心驱替具有最积极的影响。来自阿莫特岩心数据的敏感性分析表明，TDS和油粘度对采收率性能具有最消极的影响。此外，对于岩心驱替实验，最消极的影响属于孔隙度和渗透率。
