Natural gas which consists mainly of methane (usually more than 90% in volume), is becoming increasingly an important and efficient source of energy because of the lower greenhouse gas emissions and air pollution. Achieving satisfactory recovery factors in gas reservoirs is sensitive to the methane-brine/water interfaces induced by the interfacial tension (IFT) between these systems. Accordingly, accurate determination of IFT of the systems methane-brine/water is extremely important for natural gas production. In this paper, several intelligent models were implemented to accurately estimate interfacial tension (IFT) of the systems brine/pure water-methane under wide temperature, pressure and salinity ranges of (278.1–477.59 K), (0.01–260 MPa) and (0–200,000 ppm), respectively. The established models were based on an extensive databank including 879 experimental measurements. The implemented intelligent models in this study were Extreme Learning Machine (ELM), Radial Basis Function (RBF) neural network, Multilayer Perceptron (MLP), Least Square Vector Machine (LSSVM), and Generalized Regression Neural Network (GRNN). Various optimization algorithms were applied for improving the learning phase of these models. Furthermore, a Committee Machine Intelligent System (CMIS) scheme was proposed by linking the best-found paradigm under a linear single model. The results showed that all the developed intelligent-based paradigms exhibit reliable prediction abilities. In addition, it was found that CMIS and GRNN are the fittest paradigms with overall absolute average percent relative error (AAPRE) values of 1.117% and 1.003%, respectively. Besides, the performance assessment revealed that our best paradigms outperform the existing approaches. Finally, the sensitivity analysis revealed that salinity has a slight impact on IFT.

由于温室气体排放量和空气污染降低，主要由甲烷（通常占体积的90％以上）组成的天然气正变得越来越重要和高效。在气藏中获得令人满意的采收率对这些系统之间的界面张力（IFT）引起的甲烷-盐水/水界面很敏感。因此，准确确定甲烷-盐水/水系统的IFT对天然气生产极为重要。本文采用了多种智能模型来准确估算宽温度，压力和盐度范围（278.1–477.59 K），（0.01–260 MPa）和（ 0–200,000 ppm）。建立的模型基于包括879个实验测量值的广泛数据库。本研究中已实现的智能模型是极限学习机（ELM），径向基函数（RBF）神经网络，多层感知器（MLP），最小二乘向量机（LSSVM）和广义回归神经网络（GRNN）。应用了各种优化算法来改善这些模型的学习阶段。此外，通过链接线性单一模型下的最佳范例，提出了委员会机器智能系统（CMIS）方案。结果表明，所有已开发的基于智能的范例都具有可靠的预测能力。此外，还发现CMIS和GRNN是最合适的范例，总体绝对平均相对百分比误差（AAPRE）值分别为1.117％和1.003％，分别。此外，性能评估表明，我们最好的范例优于现有方法。最后，敏感性分析表明盐度对IFT有轻微影响。