Dolatyari Alireza, Ahmady Mohammad, Kazemi Alireza
Department of Petroleum Engineering, Amirkabir University of Technology, Tehran, Iran.
Department of Petroleum and Chemical Engineering, Sultan-Qaboos University, Muscat, Oman.
Sci Rep. 2024 Oct 1;14(1):22836. doi: 10.1038/s41598-024-73573-8.
Viscosity is crucial in subsurface and surface transport, used in engineering domains like heat transfer and pipeline design. However, measurements are limited, necessitating predictive viscosity relationships. Existing models lack precision or pertain to limited fluids, and accurately forecasting dead oil viscosity remains challenging due to errors. The study presents a mathematical algorithm to accurately estimate viscosity values in hydrocarbon fluids. It uses a robust non-linear regression technique to establish a reliable relationship between fluid viscosity and temperature within a specific temperature range. The algorithm is applied to extra-heavy to light crude oil samples from Iranian oilfields, revealing viscosity values ranging from 0.29 cp to 5328.74 cp within a dataset of 243 viscosity data points. After modeling each of these five fluids, the highest values obtained for the maximum absolute error and relative error are related to the fluid with an API gravity of 12.92. The maximum absolute error and relative error for this fluid sample are 1.25 cp and 6.04%, respectively. The algorithm offers acceptable precision in outcome models, even with limited training data, demonstrating its effectiveness in training models with less than 30% of available data. Moreover, these models end up with a near-unity coefficient of determination in testing data, reaffirming their proficiency at reflecting empirical data with remarkable accuracy.
粘度在地下和地表输运中至关重要，应用于传热和管道设计等工程领域。然而，测量数据有限，因此需要预测粘度关系。现有模型缺乏精度或仅适用于有限的流体，由于误差，准确预测死油粘度仍然具有挑战性。该研究提出了一种数学算法，用于准确估计烃类流体中的粘度值。它使用强大的非线性回归技术，在特定温度范围内建立流体粘度与温度之间的可靠关系。该算法应用于伊朗油田的超重质到轻质原油样品，在243个粘度数据点的数据集中，粘度值范围为0.29厘泊至5328.74厘泊。对这五种流体中的每一种进行建模后，获得的最大绝对误差和相对误差的最高值与美国石油学会（API）重度为12.92的流体有关。该流体样品的最大绝对误差和相对误差分别为1.25厘泊和6.04%。即使训练数据有限，该算法在结果模型中也具有可接受的精度，证明了其在使用不到30%的可用数据训练模型方面的有效性。此外，这些模型在测试数据中的决定系数接近1，再次证实了它们以极高的准确性反映经验数据的能力。
