Modeling viscosity of crude oil using k-nearest neighbor algorithm
Corresponding Author and Email:Abdolhossein Hemmati-Sarapardeh, hemmati@uk.a.c.ir & aut.hemmati@gmail.com; ORCID: https://orcid.org/0000-0002-5889-150X
Citation:Mahdiani, M.R., Khamehchi, E., Hajirezaie, S., Hemmati-Sarapardeh, A. Modeling viscosity of crude oil using k-nearest neighbor algorithm. Advances in Geo-Energy Research, 2020, 4(4): 435-447, doi: 10.46690/ager.2020.04.08.
Article Type:Original article
Abstract:
Oil viscosity is an important factor in every project of the petroleum industry. These processes can range from gas injection to oil reservoirs to comprehensive reservoir simulation studies. Different experimental approaches have been proposed for measuring oil viscosity. However, these methods are often time taking, cumbersome and at some physical conditions, impossible. Therefore, development of predictive models for estimating this parameter is crucial. In this study, three new machine learning based models are developed to estimate the oil viscosity. These approaches are genetic programing, k-nearest neighbor (KNN) and linear discriminant analysis. Oil gravity and temperature were the input parameters of the models. Various graphical and statistical error analyses were used to measure the performance of the developed models. Also, comparison study between the developed models and the well-known previously published models was conducted. Moreover, trend analysis was performed to compare the predictions of the models with the trend of experimental data. The results indicated that the developed models outperform all of the previously published models by showing negligible prediction errors. Among the developed models, the KNN model has the highest accuracy by showing an overall mean absolute error of 8.54%. The results show that the new developed models in this study can be potentially utilized in reservoir simulation packages of the petroleum industry.
Keywords:Oil viscosity, machine learning, k-nearest neighbor, genetic programming, linear discriminant analysis.
Download this Manuscript:PDF
