Prediction of crude oil refractive index through optimized support vector regression: a competition between optimization techniques

J Petrol Explor Prod Technol (2017) 7:195–204 DOI 10.1007/s13202-016-0229-7 ORIGINAL PAPER - PRODUCTION ENGINEERING Prediction of crude oil refractive index through optimized support vector regression: a competition between optimization techniques Amin Gholami1 • Hamid Reza Ansari2 • Seyedsajad Hosseini3 Received: 13 September 2015 / Accepted: 17 January 2016 / Published online: 1 February 2016 Ó The Author(s) 2016. This article is published with open access at Springerlink.com Abstract Refractive index (RI) provides valuable information about various reservoir engineering calculations, making it a key parameter for characterizing crude oils. Determination of this index through experiment is capitalintensive, time consuming, and also toil. Hence, it is essential to search for an efficient and accurate estimation of crude oil RI. In this study, an intelligent approach, based on optimized support vector regression (SVR), is introduced to find a quantitative correlation between crude oil RI and SARA (saturate, aromatic, resin, and asphaltene) fraction data. Optimization of SVR is implemented through three searching approaches, viz. hybrid of grid and pattern search (HGP), genetic algorithm (GA), and imperialist competitive algorithm (ICA). Using these approaches, three models are constructed and tested on experimental data gathered from open source literature. To evaluate the performance of these models, their outputs are compared with corresponding experimental data in terms of statistical criteria. The comparative study clearly shows the advantage of ICA over its rivals (GA and HGP) in optimizing the SVR parameters. ICA optimized support vector regression results in an R2 of 0.9971 and MSE of 1.48548e-05 demonstrating its efficacy in obtaining crude oil refractive index form SARA data. & Amin Gholami ; 1 Reservoir Engineering Division, Iranian Offshore Oil Company, Tehran, Iran 2 Abadan Faculty of Petroleum Engineering, Petroleum University of Technology, Abadan, Iran 3 Department of Petroleum Engineering, Sharif University of Technology, Tehran, Iran Keywords Refractive index (RI)  Support vector regression (SVR)  Hybrid grid and pattern search (HGP)  Genetic algorithm (GA)  Imperialist competitive algorithm (ICA) Introduction Refractive index (RI) is one of foremost representative properties of crude oils. This optical parameter is a fundamental physical property pertaining to the speed of light in crude oil (Chamkalani et al. 2014). Recently, this parameter has achieved growing attention in petroleum industry, mainly because of its application in various calculations regarding crude oil specific compositions (El Ghandoor et al. 2003). Touba et al. (1997) investigated the applicability of crude oil RI for defining various reservoir fluid characterizing properties. Their results demonstrated that crude oil RI can be used as a suitable property for predicting other crude oil properties including those controlling PVT behavior of hydrocarbon and surface tension. Taylor et al. demonstrated that the onset of asphaltene precipitation could be detected from RI measurement (Taylor et al. 2001). Fan et al. (2002) examined the relation between crude oil RI and asphaltene stability in crude oil. They suggested that the difference between crude oil RI and crude oil RI at the onset of asphaltene precipitation could be employed as a decisive factor for the assessment of asphaltene stability in crude oils (Fan et al. 2002). Buckley and Wang (2002) showed that the wettability altering tendency is related to crude oil RI. Riazi and AlOtaibi (2001) attempted to shed a new light on the relation between the viscosity and RI. They proposed a simple mathematical-based approach which can estimate the viscosity of crude oils from their RI data. 123 196 J Petrol Explor Prod Technol (2017) 7:195–204 Considering the importance of crude oil RI, introduction of a reliable technique for accurate determination of this parameter is valuable. Conventional refractometers have long been employed as the leading method of direct measurement for determination of crude oil RI. However, the conventional refractometer is incapable of measuring the RI of heavy crude oils, mainly due to its opacity (Taylor et al. 2001). Recently, a novel RI measurement technique, the so-called capillary tube interferometer, has been introduced by El Ghandoor et al. (2003) for laboratory determination of heavy crude oil RI. This method can determine the RI of heavy crude oil, as well as light crude oil, with high precision. Although aforementioned experimental technique measures the RI of crude oil with high accuracy, utilizing this method for quantifying crude oil RI is associated with some difficulties as experimental implementation is costly, time consuming, and also labor intensive. Due to drawbacks encountered in various methods of experimental measurement, development of a crude oil RI model could facilitate exact calculation of crude oil RI from readily available experimental data. Several researchers have tried to develop a predictive model for determining the RI (Chamkalani et al. 2014; Fan et al. 2002; Buckley and Wang 2002; Chamkalani et al. 2012; Chamkalani 2012; Gholami et al. 2014a, b; Vargas and Chapman 2010; Zargar et al. 2015; Tatar et al. 2015). The published models available in the literature for estimating the value of crude oil RI can be divided into two main categories. In the first group of models, RI of crude oils is estimated using their density. Buckley and Wang (2002) developed a correlation relating crude oil RI to its density through linear regression. Chamkalani (2012) also presented a similar correlation to that of Buckley and Wang (2002). Vargas and Chapman (2010) evaluated the applicability of the One-Third rule in hydrocarbon and crude oil systems. They introduced a novel model, so-called Lorentz–Lorenz expansion, for building a correlation between RI and density of crude oils. In the second group of models, the RI of crude oil is calculated using SARA (saturate, aromatic, resin, and asphaltene) fraction data. The concept of the mutual dependence of crude oil RI and SARA fraction data was introduced by Fan et al. (2002). They developed an empirical correlation between the SARA fraction data and RI of crude oil. Chamkalani (2012) established another empirical correlation to estimate the value of crude oils RI. Although aforementioned empirical correlations are valuable, they are faced with some shortcomings, the main one being unsatisfactory accuracy. Recently, intelligence based models have emerged as powerful techniques of solving complex problems encountered in petroleum industry. These models have achieved encouraging results in modeling phenomena and parameters such as asphaltene precipitation, minimum 123 miscible pressure, viscosity, porosity, permeability, bubble point pressure (Asoodeh et al. 2014a, b; Bagheripour et al. 2015; Gholami et al. 2014c; Rasuli Nokand (...truncated)