Support Vector Machines Framework for Predicting the PVT Properties of Crude-Oil Systems

Abstract

PVT properties are very important in the reservoir engineering computations. There are many empirical approaches for predicting various PVT properties using regression models. Last decade, researchers utilized neural networks to develop more accurate PVT correlations. These achievements of neural networks open the door to both machine learning and data mining techniques to play a major role in both oil and gas industry. Unfortunately, the developed neural networks correlations have some limitations as they were originally developed for certain ranges of reservoir fluid characteristics and geographical area with similar fluid compositions. Accuracy of such correlations is often limited and global correlations are usually less accurate compared to local correlations. Recently, support vector machines have been proposed as a new intelligence framework for both prediction and classification based on both structure risk minimization criterion and soft margin hyperplane. This new framework dealt with kernel neuron functions instead of sigmoid-like ones, which allows projection to higher planes and solves more complex nonlinear problems. It has featured in a wide range of medical and business journals, often with promising results. The objective of this research is to assess the benefit of support vector machines as decision making tools in the field of oil and gas industry. To demonstrate the usefulness of the support vector machines technique in petroleum engineering area, we describe both the steps and the use of support vector machine modeling approach for predicting the PVT properties of crude oil systems. A comparative study will be carried out to compare their performance with the performance of the neural networks, nonlinear regression, and the empirical correlations algorithms. A preliminary results show that the performance of support vector machines will be accurate, reliable, and outperform most of the existing approaches. Future work can be achieved by using this new framework as a modeling approach for solving oil and gas industry problems, such as, permeability and porosity prediction, identify liquid-holdup flow regimes, and other reservoir characterization. Introduction Reservoir fluid properties are very important in petroleum engineering computations, such as, material balance calculations, well test analysis, reserve estimates, inflow performance calculations, and numerical reservoir simulations. Ideally, these properties are determined from laboratory studies on samples collected from the bottom of the wellbore or at the surface. Such experimental data are, however, very costly to obtain. Therefore, the solution is to use the empirically derived correlations to predict PVT properties, Osman et al.38. There are many empirical correlations for predicting PVT properties, most of them were developed using equations of state (EOS) or linear/non-linear multiple regression or graphical techniques or feedforward neural networks (ANN or FFN). However, they often do not perform very accurate and suffer from a number of drawbacks. Each correlation was developed for a certain range of reservoir fluid characteristics and geographical area with similar fluid compositions and API oil gravity. Thus, the accuracy of such correlations is critical and not often known in advance. Among those PVT properties is the bubble point pressure (bpp), Oil Formation Volume Factor (Bob), which is defined as the volume of reservoir oil that would be occupied by one stock tank barrel oil plus any dissolved gas at the bubble point pressure and reservoir temperature. Precise prediction of Bob is very important in reservoir and production computations. The objective of this study is to develop a new support vector machines prediction model for both bpp and Bob based on the kernel function scheme using worldwide experimental PVT data.

Machine learning on field data for hydraulic fracturing design optimization

“…Numerous efforts have been made by researches to implement data science to lab cost reduction issues. PVT [23] correlations correction for crude oil systems were comparatively studied between ANN and support vector machine (SVM) algorithms. Then, the problem is formulated as follows: one may suppose that a typical hydraulically-fractured well does not reach its full potential, because the fracturing design is not optimum.…”

Hybrid functional networks for oil reservoir PVT characterisation

“…The "black box" representation of ANN has made it unattractive for adoption in industrial PVT application. Sequel to this, some representative ML and evolutionary techniques have been used, such as support vector machine (SVM), genetic algorithm (GA), adaptive neuro fuzzy system (ANFIS), functional networks (FN) and so on (Khoukhi et al, 2011;El-Sebakhy et al, 2007;Hajizadeh, 2007).…”

“…For more reliable and improved prediction performance of these PVT properties, other researchers have implemented machine learning [ML] algorithms to predict different PVT properties (Osman & Al-Marhoun, 2005;Khoukhi et al, 2011;Gharbi et al, 1999;El-Sebakhy et al, 2007). Also, to address the problem in modelling ANN, a recursive least squared algorithm has been used for learning feedfoward ANN to model crude oil blending process (de Jesús Rubio, 2016).…”

Moving boundary approach to forecast tight oil production

“…Surrogate‐based models 21,22 such as polynomial response surface 23‐26 and artificial intelligence method such as least square support vector machine 27‐29 and artificial neuron network are gaining interest due to high computational efficiency. In surrogate models, important parameters 30 are selected based on their hierarchy to affect the production.…”