Permeability prediction for carbonate reservoir using a data-driven model comprising deep learning network, particle swarm optimization, and support vector regression: a case study of the LULA oilfield

Abstract

Listen

Translate article

Permeability is universally considered as an important parameter since its data is critical for some basic geological work, such as constructing a pore-throat system of reservoir, evaluating flowing capability of formation, etc. Accordingly, how to predict permeability becomes a primary research in the realm of geoscience. Lots of physical models specifically used for permeability prediction have been created in recent decades, but calculation parameters involved in the models must be determined through core experiments. In order to reduce reliance on the usage of the data generated by core experiments, statistical prediction methods for permeability are developed. Support vector regression (SVR) is one of the optimal prediction approaches that can perfectly reveal the nonlinear relationship between permeability and other geological parameters, while its performance is severely limited by qualities of input data and calculation parameters. Colinear issue regarding variables of input data and bad initialization of calculation parameters can cause failure of fitting equation establishment. Thus, in view of those demerits, two techniques, deep learning and particle swarm optimization (PSO), are introduced to enhance the calculation capability of SVR. Then, a hybrid data-driven model which consists of deep learning network, PSO, and SVR is proposed. Data for method validation is recorded by three wells of the LULA oilfield. Two experiments are designed by the validation data. Experiment results prove that the proposed method has the capability to produce more accurate predicted results than those provided by single SVR and PSO-SVR. Consequently, the new hybrid data-driven model is effective in predicting permeability under processing real data.