A hybrid PSO-SVM-based model for determination of oil recovery factor in the low-permeability reservoir

Oil recovery factor is one of the most important parameters in the development process of oil reservoir, especially in the low-permeability reservoir. In general, the determination of recovery factor can be obtained either experimentally or numerically. Experimental method is often time-consuming and expensive, while numerical method has been always confined to narrow range of application or relatively large error. Recently, an intelligent method has been proven as an efficient tool to model the complex and nonlinear phenomena. In this work, an intelligent model based on support vector machine in combination with the particle swarm optimization (PSO-SVM) technique was established to predict oil recovery factor in the low-permeability reservoir. Input variables of the proposed PSO-SVM model with the aid of a grey correlation analysis method are permeability, well spacing density, production-injection well ratio, porosity, effective thickness, crude oil viscosity and output parameter is oil recovery factor of low-permeability reservoir. The accuracy and reliability of the proposed model were evaluated through 34 data sets collected in the open literature and compared with PSO-BP neural network, empirical method from Oil and Gas Company. The results indicated that the PSO-SVM model gives the best results with average absolute relative deviation (AARD) of 3.79%, while AARDs for the PSO-BP neural network and empirical method are 9.18% and 10.0%, respectively. Furthermore, outlier detection was used on the basis of whole data sets to definite the valid domains of PSO-SVM and PSO-BP models by detecting the probable doubtful recovery factor data in the low-permeability reservoir.