Modeling study of sandstone permeability under true triaxial stress based on backpropagation neural network, genetic programming, and multiple regression analysis

Permeability evolution of sandstone is of great significance in the development of tight sandstone gas reservoirs. Traditional laboratory tests have the disadvantages of high cost and long testing time. Therefore, the present study employed use artificial intelligence systems, i.e., backpropagation neural network (BPNN), genetic programming (GP), and multiple regression analysis to construct prediction models of sandstone permeability based on the coupling effect of true triaxial stress field and pore pressure. The results showed that the permeability prediction obtained from the systems fit well with the experimental data, and evidenced that permeability increased with pore pressure and decreased with increase in principal stress. Sensitivity analysis showed that the pore pressure has the greatest influence on sandstone permeability under different true triaxial stress. The effect of anisotropic principal stress on permeability exhibited sigma(1) > sigma(2) > sigma(3) under fixed pore pressure. Further assessment based on a combination of five evaluation indexes showed that the prediction accuracy of the BPNN model was better.

Automated summary

This study used BPNN, GP, and multiple regression analysis to construct prediction models for sandstone permeability, and found that pore pressure and principal stress have significant effects on sandstone permeability, with pore pressure showing the greatest influence. Comprehensive evaluation results indicated that the prediction accuracy of the BPNN model was superior.