Implementation of regression algorithms for oil recovery prediction

This paper presents the work of predicting oil production using machine learning methods. As a method, multiple linear regression algorithm with polynomial properties was implemented. Regression algorithms are suitable and workable methods for based on data-driven approach. The synthetic dataset obtained Buckley-Leverett mathematical model, which is used to calculate hydrodynamics determine saturation distribution in problems. Various combinations parameters problem were chosen, where porosity, viscosity phase absolute permeability rock taken as input learning. And value recovery factor chosen output parameter. More than 400 thousand data test algorithms. To estimate quality algorithms, mean square error metrics coefficient determination used. It found that does not cover all patterns due underfitting. degrees deployed tested, it also our data, quadratic model trains quite well perfectly predicts factor. solve overfitting problem, L1 regularization known Lasso method applied. For 0.96, pretty good result data. Thus, assumed discussed can be useful practical from fields at stages