Study On Machine Learning Prediction And Conditional Random Simulation Of Formation Physical Property Parameters

It is one of the key tasks to obtain accurate physical porosity distribution in the process of oil and gas exploration and development.If the formation heterogeneity is strong and the relationship between logging curve and reservoir characteristics is highly non-linear,the traditional prediction method of physical parameters based on single variable and linear assumption can not meet the production demand.Although the results obtained from formation nuclear magnetic resonance experiments are of high accuracy,the cost is also high.Therefore,it is necessary to study the prediction methods of porosity.Machine learning algorithm has become one of the research hotspots in the current data age.Because of its excellent performance in dealing with non-linear problems,it has attracted wide attention in the field of stratum physical parameters.In this paper,seven logging curves,including acoustic time difference(AC),natural gamma ray(GR),deep investigate double lateral resistivity log(RD),shallow investigate double lateral resistivity log(RS),total organic carbon(TOC),density(DEN),compensated neutron(CN),are selected.Combining with geophysical inversion technology,Random Forest regression and support vector regression models are established to predict physical parameters-porosity of four wells in the study area.Programming implementation with R language,the prediction results show that the average prediction error of support vector regression is 17%,while random forest regression is 15% in small samples and less than10% in large samples.Random forest regression is better than support vector regression.While using the random forest regression model to predict the porosity,giving the order of the importance of a variable.The results show that acoustic time difference(AC),natural gamma ray(GR),density(DEN)and total organic carbon(TOC)have greater influence on porosity,while shallow bilateral resisitivity loggging(RS)has the least influence.The results show that the variation model has the best simulation effect when it is used as an exponential function,and has the best consistency with the originalstochastic forest porosity prediction value,so as to maximize the information of porosity distribution.The results show that the random Forest regression model has higher prediction accuracy,better stability,stronger extrapolation ability and higher efficiency than the support vector regression model.It provides an economic and efficient new method for predicting porosity of physical parameters in a certain range and condition,and has certain reference value in practical engineering application.