Research On Geological Stratification Recognition Based On AdaBoost And GRU Method

The main task of logging interpretation is to use the logging data to analyze the lithology of the formation,to judge the oil,gas and water layers,to calculate the geological parameters such as porosity,saturation and permeability,and to divide the reservoir on this basis to make oil Mining provides guidance.The traditional interpretation of traditional logging data often relies on manpower and uses physical modeling to analyze,Which not only has a large Workload,but also has multiple solutions.With the development of computing technology,machine learning algorithms are gradually applied in the field of logging interpretation.By analyzing a large number of logging exploration data and constructing a suitable machine learning model,not only the prediction of geological parameters and the division of reservoirs can be realized,but also the workload of traditional manual marking formation can be greatly reduced.This paper studies the current status of petroleum logging interpretation and analyzes the existing problems.On this basis,based on the characteristics of petroleum logging data,the geological layered identification model of AdaBoost_GRU method is studied,and the automatic identification and column of the stratum is realized.The main research contents of this paper are as follows:(1)A data preprocessing method is proposed for logging data characteristics.Due to environmental,equipment or human factors,the logging data has a large amount of data missing and incomplete,which seriously affects the subsequent data interpretation.Moreover,the original logging data has a high dimension and there is a correlation between the data attributes.Based on this,this paper rejects the original invalid logging data,data completion and feature extraction,so as to maintain the data feature integrity and reduce the amount of data.(2)A three-dimensional Kriging space interpolation algorithm is studied.The traditional Kriging interpolation mainly predicts the geological properties of a certain location based on the geological properties of known points in the two-dimensional space.However,in the reconstruction of geological properties of drilling logs,the attribute values of a certain position in space are related to the attributes of surrounding locations,and data prediction is needed in three-dimensional space.According to the basic principle of Kriging,this paper improves the interpolation algorithm of two-dimensional Kriging and realizes the geological property of using the Kriging algorithm to predict the spatial position in three-dimensional space.(3)A geologic stratification recognition model based on AdaBoost and GRU methods is proposed.In this paper,the geological theory and GRU algorithm characteristics are combined to construct a geological layered recognition model,and the model weight coefficient formula and sample weight update formula are derived.By comparing the experimental results with BP,SVM,GRU and AdaBoost methods,the advantages of this model are proved.