Well Logging Evaluation Of Carbonate Reservoir Using Machine Learning Methods

The properties of carbonate reservoirs are relatively complex,which are manifested as: deep buried reservoirs,complex storage spaces,and heterogeneity of reservoir physical properties.Existing mature evaluation techniques used in conventional sandstone reservoir have a poor performance when used in carbonate reservoirs.With the advent of the “geological big data” era,the potential of artificial intelligence(AI)in the field of oil and gas exploration is increasingly recognized.Compared with traditional methods that rely on expert experience and prior models,intelligent algorithms represented by machine learning(ML)have advantages such as data mining,complex problem interpretation,and automated prediction.This paper takes the application of machine learning in the logging evaluation of carbonate reservoirs as the main contents,and starts work on the basis of the data from the T6-5 well group in Tahe Oilfield,China.And a comprehensive assessment of various machine learning methods was carried out,which providing new ideas for intelligent reservoir evaluation.The main research results of the paper includes the following aspects:(1)Classification of carbonate reservoir types.Combining field outcrop observations,conventional logging response characteristics,FMI imaging logging images and production characteristics,the reservoirs in the target area are divided into two categories: karst cave and corroded pores/fracture.The karst caves could be divided into cave-filling with sand and mud,dissolution fracture,and unfilled cave.Finally,we summarized a set of specific methods for the classification of reservoirs for labeling.(2)Plotting of fracture density curves in carbonate reservoirs.Well S6-7 in the target area has a relatively high degree of fracture development.Combined with FMI imaging logging data,the numbers of fractures on single wells have been counted,and the fracture density curves of Well S6-7 have been drawn(depth,dip,azimuth).Through correlation analysis,it can be seen that the acoustic(AC)and density(DEN)logging parameters have poor correlation with fracture development.The degree of crack development is highly heterogeneous,and the development position is relatively concentrated.(3)The BP neural network method based on sample optimization predicts reservoir fracture density.Correlation analysis of fracture density and different logging parameters is carried out.Logging parameters with larger correlation coefficients are selected as input,and the BP neural network model in machine learning is used to predict reservoir fracture density.Taking into account the large difference among samples that affects the accuracy of the model results,the Kmeans clustering algorithm is introduced to optimize the distribution of the samples,and a more refined neural network model is established according to the clustering results,thereby improving the accuracy of fracture density prediction.(4)the types of carbonate reservoirs are predicted based on multiple ensemble learning strategies.Introducing the idea of integrated learning in machine learning,and based on the logging response patterns established by multiple data,we propose an improved method that combines the two integrated strategies of Boosting and Bagging to apply to the classification of carbonate reservoir types.The Ada Boost.M2 algorithm is used to construct 3 strong classifiers based on the support vector machine(SVM),decision tree(DT),and artificial neural network(ANN)in machine learning.Combined with the Bagging parallel strategy for combinatorial optimization,the final classification result of the reservoir type could be obtained.