Research And Application Of Lithology Identification Based On Ensembling Learning

In oil and gas exploration,lithology identification is an important basis for reservoir division,the premise for accurate determination of porosity and oil saturation,the basis for reservoir characteristics research,reserve calculation and geological modeling,and the powerful evidence for subsequent determination of lithology spatial distribution targets and complex THREE-DIMENSIONAL modeling.There are various lithology identification methods,among which the identification of lithology by logging curve is one of the most commonly used methods.Due to its limitations,the traditional logging lithology identification method has some problems,such as poor accuracy,low identification efficiency and human factors,which can not meet the actual production demand.In recent years,the rapid development of machine learning technology has promoted the process of lithology identification automation.The goal of machine learning is to train a model with strong generalization ability and high adaptability in sample space.However,this process has great difficulty,and sometimes the trained model cannot accurately identify and predict the lithology,so it is difficult to ensure high accuracy and high stability at the same time.As an improvement method,ensemble learning is different from traditional machine learning methods,mainly by training multiple models and combining them.High accuracy and stability.The main research contents are as follows:First,an improved Stacking ensemble learning algorithm is proposed.Aiming at the problem that the actual prediction situation is not fully considered when the traditional Stacking integrated learning algorithm trains the base model,which leads to the low prediction accuracy of the final model,a new algorithm is designed that can calculate the weight of its base model according to the PCA algorithm,and add the weight to the overall model training process.Through experiments,it is found that this method can provide more sufficient prediction information for the meta-model,thereby effectively improving the accuracy of lithology identification.Second,an improved hybrid expert model based on KMeans task division is proposed.Aiming at the problems of strong correlation and insufficient diversity among base models in traditional ensemble learning methods.An improved hybrid expert model was designed.The model divided data into different task scopes by KMeans clustering function,and trained each sub-task by different base models,which consciously enhanced the diversity of base models in the ensemble learning method.Experiments show that the algorithm is helpful to improve the accuracy of lithology identification.Thirdly,the development of lithology identification system based on integrated learning is completed.The system is developed based on CIFLog platform,showing the whole process of lithology identification using integrated learning,from data preprocessing,model training to application,and realizing many functions including logging curve drawing,logging curve data preprocessing,intelligent model training,intelligent model application and identification result visualization.The algorithmatic models in the system offer a variety of models including improved Stacking integrated learning,improved hybrid expert models,and a variety of single machine learning for users to compare and choose from.It has been tested and used by the CIFLog platform,and the overall effect is good,and it has considerable use value and promotion value.