Formation resistivity inversion and uncertainty analysis based on hybrid machine learning model
In the process of extended reach wells and horizontal wells,it is of great significance for geosteer-ing work to quickly and accurately extract stratigraphic information from the logging data of LWD tools by es-tablishing an effective inversion model.Machine learning has emerged as a widely adopted approach to ad-dress geophysical inversion problems.However,evaluating the reliability and uncertainty of inversion results proves challenging due to the predominant focus on deterministic methods.Hence,assessing the reliability of inversion outcomes becomes critical,and this can be accomplished through uncertainty estimation.In this pa-per,we utilize the NGBoost algorithm to construct a probabilistic inversion model to quantify the uncertainty associated with inversion results.By selecting a suitable machine learning model as the base learner of the NG-Boost algorithm,a hybrid machine learning model is constructed,thereby enhancing the accuracy of the inver-sion results.The performance of six different machine learning models for formation resistivity inversion is compared using the logging data of the drilling azimuthal electromagnetic induction logging instrument in lay-ered isotropic formations.The experiment resluts demostrate the pronounced advantage of XGBoost algo-rithm in terms of inversion accuracy and speed.The XGBoost algorithm as a base learner is combined with the NGBoost algorithm framework to construct the N-XGBoost probabilistic inverse model,Simulative ex-periments are conducted to verify the accuracy,reliability,and robustness of the N-XGBoost probabilistic in-version model.The results substantiate the effectiveness of the proposed model in evaluating the uncertainty of inversion results and yielding dependable inversion outcomes.This approach will provide reliable logging interpretation for geosteering work.
国家科技期刊平台
NSTL
万方数据
