Application Of Machine Learning In Shale Oil Logging Evaluation

In the 21 st century,the oil shale industry has recovered due to rising global oil prices.In 2019,the U.S.has extracted more than half of its crude oil production from shale tight oil.China’s 12 th Five-Year Plan and 13 th Five-Year Plan have also elevated the oil shale industry to an important strategic position.However,shale oil resources are more difficult and costly to extract.Logging evaluation has important research significance as an important part of the oil shale industry.The application of machine learning to logging evaluation can help to further improve the accuracy of logging evaluation based on traditional methods and provide new ideas for discovering the inherent laws of logging data.Reservoir evaluation in logging evaluation is a supervised-learning multiclass classification problem in machine learning.The reservoir was evaluated using data from the test oil of the long 7 hydrocarbon source rock in the Ordos Basin.After cleaning the data,the features are then processed using feature engineering.After pre-processing the data,the training set is trained with K-neighborhood,logistic regression,support vector machine,decision tree,and random forest algorithm,the model with better effect is selected,its parameters are further adjusted,and finally its results are verified on the test set.At the same time,a plug-in for professional loggers has been developed based on the CIFlog logging platform.The plug-in uses a vector machine algorithm that can directly read data from the CIFlog platform for processing.On the test set,the random forest model was 85.31%,88.86% and 85.85% for accuracy,precision and recall,respectively.The precision of the oil layer can reach 85.35% and the recall rate can reach 87.27%,in line with the expected design indicators.A support vector machine plug-in based on the CIFlog platform that can be applied to more scenarios in logging evaluation,providing reference for professional loggers.