Optimized feature selection assists lithofacies machine learning with sparse well log data combined with calculated attributes in a gradational fluvial sequence

Machine learning(ML)to predict lithofacies from sparse suites of well-log data is difficult in laterally and vertically heterogeneous reservoir formations in oil and gas fields.Meandering,braided fluviatile depositional environments tend to form clastic sequences with laterally discontinuous layers due to the continuous shifting of relatively narrow sandstone channels.Three cored wellbores drilled through such a reservoir in a large oil field,with just four recorded well logs available,are used to classify four lithofacies using ML models.To augment the well-log data,six derivative and volatility attributes were calculated from the recorded gamma ray and density logs,providing sixteen log features for the ML models to select from.A novel,multiple-optimizer feature selection technique was developed to identify high-performing feature combinations with which seven ML models were used to predict lithofacies assisted by multi-k-fold cross validation.Feature combinations with just seven to nine selected log features achieved overall ML lithofacies accuracy of 0.87 for two wells used for training and validation.When the trained ML models were applied to a third well for testing,lithofacies ML prediction accuracy declined to 0.65 for the best performing extreme gradient boosting model with seven features.However,an accuracy of~0.76 was achieved by that model in predicting the presence of the pay bearing sandstone and siltstone lithofacies in the test well.A model using only the four recorded well logs was only able to predict the pay-bearing lithofacies with~0.6 accuracy.Annotated confusion matrices and feature importance analysis provide additional insight to ML model performance and identify the log attributes that are most influential in enhancing lithofacies prediction.

Predicting total organic carbon from few well logs aided by well-log attributes

Derivative/volatility well-log attributes from very few commonly recorded well logs can assist in the prediction of total organic carbon(TOC)in shales and tight formations.This is of value where only limited suites of well logs are recorded,and few laboratory measurements of TOC are conducted on rock samples.Data from two Lower-Barnett-Shale(LBS)wells(USA),including well logs and core analysis is considered.It demonstrates how well-log attributes can be exploited with machine learning(ML)to generate accurate TOC predictions.Six attributes are calculated for gamma-ray(GR),bulk-density(PB)and compressional-sonic(DT)logs.Used in combination with just one of those recorded logs,those attributes deliver more accurate TOC predictions with ML models than using all three recorded logs.When used in combination with two or three of the recorded logs,the attributes generate TOC prediction accuracy comparable with ML models using five recorded well logs.Multi-K-fold-cross-validation analysis reveals that the K-nearest-neighbour algorithm yields the most accurate TOC predictions for the LBS dataset.The extreme-gradient-boosting(XGB)algorithm also performs well.XGB is able to provide information about the relative importance of each well-log attribute used as an input variable.This facilitates feature selection making it possible to reduce the number of attributes required to generate accurate TOC predictions from just two or three recorded well logs.