Prediction Method of Lithology Log Based on XGBoost Algorithm

Abstract

The photoelectric absorption cross section index is an important parameter for lithology logging, and its measurement value is affected by environmental factors such as well diameter, drilling fluid density, and the gap between the logging instrument and the well wall. Therefore, it must be corrected. Traditional environmental correction charts for photoelectric absorption cross section index correction have low efficiency and poor accuracy. To solve this problem, an XGBoost algorithm-based model for predicting the photoelectric absorption cross-section index is established, and its prediction effect is compared with that of a neural network algorithm model. The P6 well and the TS1 well data used for model training and prediction come from the same block of an oilfield in China. The learning samples are constructed using the 2 energy window count rates of the near-source detector and the 3 energy window count rates of the far-source detector of a dual-detector lithology density logging instrument. Noise is added to the samples to improve the robustness of the model. The 8 hyperparameters of the XGBoost algorithm are optimized using grid search. After completing the model training, the photoelectric absorption cross-section index prediction model is validated using the P6 well data and the photoelectric absorption cross-section index is predicted for the TS1 well. The research results show that the photoelectric absorption cross section index prediction model based on the XGBoost algorithm has higher accuracy than the model based on the neural network algorithm. The XGBoost algorithm has been applied in the logging field, and it provides a new method for improving the accuracy of lithology logging data.

Keywords

• lithologic logging
• photoelectric absorption cross section index
• machine learning
• xgboost
• hyperparameter optimization