In-Situ Stress Prediction Model for Tight Sandstone Based on XGBoost Algorithm
Du Tong, Li Yuwei
School of Environment, Liaoning University, Shenyang, China
Keywords: In-situ stress, XGBoost, tight sandstone, machine learning
Abstract
This article uses XGBoost algorithm to calculate rock in-situ stress. By using Pearson correlation coefficient method, it is determined that the logging parameters with the best correlation with minimum horizontal principal stress are Depth, GR, LLD, ILD, AC, VCA, with maximum horizontal principal stress are: Depth, GR, SP, CAL, DEN. In order to verify the performance of the model, linear regression, support vector machine, and random forest models are used for comparison. In order to improve the generalization performance, the k-fold cross-validation method is used. The results show that using XGBoost algorithm to predict rock in-situ stress with a small amount of data has a high average accuracy of 94% and good generalization performance. The linear regression model has a faster fitting speed, but the fitting accuracy is the lowest. The random forest and support vector machine models are in-between. The result confirms that the research method in this article has certain universality and can be extended to solve other rock in-situ stress prediction problems.
|
