The Prediction of Coalbed Methane Layer in Multiple Coal Seam Groups Based on an Optimized XGBoost Model

Abstract

The prediction of the optimal coalbed methane (CBM) layer plays a significant role in the efficient development of CBM in multiple coal seam groups. In this article, the XGBoost model optimized by the tree-structured Parzen estimator (TPE) algorithm was established to automatically predict the optimal CBM layer in complex multi-coal seams of the Dahebian block in Guizhou Province, China. The research results indicate that the TPE XGBoost model has higher evaluation metrics than traditional machine learning models, with higher accuracy and generalization ability. The optimal coalbed methane layer predicted by the model for the Dacong 1–3 well is the 11th coal seam. In addition, the interpretation results of the model indicate that sonic (AC) and caliper logging (CAL) are relatively important in determining the optimal CBM layer. The favorable layers for coalbed methane development are distributed in coal seams with developed fractures and high gas content. The TPE-XGBoost model can help us objectively analyze the significance of different types of logging, quickly predict the optimal layer in complex multiple coal seam groups, and greatly reduce costs and subjective impact. It provides a new approach to predict the best CBM layer in multiple coal seam groups in the Guizhou Province in the southwest of China.

Keywords

• tree-structured parzen estimator
• XGBoost
• multiple coal seam groups
• coalbed methane layer prediction
• geophysical logging