A Comparative Study of Machine Learning Methods for Gas Hydrate Identification in Qilian Mountain Permafrost, China

Determining gas hydrate reservoirs using well‐logging data is necessary for gas hydrate resource assessment. Conventional well‐logging data interpretation can be tedious and time‐consuming. Machine learning models can automate the well‐logging interpretation process, saving time and reducing the need for the expertise of trained engineers. In this research, gas hydrate‐bearing sediments in Qilian mountain permafrost of the Qinghai–Tibet Plateau were investigated. An extreme gradient boosting (XGBoost) ensemble learning model was constructed to identify gas hydrate‐bearing sediments, with its performance benchmarked against typical machine learning algorithms including random forest (RF), K‐nearest neighbors (KNNs), and gradient boosting decision tree (GBDT). The XGBoost model achieved the highest precision (99.0%) in identifying the gas hydrate of minority class in permafrost regions, demonstrating significant improvements over RF, KNN, and GBDT models. Through dimensionality reduction and sensitivity analysis of well‐logging parameters for gas hydrate‐bearing layers, the optimal parameter combination for identification was determined as caliper, resistivity, and bulk density. The optimized identification model based on ensemble learning algorithms provides theoretical foundations and technical support for detecting gas hydrate reservoirs in permafrost regions.