Optimization of an XGBoost classification model using grid search and genetic algorithm for lithology identification based on well logging data

Objective Lithology identification is very important in oil and gas exploration and geological research. Traditional lithology identification mainly relies on expert experience and basic statistical analysis, which is difficult to adapt to the nonlinear characteristics of complex geological conditions.

Methods In order to improve the accuracy and model stability of lithology identification, the XGBoost classification model is constructed by feature screening and normalization of logging data Combined with grid search (GS) and genetic algorithm (GA), a hybrid optimization strategy is constructed to optimize the hyperparameters of the model globally and locally, so as to enhance the learning ability and generalization ability of the model. Taking Accuracy, Precision, Recall and F1 scores as the model evaluation criteria, taking Xi36 well as an example, 14 wells in Xilinhaolai area were tested and verified.

Results The experimental results show that compared with traditional methods such as KNN, SVM, RF, XGBoost and LightGBM, the optimized model has improved classification accuracy, model stability and generalization ability, and can more accurately identify lithology types in different reservoir environments.

Conclusions This research provide more reliable technical support for reservoir evaluation and geological modeling in oil and gas exploration.