Abstract: Traditional lithology identification methods, such as cuttings logging, drilling coring, and logging data interpretation techniques, are highly dependent on logging quality, have low identification accuracy and efficiency, and have poor generalization capabilities. With the rapid development of computer technology, combining logging data with computer technology to carry out lithology research has become an effective means of lithology identification. This paper proposes a lithology recognition method based on gradient boosting algorithms XGBoost and LightGBM. Taking the lower carbonate reservoir in Block 41-33 of Sudong gas field in Sulige gas field as an example, test and verify it, using the acoustic time difference, natural gamma, photoelectric absorption cross-section index, density, and deep lateral resistivity in the logging data. Lithology identification is carried out with six parameterssuch as compensation neutron, and compared with traditional algorithms such as KNN, naive Bayes and support vector machine. The results show that the accuracy of lithology identification of the three traditional algorithms is 78.45%,74.43% and 78.72%, the recognition accuracy rates of XGBoost and LightGBM based on gradient boosting algorithms reached 98.90% and 98.72% respectively, which are much higher than traditional algorithms.

Key words: lithology identification, gradient boosting algorithm, carbonate rock, decision tree