Using logging data to identify stratum lithology is a key link in petroleum exploration engineering. The lithology classifier based on machine learning has become a current research hotspot due to its excellent performance. Aiming at the problem that existing methods rely too much on relevant personnel's parameter adjustment experience, combined with the strong robustness of the XGBoost integrated classifier and the fast search ability of the differential evolution algorithm, a machine learning lithology identification method that can automatically adjust parameters is proposed. Differential evolution iterates the parameter population of XGBoost classifier to obtain the global optimal lithology recognition model. Based on the actual logging data of Daqing Oilfield, comparative experiments on lithology identification based on BP network, Randomforest, XGBoost have been carried out. The results show that the proposed method has good performance in overall accuracy, stability and recognition accuracy of each sub-category.