Logging Lithology Identification of Unconventional Reservoir Based on Hybrid Model

Lithology identification is an important foundation for comprehensive evaluation of shale reservoir logging. Shale reservoirs have complex and significant heterogeneity, it is of great significance to effectively recognize their lithology. However, under complex geological conditions, the relationship between logging response and lithology is weak, and lithology identification relies on the geological experience of logging interpreters; The cost of lithology data is high, sampling is difficult, and accurate characterization of lithology is difficult in the case of lack of sample size. Therefore, in this paper, we propose a lithology identification method based on the combination of convolutional neural network (CNN) and support vector machine (SVM). On the one hand, the feature information between logging data and lithology is automatically extracted by CNN, and on the other hand, SVM is utilized to solve the problem of small sample size. Using seven conventional logging curves such as spontaneous potential (SP) and deep, shallow lateral resistivity (RD, RS) as training data. The results show that compared with CNN and SVM, the performance of the hybrid model is more robust. The recognition accuracy of well H2 reaches 91.95%, which confirms the higher recognition accuracy and generalization ability of the new method.