Abstract: Lithology recognition plays an important role in reservoir logging evaluation, influencing the accuracy of critical parameters such as development degree and porosity. In Huizhou  26-6 well block within the Pearl River Mouth basin in the northern South China Sea, the lithology of igneous rocks is intricate, with widespread alteration significantly impacting conventional logging data. As a result, the conventional lithology identification faces difficulty in satisfying the exploration needs. To enhance the accuracy of identifying altered igneous rocks, we integrate conventional logging and element cutting logging to establish lithology identification methods through diverse machine learning algorithms. A comparative analysis leads to a comprehensive identification method of discerning altered igneous rocks. Initially, a core element data-based correction method for element cutting logging is established to obtain reliable data. Subsequently, the k-nearest neighbor (KNN) method and the support vector machine (SVM) method are employed to identify the lithology of six igneous rocks in the study area—diorite, tectonic schist, altered diabase, granodiorite, altered granite, and granite.  In the target layer of four wells with rock slice identification data in Huizhou 26-6 well block, data points (145 in total) are extracted according to the corresponding depth, of which 80% are used as training samples and the remaining 20% as test samples. Taking  sample test accuracy and  whole well lithology recognition effect as  evaluation indicators, the results of comparing the two algorithms indicate that the recognition accuracy of KNN and SVM algorithms is both 92.65%, but the whole well recognition effect of KNN algorithm is more in line with the distribution characteristics of stratigraphic lithology, indicating that the comprehensive lithology recognition based on KNN algorithm is more suitable for the study area.

Key words: altered igneous rock, SVM, k-nearest neighbor, element cutting logging, comprehensive lithological identification