Lithology identification of intrusive rocks based on petrochemical big data

To address the high cost and low efficiency of lithology identification for intrusive rocks in oil and gas exploration, this study proposes a hierarchical identification method based on petrochemical big data and machine learning. By integrating global geochemical databases, a standardized sample set covering ultramafic to felsic intrusive rocks was constructed, and a three-level classification system (“group–subgroup–specific lithology”) was established. Visual discrimination charts and an automatic identification model were developed using Linear Discriminant Analysis and Multilayer Perceptron, respectively. The results show that the method achieves over 90% accuracy in the first- and second-level classifications, effectively identifying major rock types such as gabbro, diorite, and granite. Although the accuracy fluctuates in the third-level classification due to compositional overlap and data quality issues, the method exhibits good interpretability and generalizability. This approach provides a low-cost and efficient technical solution for rapid lithology identification and reservoir evaluation, demonstrating potential application in deep and unconventional hydrocarbon exploration.