Logging identification of lithology in fine-grained sedimentary rocks based on the FSSA-HKELM model: a case study of the Qingshankou Formation in the Songliao Basin (NE China)

Abstract

The lithology distribution of fine-grained sedimentary rocks in regional space plays a crucial role in guiding the exploration of oil shale layers. Identifying lithology by conventional logging information is efficient and cost-effective. However, the strong inhomogeneity of fine-grained sedimentary rocks leads to a complex nonlinear relationship between lithology and logging data, making conventional linear methods no longer applicable. This study proposes a hybrid model for logging lithology identification of fine-grained sedimentary rocks from the first member of Qingshankou Formation in the Songliao Basin, NE China. This model is based on the hybrid kernel extreme learning machine (HKELM), and the firefly perturbation strategy is introduced into the sparrow search algorithm (FSSA) for optimization. The lithologic distribution is determined using cores, thin sections, and total organic carbon (TOC), while a total of four logging curves, the acoustic (AC), density (DEN), resistivity (RT), and natural gamma (GR) curves, were collected. FSSA-HKELM was compared with five algorithms, ELM, KELM, HKELM, PSO-HKELM, and SSA-HKELM, for lithology prediction effectiveness. The proposed hybrid method outperformed the other algorithms, achieving an accuracy of 81.78%, precision of 81.65%, recall of 87.78%, and a weighted F1 score of 82.16%. FSSA-HKELM is a very effective lithological identification method, which provides a basis for the lithological prediction of oil shale-bearing formations.

Keywords

• lithology identification
• sparrow search algorithm
• firefly perturbation strategy
• hybrid kernel extreme learning machine
• fine-grained sedimentary rocks