Lithology Identification Based on Deep Neural Network and Theoretical Rock Physics Modeling

Abstract

Summary Lithology identification is one of the important tasks in reservoir evaluation and is the basis for solving reservoir parameters. However, the traditional lithology identification methods have problems such as large workload, poor generalization and difficulty in obtaining labels, resulting in low accuracy of lithology identification. In this abstract, multivariate properties including elastic parameters, physical parameters and fluid parameters are incorporated into lithology identification by theoretical rock physics models, and we use the lithology label data generated by the theoretical rock physics models to train a deep lithology identification network, which can both improve the accuracy of lithology identification in practical production and enhance the generalization of the neural network. The results show that the established deep lithology identification network achieves 95.82% accuracy for lithology identification of logging data and 73.8% accuracy for lithology identification of seismic data. Lithology identification based on theoretical rock physics models and deep neural network provides a new idea of lithology identification based on a two-wheel drive of data and model.