A method for estimating the mineral contents from well logs using physics-informed neural networks

The composition contents of various minerals in the rock are a key concern in geophysical exploration and development. It is essential for lithology classification, the quantitative assessment of mineral resource potential, and reserves prediction. However, accurately calculating these mineral components is often highly challenging for formations with complex lithology, particularly when core samples and formation elemental logging data are scarce. In recent years, with the rapid development of artificial intelligence, utilizing big data and deep learning technologies to improve the accuracy and efficiency of well logging interpretation has become a research hotspot. Nevertheless, traditional data-driven models suffer from a lack of interpretability, which imposes certain limitations on their practical application. As a novel model integrating physical laws, Physics-Informed Neural Networks (PINNs) can constrain prediction results, rendering them more physically meaningful.

In this study, we propose a mineral content prediction model specifically designed for formations with complex mineral types. The model is capable of accurately calculating mineral contents using conventional logging data. First, based on the mineral types present in the formation, forward modeling is used to generate data and construct the training dataset. Subsequently, a CNN (Convolutional Neural Network) model is employed to predict the mineral content. By simultaneously constructing data loss and physical loss functions, the interpretability of the prediction results is ensured. The physical loss is mainly constructed by the volume model. The validity of the model is verified using forward modeling data. Finally, the model is applied to the processing of real logging data. The prediction results demonstrate good consistency with the mineral content obtained from X-ray Diffraction (XRD) analysis of core samples indicating that the model can accurately reflect the variations of complex mineral contents. This study provides a new method for the evaluation of mineral content, which is expected to offer a potential technological pathway for the identification of deep-seated ore bodies and the estimation of resource reserves.

This work is supported by National Science and Technology Major Project for Deep Earth Probe and Mineral Resources Exploration under Grant 2025ZD1008500.