Engineering Applications of Artificial Intelligence | Lithology identification with limited data based on deep learning serial ensemble optimization

By  李勇    2026-10-16    Visited 10 times

Abstract

The artificial intelligence methods represented by deep learning have been widely used in lithology identification, which requires a large number of samples. In the early stages of actual exploration and development, the number of deep-water wells drilled is relatively limited, and the number of samples available for training is small, leading to the problem of overfitting and insufficient generalization that can easily occur in the practical application of deep learning techniques. This paper proposes a serial ensemble based on deep learning optimization method designed to enhance the learning capability of deep learning models with limited sample data. In this paper, multiple sub-nets are connected in series and a local-loss function is constructed for each sub-net, which induces the current sub-net to focus more accurately on the samples where the prediction of the previous sub-net is in error. On this basis, an optimization model for exponential interaction sequence ensemble gradient boosting (eGB) is successfully constructed by integrating a global loss function with an exponential interaction mechanism among the local residuals of multiple sub-networks. In this paper, two deep learning networks, deep neural networks (DNN) and long short-term memory (LSTM), are tested for ensemble optimization. The test results show that the serial ensemble of multiple deep learners can more effectively exploit the nonlinear mapping ability of deep learning and improve its prediction ability in limited samples. In addition, the optimization strategy of exponential interaction greatly improves the training efficiency of multiple deep learners while guaranteeing accuracy, which further promotes the application of deep learning in limited samples.

Paper Information:

Yubin Ma, Jiahao Zhao, Zhiyu Hou, Danping Cao. 2026. Lithology identification with limited data based on deep learning serial ensemble optimization. Engineering Applications of Artificial Intelligence, 176, 114654. doi:https://doi.org/10.1016/j.engappai.2026.114654.