Abstract
<ns0:p>This paper summarizes a machine-learning competition, which is the first public benchmark comparing machine-learning (ML)-based automatic depth-shifting methods for well logs. Accurate depth determination of subsurface formations is critical for various applications in the oil and gas industry. However, achieving accurate depth registration of well logs can be challenging due to inherent ambiguity and subjectivity. Conventional depth-shifting methods relying on manual bulk shift based on peaks and troughs can be tedious and time consuming. The 2023 Machine-Learning Competition aimed to evaluate and compare various data-driven techniques for automatically aligning well logs to a reference log and correcting depth misalignments. It was organized by the Petrophysical Data-Driven Analytics (PDDA) Special Interest Group (SIG) of the Society of Petrophysicists and Well Log Analysts (SPWLA). Various machine-learning algorithms and deep-learning architectures from the top five teams were explored, including techniques like dynamic time warping (DTW), convolutional neural networks (CNNs), recurrent neural networks (RNNs), and ridge regression, known for their ability to learn complex patterns from data. The performance of the submitted solutions was evaluated using metrics like root mean squared error (RMSE) and mean absolute deviation (MAD). The contest results, presented in this paper, highlight the effectiveness of data-driven methods in automatically rectifying well logs and provide insights into the strengths and weaknesses of different approaches for this task. The findings contribute to the advancement of automated depth-shift techniques and their potential application in subsurface characterization and reservoir management. The code and data set are available at https://github.com/pddasig/Machine-Learning-Competition-2023. This paper only gives a high-level summary of this competition, and we refer readers to the GitHub repository for a more in-depth understanding of the works.</ns0:p>