Abstract
Fluid prediction is important in exploration work, helping to determine the location of exploration targets and the reserve potential of the estimated area. Machine learning methods can better adapt to different data distributions and nonlinear relationships through model training, resulting in better learning of these complex relationships. We first use the Gram angle field (GAF) to convert one-dimensional logging data into two-dimensional images. GAF can better capture the nonlinear structure and patterns in time series data by using trigonometric transformation. After that, we used the Swin Transformer model to classify the converted images. It captures the locality and timing of the image by moving the window. Swin Transformer uses a staged attention mechanism that allows the model to efficiently capture feature information at different scales. This allows the model to capture both local and global information in the image, contributing to a better understanding of the image content. The multi-scale feature capture capability of the Swin Transformer enables it to effectively capture different scales and spatial relationships in fluid prediction tasks. Tested in real data from Tarim Oilfield, the GAF-Swin Transformer model has better performance than other machine learning models. This study provides a new perspective in the field of fluid prediction.
Subject
Condensed Matter Physics,Fluid Flow and Transfer Processes,Mechanics of Materials,Computational Mechanics,Mechanical Engineering