Dynamic graph convolutional networks for fluid identification of well logging data transformed through the gram angle field

Accurately predicting the fluids holds immense significance in exploration work, assisting in the identification of exploration targets, estimation of reserve potential, and evaluation of reservoirs. In our research, we employed an innovative approach by using the gram angle field (GAF) to transform logging parameters. By adeptly capturing time series information and converting one-dimensional data into two-dimensional matrix representations, GAF takes into account not only the values at each time point but also their relative position and order. This method effectively preserves the temporal evolution characteristics of the original data. The resulting Gram Angle Field matrix can be viewed as a two-dimensional image, facilitating visualization and analysis through image processing techniques. Additionally, we introduced the dynamic graph convolutional network (DGCN) to segment the transformed images. The DGCN structure, employed for feature learning, can extract more comprehensive and representative feature representations from the logging data. Since logging data demonstrate a time series relationship, indicating a temporal correlation between logging curves at different depths, DGCN utilizes dynamic graph structures to capture and comprehend this time series information. This capability enables DGCN to model the evolution process of well log data effectively. DGCN assigns varying weights to nodes and edges at each time step, updating the current node representation with information from neighboring nodes. This localized approach enables DGCN to meticulously focus on significant features at each time step, facilitating the identification of potential patterns and trends in the logging data. Our research not only paves the way for advancements in the field but also provides valuable insights for geologists and professionals engaged in oil and gas exploration.