Reservoir Fluid PVT High-Pressure Physical Property Analysis Based on Graph Convolutional Network Model

In this paper, the high-pressure physical property analysis of reservoir fluid PVT (pressure–volume–temperature) was studied to improve the accuracy and efficiency of reservoir fluid property prediction. Due to the limitations of traditional laboratory measurement and theoretical model prediction methods, the graph convolutional network (GCN) model was introduced in this paper, and the enhanced ChebNet model was used to analyze the complex relationship between the high pressure physical property parameters. The key parameters such as bubble point pressure, volume coefficient, and crude oil viscosity were accurately predicted by using Chebyshev polynomial approximation and the matrix product optimization ChebNet model, which was constructed to represent the high pressure physical property parameters and their relationships. The experimental results showed that compared with linear regression, linear discrimination, random forest, and ordinary ChebNet models, the enhanced ChebNet model introduced in this paper presented significant advantages in evaluation indicators, and the AUC value reached the optimal value. This paper provides a new perspective and method for reservoir fluid PVT high-pressure physical property analysis and explores a new possibility for the application of graph convolutional networks in oil and gas exploration and development.