The application of a graph neural network to forecast the reciprocal impact of wells within an oil field
Keywords: Well interconnection, graph neural networks, GraphSAGE, LSTM, GGNN, oil flow rat
TL;DR: Graph neural networks for data processing
Abstract: The objective of this paper is to explore a novel approach for predicting the interdependencies between production and injection wells. The method under examination involves the application of graph neural networks.
The significance of this research stems from the necessity to understand the overall impact of enhanced oil recovery methods on field production, rather than focusing solely on the individual well where the activity is implemented. Currently, there is a lack of accurate, rapid, and cost-effective methods for identifying inter-well influences. Given the pressing nature of the challenge in determining mutual influences between wells, the authors have opted to address it through the utilization of graph neural networks.
The focus of this study is on various architectures of graph neural networks.
Research methodology involves the development of a graph neural network designed to predict the mutual influence between wells by extracting weights from the trained model.
As a result, several neural network architectures have been successfully trained, with the GraphSAGE architecture utilizing a long-short term memory (LSTM) aggregation function achieving the highest accuracy. The performance metrics for this neural network are as follows: r² - 0.97, MSE - 171.39, MAE - 5.46, RMSE - 13.09, based on an average oil flow rate of 272.33.
In conclusion, this study demonstrates the promising potential of employing graph neural networks for predicting the mutual influence between wells.
Submission Number: 52
Loading