Virtual Reality (VR) Implementation of Reservoir Models and Integration with Artificial Intelligence (AI)

Die Hu


In order to support visualizations of non-3D information and multiple reservoir models, multi-dimensional and multi-representation features are developed. The multi-dimensional feature supports the view of 3D reservoir models and the corresponding 2D information. Multiple 2D panels can be displayed together and show information on different wells. Non-3D data is represented in 3D mode using color, height, size factors.

Two prototypes are proposed as preliminary explorations of integrating VR and AI. The first one is to integrate time series forecasting algorithms with the graph user interface in the VR visualization system. LSTM and Transformer show similarly acceptable prediction results. Because of the computational cost and current data limitation, LSTM is used as a default algorithm to predict monthly oil production in the VR visualization system. The second integration attempt applies the fuzzy set theory to select refracturing candidate wells. If good data is put into calculations, the fuzzy set theory provides solutions based on engineers’ experience. In the study, field history data and physics from a reservoir simulation model are studied using the fuzzy set theory. Though the two examples are not fully or automatically integrated into the VR visualization system, they are good starts for the research.

In the future, both visualization and AI parts of the visualization will be optimized. For visualization, the graph theory-assisted visualization enhancement will be further developed. Important indexes, such as water breakthrough time and connectivity within well groups, will be simulated in graphs and visualized. For AI parts, more AI functions can be applied to digital reservoir models for data mining. Algorithms to discover patterns, identify clusters and forecast among temporal and spatiotemporal data will be implemented and integrated into reservoir engineering workflows.
Feedback from our industrial friends suggest that a two- or three-axis coordinate system can be built for custom analysis, allowing users to assign parameters to the system and analyze a correlation between different parameters.