Case Description
The entire offshore production asset consists of 12 gas reservoirs, 47 wells, 3 platforms, and 2 terminals connected via 2,200KM pipelines. Visualization, modeling, simulation, and ultimately to perform rapid diagnosis and production optimization of this coupled system is a huge challenge.
In this project, we start with the integration of all G&G&E data, including geometries, model parameters, historical and real-time measurements across all objects. Highly efficient 3D visualization is accomplished using a combination of novel technologies. Surrogate simulation models for each reservoir and pipeline are trained using a time-variant, deep convolutional network. This AI powered simulation model for the full asset offers 100-1,000x speedups with a high degree of accuracy. Finally, this digital twin provides a tireless engine to capture potential risks, and to deliver optimal production allocation strategies at real time.
Reservoir simulation surrogate: a E2C network consists of an "encoder-decoder" neural net with convolutional layers, and a transformer network of fully connected layers
Solution
With the overall surrogate model for the entire reservoir-well-pipeline system, scenario evaluation on thousands of variable combination become affordable. Particle swarm optimization (PSO) is then applied to yield recommendations on gas production allocation and well control scheduling.