Industry leading technology for Reservoir Simulation and Optimization
Complete reservoir simulations in seconds.
Up to 10,000x Faster than conventional simulation, Physics-Informed Neural Networks (PINNs) have the potential to serve as incredibly efficient, on-demand simulators for physical processes.
Our new PINN methodology accelerates reservoir simulation dramatically. The combination of our new neural architecture and the introduction of partial differential equation (PDE) residuals allow us to learn the complex behavior of nonlinear PDEs. Our new methodology is able to perform full reservoir simulations with the same level of accuracy as traditional simulator techniques (e.g. finite element), but several orders of magnitude faster.
Assisted History Matching (AHM) orders of magnitude faster than conventional approaches.
Accurate history matching is mission critical to having confidence in your model. Traditionally, history matching was completed manually, and extremely slowly. In recent years, software providers have released tools to aid in the process, but these have resulted in their own challenges, often relying on black box simulations, and very high computational demands.
Full history match of models in subsurface systems are challenging due to the large number of reservoir simulations required, and the need to preserve geological realism in matched models. This drawback increases exponentially in large real fields due to the high heterogeneity of the geological models, the reservoir simulation computational time, and the limited amount of information available. PROTEUS-AIRE relies on a new method based on artificial intelligence to solve these limitations. Our workflow is comprised of three main components:
1) Our PINN Methodology, as outlined in the previous section; 2) A new combination of model order reduction techniques (e.g. PCA, k-PCA) and artificial intelligence for parameterizing complex 3D geomodels, called OriGen Geo-Net. Our new approach is able to create complex high-dimensional heterogeneous reservoirs in seconds, respecting higher-order statistics, hard-data constraints and geological plausibility; 3) Finally, a deep reinforcement learning (DRL) optimization framework is used to complete the assisted history match (AHM). This new approach allows us to carry out sequential optimization to do local changes in the reservoir while conserving geological plausibility.
A field development plan (FDP) determines the production strategy, specifying, among other things, the location and drilling schedule of wells. Reservoir flow simulators are traditionally used to forecast field production rates for any user-set production strategy and petrophysical properties. This allows the estimation of the economic value of an FDP candidate.
With that in mind, traditional simulators are SLOW. Simulations can take days, and full workflows can take weeks. Decisions and iterations drag on, and in the end decisions must be made on less than optimal information. Compute costs are also prohibitive for very large models (1M+ cells). The impacts of such time and compute limitations prevent the highly granular understanding necessary to achieve true optimization.
PROTEUS-FDP delivers results thousands of times faster than conventional simulators. The bottleneck of waiting on reservoir modeling is eliminated. Instead of data entry errors costing hours or possibly days, the model can simply be corrected and run again instantaneously. Instead of looking at 100 potential options for a field development plan, millions of options can be analyzed, massively enhancing solution optimization, and reducing uncertainty.