The SRT team and our development partners Eni presented talks about ECHELON software version 2.0 during a webinar we hosted recently.
The SRT team and our development partners Eni presented talks about ECHELON software version 2.0 during a webinar we hosted recently. The audience asked many interesting questions and we have shared them, and our answers, here.
Eni - Except for some advanced features such as multi-reservoir coupling or integration with the facility network solver, you can use your existing ECLIPSE™ data decks to run models with ECHELON reservoir simulation software. The former requires some ECHELON specific modifications. ECHELON binary outputs conform to the ECLIPSE™ binary format, they are therefore compatible with other software like Petrel, Office, Floviz, Tecplot, RE-Studio and S3-graph.
SRT - We have already experimented with NVIDIA MPS and MIG technologies in the context of running several small to medium models simultaneously on a single GPU to improve overall simulation throughput. For very small models (<30k cells) we observe 700% overall improvement, while the overall improvement is about 60% for models with 1M cells. These GPU reservoir simulation results will be presented at the Fifth EAGE Workshop on High Performance Computing for Upstream in September 2021.
SRT - Currently, in the market, there are solutions that allow the saving, manipulation and rendering of very large models. When it comes specifically to ensemble-based modeling, we see more of the complexity and business value in industry emerging, this is an area that we are also working on, bringing a better analysis framework to our customers where decisions can be made.
SRT - ECHELON software can simulate CO2 injection for EOR, but the current release supports dissolution of CO2 only in the hydrocarbon phases. Support for dissolution of CO2 (and other components) in brine is currently under development and is planned for commercial deployment next year. ECHELON reservoir simulation software does not currently have any special consideration from the EPA.
SRT - For representing unconventional reservoirs with hydraulic fractures, ECHELON reservoir simulation software has been used with:
1) local grids using our easy to use LGR template
2) very fine grids with a large number of cells and enhanced permeability for fractures
3) models requiring dual-permeability features.
ECHELON software supports compaction tables as is required for approximating fracture closure.
There are 3 papers jointly published with iReservoir.com using dual-permeability idealizations. The first is referenced as a case study on our website. The others can be found on our resources page and on OnePetro.
We support on-demand and cloud services. For more details, please contact us.
Eni - Experimental design
SRT - Thermal simulations are highly suited to GPU reservoir simulation considering the nature of the coupled system, however, we do not have any plans in the short-medium term to develop this capability. If it is an essential requirement for a particular customer we would be pleased to discuss options.
Eni - It is more feasible than with other simulators, but it is still challenging.
SRT - ECHELON software supports most standard models of both relative permeability and capillary pressure hysteresis. As ECHELON is exceptionally fast, even on large models, there hasn't been a reason for a coarsen feature thus far.
Eni - The question is twofold and therefore it deserves two answers:
1. The agreement between Eni and Stone Ridge Technology is, injecting reservoir engineering know-how and driving the development towards the solution of relevant problems for the energy industry. What is developed by the two companies together becomes commercially available throughout the usual sequence of software releases, typically, a major release followed by two minor releases. This approach is aimed at giving the possibility to anybody to replicate our results.
2. ECHELON software results are systematically compared against ECLIPSE™ and Intersect results, using the agreement with these simulators as the litmus test for the reliability of the solution achieved.
SRT - The webinar, including an introduction to ECHELON Software 2.0, is available here and our blog post on ECHELON 2.0 has a substantial description of ECHELON’s latest features. We will also present, jointly with Eni, a comprehensive description of ECHELON’s compositional formulation during the SPE Reservoir Simulation Conference in October of this year.
Our view is that benchmarking studies are best performed by the clients. Please feel free to contact us to arrange a trial.
SRT - Multi-segmented wells are presently under development together with a model for non autonomous (passive) ICDs for production wells that limit the inflow over reservoir well connections without the need for additional discretization.
SRT - ECHELON Software 2.0 supports low salinity flooding in black oil. Support for low salinity flooding in compositional, as well as polymer flooding, will be released towards the end of the year. Currently there is no plan, in the short-term, to develop a fully-fledged surfactant flooding support accounting for a microemulsion phase. Implementation of a simpler model may be discussed based on clients' requests.
SRT - We are actively developing an optimization capability for FNS where we try to strike a balance between finding an optimal solution for the well allocation problem under constraints and maintaining superior runtime speed. To that extent, we presently support local, gradient based algorithms that make use of the network adjoint (not in the commercial release). We will present first results during the SPE Reservoir Simulation Conference in October.
SRT - Currently, ECHELON reservoir simulation software allows users to update a surface network configuration at fixed dates during a simulation run. This logic will become more flexible in the upcoming release, making it possible to update networks based on the values of other measurable simulation quantities (e.g. field production rates).
SRT - We do not have any plans (in the short to medium term) to implement a fully implicit coupled reservoir solution that requires coupling the Jacobian terms from the reservoir and network. However, we plan to have an iteratively lagged scheme that, when converged, yields the implicit solution but lagged within the Newton loop i.e. network equations are solved first using previous Newton’s solution for the reservoir, followed by the reservoir system solution using the current solution of the network.
Eni - No, they are not embedded but we are also developing a framework for ensemble-based history match, optimization and sensitivity studies called 'ENCORE'. The speed-up comparison in our workflows is based on a combination of speed-up of the GPU simulator with respect to a modern multi-core simulator together with the more efficient use of the available hardware (e.g. we may run four jobs on one node in an hour, while using multi-cores we need two hours using 4 nodes).
SRT - The order of magnitude speed-up for HM, Opt & Sens (8x in the illustrative example above) corresponds to a factor of 2x to the fact that Eni's HPC5 cluster has 2 multi-core CPUs (Intel Gold 6252 with 24 cores each) and 4 GPUs (NVIDIA V100), and qualitatively a factor of 5x to account for the difference in memory bandwidth between said CPUs and GPUs.
SRT - As presented in the webinar, ECHELON Software 2.0 only supports coupling for black-oil models. Support for coupling compositional models to the facility network solver will come very soon, but no lumping-delumping of composition will be required. The surface network will, at first, still be modeled in black-oil, which is sufficient for most use cases where the network is needed for computing the wells' back pressures.
SRT - In order to answer this question, we would like to distinguish between two concepts: integration and coupling tightness.
ECHELON and FNS are integrated, meaning that there is a single executable and the user does not need to use a specific coupling software.
ECHELON and FNS are coupled explicitly. In the current release, two tightness levels are available, namely periodic and time-step coupling. Coupling at the Newton level (sometimes referred to as 'iteratively lagged') will come with the next release.
Eni - We did not use pressure maintenance, but voidage replacement in each reservoir using gas injection with maximum rate constraint and water as top-up phase.
Eni - It was turned off, mostly to save disk space.
SRT - Note that ECHELON software spins a specific CPU thread to perform I/O tasks, which allows the simulation to continue while writing to the report files.
SRT - There are three reasons for this:
The first is that performing flash on GPUs is embarrassingly parallel, hence ultra-fast. Using mass variables takes advantage of this by preconditioning the phase equilibrium at each Newton iteration, which can help, in particular, in near-critical cases.
The second is that the code is more easily extensible to add more phases if needed in the future (e.g., for Microemulsion or for an additional liquid hydrocarbon phase), since variables switching and lineup are not needed.
Finally, the linear systems generated by a compositional mass-variables formulation are more suited to GPUs than those generated by a natural variables formulation.
Eni - ENI geomodelling packages that can interface with ECLIPSE™ can be also used with ECHELON software. ECHELON is routinely used with Petrel™ in Eni.
Emily Fox is Stone Ridge Technology's Director of Communication.
What we are doing to help improve the reservoir simulation industry.
Leveraging modern GPUs and neural computing libraries, scientists are using deep learning technology to solve differential equations that emerge from a diverse set of physics-based problems. This blog introduces the topic and explores potential applications and limitations of the technology.Read article →