Virtual Flow Meters (VFMs) offer a means to estimate multiphase flow rates in oil & gas production. They are cost effective and fast to deploy since they use existing sensors to deduce flow rates, and, thus, do not require costly infrastructure investments.
FLUX VFM is a hybrid technology that can combine physics-based modeling with machine learning. Pure machine learning solutions rely solely on measurements to deduce relations between observables and require vast amounts of data. A physics-based simulator, for example our FLUX Simulator, builds on well-established laws of physics. It uses a simulation model to represent the physical system it describes. Thus, it requires additional information as compared to pure machine learning, but significantly less data to fine-tune the simulation model and ensure best performance possible. So, what does it take to set up a simulation model?
A key component for any flow modeling is the fluid description since fluid properties change with pressure and temperature. This implies that the character of the fluid at reservoir conditions differs significantly from its character when it reaches the process plant at much lower pressure and temperature.
FLUX Simulator uses tabulated fluid properties on a pressure-temperature grid. External 3rd-party PVT packages can provide these properties based on a characterized fluid description. Typically, PVT reports provide the input necessary to create such a fluid description. However, as PVT sampling and subsequent analysis is costly, it is not uncommon to encounter decade-old reports. Here it is important to notice that up-to-date fluid sampling is not a necessity for virtual flow metering. In addition, our FLUX VFM does not require periodic sampling for recalibration. The PVT report only forms a starting point for the FLUX VFM self-calibration, which automatically adjusts GOR and water cut.
Thermo-Hydraulic Simulation Model
Simulation of fluid flow in a pipe requires information about the conduit, i.e., information about the trajectory and pipe inner diameter. Given fluid properties, this accounts for the hydraulic part of the fluid flow. The fluid properties depend on the fluid temperature, which in turn depends on heat transfer and thermal interactions with the surroundings. For best performance, an accurate model for heat loss is desirable. To determine the heat transfer, you can either use a U-value or, preferably, detailed information about the well completion or pipeline. The latter option requires information about the materials and material thicknesses that encase the fluid flow. For a well, this includes information contained in a well schematic such as tubings, casings, and fluids or cement in the annuli. For a pipeline in its simplest form, this is the steel pipe and any insulation.
If we address heat transfer, information about the ambient temperature is a minimum requirement. However, to achieve best accuracy you need additional information about what surrounds the conduit. For an onshore pipeline, we must know the temperature and windspeed to determine the heat exchange with the air. Similarly, for an offshore pipeline, we need metocean data, i.e., the temperature and currents, to determine the heat exchange with the water. For wells, we need subsurface information about the formation as well as the geothermal temperature profile. If we then look at an offshore well with a dry tree, all of the above apply; subsurface information from the reservoir to the seabed, metocean data from seabed to the sea surface, and air temperature and windspeed from sea level to the platform.
Further, the simulation model should account for equipment along the conduit that impacts the thermo-hydraulics of the fluid flow. A non-exhaustive list includes chokes and valves, pumps, compressors, and heat exchangers, components which affect the pressure and temperature profiles along the conduit. To account for such effects, FLUX Simulator incorporates equipment models. These models require information about the specific equipment installed, e.g., choke curves or pump curves provided by the vendors.
The last part to address for the physics-based simulation model is the fluid inflow. Well models start at the reservoir contact and Inflow Performance Relationships (IPRs) provide the inflow. The IPR relates the inflow to the drawdown, i.e., the difference between the reservoir pressure and the flowing bottomhole pressure. The inflow phase distribution is determined through self-calibration primarily based on existing sensor data.
With a simulation model at hand, the next step is to investigate the Piping and Instrumentation Diagrams (P&ID) to understand which sensors are available. In deployment, these sensors serve multiple purposes. Firstly, they provide real-time information about boundary conditions as well as operational settings such as choke openings. Secondly, they offer points of comparison between the real and virtual systems, i.e., they offer quality assurance. Thirdly, they provide information to the self-calibration.
While the underlying physics may be well-known, uncertainties apply since it is impossible to have absolute control over all model parameters that impact a physics-based simulator. The models offer a good starting point and with additional calibration, it is possible to ensure that a simulation model describes reality as accurately as possible.
To fine-tune the thermo-hydraulic model, reference data is used. The reference dataset contains both sensor data and flow rate measurements, e.g., from well tests, and relates the pressure and temperature drops to known flow rates. By matching the simulated pressure and temperature drops to measurements for known flow rates, it is possible to calibrate both the hydraulic modeling and heat transfer models. The same approach applies to equipment where vendor-supplied information is representative rather than specific to the installed equipment.
Apart from wear and tear or major changes in flowing conditions, the physics tuning is robust and remains unchanged.
Here, it is important to notice the difference between calibration of a physics-based simulation model and training of machine learning models. Where machine learning algorithms require vast amounts of data to determine relationships between properties, physics-based models make do with much less. Nevertheless, it is worth noting that the best calibration is achieved when the reference data span a broad range of operating conditions.
While the thermo-hydraulic models and their calibration are robust, reservoir inflow requires more care. As discussed above, fluid samples are few and far between and we cannot rely on up-to-date fluid information to address changes in GOR and water cut. It is of course possible to use reference data to deduce these properties. However, it may neither be desirable nor feasible to gather reference data on a schedule that ensure the capture of rapid changes associated with, e.g., gas or water break-through.
To address these challenges, FLUX VFM incorporates self-calibration based solely on sensor data with no need for additional reference measurements.
Are You Ready for Virtual Flow Metering?
In live deployment, a Virtual Flow Meter provides real-time flow rates based on existing sensor data. With automated self-calibration, our FLUX VFM delivers accurate flow rates at all times. So, if you want to harvest the benefits of a physics-based FLUX VFM and access best-in-class flow rate predictions, start by gathering:
- Well trajectory
- Casing & tubing
- Ambient conditions
Metocean data [if applicable]
Air temperature & wind speed [if applicable]
- Equipment characteristics (from vendor) [if applicable]
Contact us to discuss how we support your field operations.
You may also be interested in the following paper:
Evaluation of Data Driven Versus Multiphase Transient Flow Simulator for Virtual Flow Meter Application (in collaboration with Petronas, presented at OTC Asia 2020). Access through OnePetro.