Patent Application
20220414303
The invention relates to measurement of fluid flow properties of a porous rock medium, and in particular to predicting a three-dimensional imbibition phase saturation profile for imbibition of the porous rock medium. The invention further relates to a measurement system for predicting a three-dimensional imbibition phase saturation profile for a porous rock medium and a method for training of the machine learning algorithm for predicting three-dimensional imbibition phase saturation profile.
Extracting hydrocarbons in the form of fluids from subsurface rocks in a subsurface involves an understanding and ability to predict a fluid's movement through the subsurface rocks. Fluid flow of the hydrocarbons and water in the subsurface is significantly impacted by rock heterogeneity of the porous subsurface rocks. An understanding of the fluid flow movement of the hydrocarbons inside the heterogeneous porous medium of the subsurface rocks is a relevant factor in field development planning for oil and gas fields. A wide range of lab experiments are therefore conducted on rock samples, such as core plugs extracted from the subsurface rocks, in order to determine properties of the fluid flow within these subsurface rocks.
Special core analysis laboratory data, oftentimes also referred to as “SCAL-data”, is used to understand the fluid flow movement inside the rock samples. This SCAL-data includes, for example, capillary pressure and relative permeability curves of the rock samples. These capillary pressure and relative permeability curves are oftentimes obtained from post processing analysis of experimental measurements performed on the rock samples that are extracted from the subsurface for analysis and evaluation. The SCAL-data such as relative permeability curvature and end-point values are direct indicators of the fluid flow of the hydrocarbon and the water inside the porous rock medium of the rock samples in the subsurface.
Linear X-Ray core flooding equipment is used to determine the fluid flow properties of the porous rock medium. This determining of the fluid flow properties includes capturing information on a one-dimensional phase saturation inside the porous rock medium. Even small variations in the capillary pressure characteristics of the porous rock medium can, however, lead to a saturation heterogeneity. These variations in the capillary pressure characteristics impact results for the relative permeability curves derived from the X-Ray core flooding experiment. The one-dimensional phase saturation information determined from the X-Ray core flooding experiments does therefore not suffice to estimate the relative permeability and the capillary pressure in the rock samples, for example in cases where the rock samples show a strong heterogeneity. This heterogeneity of the porous rock medium impacts the flow of the fluids and trapping of the hydrocarbons in the subsurface on a local level and on a field scale. Understanding and predicting these fluid flow properties is relevant for the extraction of the hydrocarbons from the subsurface rocks. The effect of the heterogeneity on the SCAL-data needs to be characterized and incorporated into a field development plan.
Therefore, capturing three-dimensional water saturation profiles enable an understanding of the fluid flow properties within the porous rock medium. Estimating the capillary heterogeneity and relative permeability also requires a knowledge of distribution of the fluid saturations in the porous rock medium. Medical computer tomography (medical-CT) equipment is therefore used for conducting of so-called “core-flooding experiments” aiming at capturing three-dimensional phase saturation information of the core plugs extracted from the porous rock medium. This acquired three-dimensional phase saturation information is then used in numerical reservoir simulation to quantify capillary heterogeneity and relative permeability of the rock medium. However, extracting a three-dimensional phase saturation profile for each fractional flow is challenging, time-consuming, and expensive. There therefore is a strong need for a time-efficient and cost-effective approach to generate three-dimensional phase saturation profiles.
Various approaches are known in the art for simulation of fluid flow properties within the porous rock medium of the subsurface area. International patent application WO 2019/210102 A1 (Dogru; assigned Aramco Services Co., Saudi Arabian Oil Co.) describes a computer implemented method for the reservoir simulation of fluid flow in a producing reservoir. The method is used for improving the convergence of the determination of the pressure distribution within the reservoir. The pressure distribution is then used to determine initial pressures in the reservoir and estimate the fluid flow in the reservoir. The reservoir comprises rock having heterogeneous properties of permeability and porosity. The reservoir is organized into a three-dimensional grid of reservoir cells. The simulation of the reservoir is performed for a sequence of time steps based on fluid pressures from wells in the reservoir. The method and system do not disclose the use of machine learning for the calculation of the fluid flow.
U.S. Pat. No. 10,718,188 B2 (Dinariev et al; assigned to Schlumberger Technology Corp.) relates to the simulation of a portion of a subterranean formation containing a field (oil or gas reservoir). The method and system disclose the construction of a digital model to represent a portion of the field for the purposes of making decisions regarding the development of the field. The computer model represents the physical space of the reservoir by an array of discrete grid blocks, delineated by a grid which may be regular or irregular. Each block in the grid represents a subsurface volume. The array of grid blocks may be two-dimensional or three-dimensional. Values for physical attributes, such as porosity, permeability and liquid or vapor hydrocarbon saturation, may be associated with each grid block. The value of each attribute may vary across the reservoir volume, but the value is applied uniformly throughout the volume of the grid block. As an example, simulations may solve a complex set of non-linear partial differential equations that model the fluid flow in porous media over a sequence of simulation time points. Grid resolution may impact the accuracy of simulation results, such as in a highly heterogeneous reservoir. However, for practical applications, upscaling techniques may be used to capture fine scale phenomena while using relatively coarse grids. In one or more examples in the document, an upscaling technique is used for simulation of fluid hydrocarbons flow through a heterogeneous reservoir to estimate the volumes of fluid hydrocarbons that may be recoverable. In a further example, the upscaling process may be initiated by the results of a set of pore scale models simulations, whereby the pore scale models used in the simulations represent the pore geometry of the reservoir rock. The method and system do not, however, disclose the use of AI and/or machine learning for the calculation of the fluid flow.
U.S. Pat. No. 10,198,535 B2 (Li et al; assigned to ExxonMobil Upstream Research Co.) discloses a method and system for modeling a hydrocarbon reservoir that includes generating a reservoir model. The reservoir model contains a plurality of volume elements. At least one of the volume elements is simulated using a training simulation to obtain a set of training parameters. The training parameters are, for example, state variables and boundary conditions of the sub region. A machine learning algorithm is then used to approximate an inverse operator of a matrix equation that provides a solution to fluid flow through a porous media. The approximation is done based on the set of defined training parameters. The flow in the (hydrocarbon) reservoir can then be simulated using the inverse operator approximated for the at least one volume element. The method also includes generating a data representation of a physical hydrocarbon reservoir that can be generated in a computer-readable medium based, at least in part, on the results of the simulation. The system and method do not, however, offer a solution for identifying a relationship between the oil distribution during drainage and imbibition of the subsurface area.
The present document describes in-situ three-dimensional visualization of phase saturation during water displacing oil process similar to that of three-dimensional saturation profile generated by medical-CT based core flooding equipment.
A computer-implemented method for approximating a predicted three-dimensional imbibition phase saturation profile for imbibition of a porous rock medium is described. The approximating of the three-dimensional imbibition phase saturation profile is done using a machine learning algorithm. The method comprises a first step of inputting at least one of a measured three-dimensional drainage phase saturation profile into the trained machine learning algorithm. The method further comprises the additional steps of inputting a derived one-dimensional drainage phase saturation profile into the trained machine learning algorithm and inputting a one-dimensional imbibition phase saturation profile into the trained machine learning algorithm. The method also comprises the further step of approximating the predicted three-dimensional imbibition phase saturation profile using the using the trained machine learning algorithm, wherein the approximating is done using the processor.
The machine learning algorithm for use in the method is trained using a simulated one-dimensional drainage phase saturation profile, a simulated three-dimensional drainage phase saturation profile, a simulated one-dimensional imbibition phase saturation profile, and a simulated three-dimensional imbibition phase saturation profile.
A measurement system for prediction of a three-dimensional imbibition phase saturation profile is also disclosed. The measurement system for the imbibition of a porous rock medium comprises a processor, a memory, and a detection device. The processor is used for executing at least one of a porous media fluid flow simulator and a machine learning algorithm. The memory is electronically connected to the processor and is used for storing items of data. These items of data stored in the memory comprise at least one of an oil-water contact angle, a relative permeability, and a capillary pressure of a target core plug.
The porous media fluid flow simulator is used to generate SCAL-data which is fed to reservoir simulation models for the prediction of long term oil and gas field behavior as part of a field development planning. The detection device of the system comprises at least one of a linear X-Ray scanner and a medical-CT core flooding equipment.
A computer-implemented method for training a machine learning algorithm is also disclosed. The machine learning algorithm is used to approximate a three-dimensional imbibition phase saturation profile for imbibition of a porous rock medium. The method for the training of the machine learning algorithm comprises the steps of deriving a derived one-dimensional drainage phase saturation profile from a measured three-dimensional drainage phase saturation profile, wherein this deriving is done using a processor. The method further comprises calculating a plurality of synthetic values for a relative permeability of the porous medium and a capillary pressure of the porous medium. This calculating is done using the processor and a measured oil-water contact angle in the porous medium. The method also comprises the feeding of a measured three-dimensional drainage phase saturation profile, the derived one-dimensional drainage phase saturation profile, a one-dimensional imbibition phase saturation profile, the measured oil-water contact angle, the calculated synthetic values for the relative permeability, the calculated synthetic values for the capillary pressure, and a rock heterogeneity state into a porous media fluid flow simulator. The method further comprises the step of generating both a simulated three-dimensional drainage phase saturation profile and a simulated three-dimensional imbibition phase saturation profile using the porous media fluid flow simulator. Using the processor, a simulated one-dimensional drainage phase saturation profile is calculated from the simulated three-dimensional drainage phase saturation profile, and a simulated one-dimensional imbibition phase saturation profile is calculated from the simulated three-dimensional imbibition phase saturation profile. The method further comprises training of the machine learning algorithm. The training is done using the simulated one-dimensional drainage phase saturation profile, the simulated three-dimensional drainage phase saturation profile, the simulated one-dimensional imbibition phase saturation profile, and the simulated three-dimensional imbibition phase saturation profile.
The method further comprises measuring a three-dimensional drainage phase saturation profile using an in-situ core flooding monitoring tool comprising a CT-scanner and the measuring of fluid properties of a crude oil.
The method also comprises a determining of a sister plug for the selected core plug and performing an ageing operation. The method also comprises measuring an oil-water contact angle in the porous medium using analytical or experimental approaches.
The method further comprises determining a rock heterogeneity state by scanning a dry core plug using an CT scanner, for example, Medical CT scanner and micro CT scanner.
The invention will now be described on the basis of the figures. It will be understood that the embodiments and aspects of the invention described herein are only examples and do not limit the protective scope of the claims in any way. The invention is defined by the claims and their equivalents. It will be understood that features of one aspect or embodiment of the invention can be combined with a feature of a different aspect or aspects and/or embodiments of the invention.
The fluid and chemical properties of the crude oil and brine are determined in step S110 using, for example, a laboratory Pressure-Volume-Temperature cell, also referred to as “PVT cell”. Using the PVT cell, the properties of the crude oil such as the molecular weight of the crude oil, the bubble-point pressure of the crude oil, or the oil formation volume factor can be determined.
In step S120, a sister plug 55 is identified for the selected core plug 50. The sister plug 55 is a sample that is extracted from the subsurface area. The sister plug has similar petrophysical properties and heterogeneity compared to the selected core plug 50. This selecting of the sister plug 55 becomes necessary because the analysis of the petrophysical properties of the selected core plug 50 may compromise the results for other laboratory experiments conducted on the selected core plug 50. For example, the measurement of the porosity on the selected core plug 50 may compromise the wettability of the selected core plug 50. The sister plug 55 therefore has to be used for certain laboratory experiments, as will be explained later.
A three-dimensional drainage phase saturation profile 25m for the selected core plug 50 is determined in step S130. The determining of the drainage phase saturation profile 25m is done using a steady state oil displacing water (also called “drainage”) core flooding lab experiment with a three-dimensional in-situ saturation monitoring tool (see Jackson et.al, 2018). The one-dimensional drainage phase saturation profile 25d is derived from the measured three-dimensional drainage phase saturation profile 25m using the processor 110. The one-dimensional drainage phase saturation profile 25d is calculated in Step S140 as a mathematical average of the three-dimensional drainage phase saturation at each circular cross section along the core plug 50.
An ageing operation is performed in an ‘ageing cell’ using representative crude oil on the sister plug 55 in step S150.
An oil-water contact angle 60 in the sister plug 55 is measured using an analytical or an experimental approach in step S160. The oil-water contact angle 60 in the porous medium 70 describes an angle of intersection of the interface between the two fluids and a porous solid at a solid surface. The oil-water contact angle 60 is geometrically defined as the angle formed by a liquid at the three-phase contact point between where the porous medium 70, the oil and the water intersect. For example, the sister plug 55 is initially cleaned and saturated with water then flooded with oil to reach a state representative of the reservoir. The sister plug 55 is aged using ‘ageing cell’. The oil-water contact angle 60 is measured from the solid surface of the porous medium 70 of the sister plug 55 through the oil phase of the porous medium 70 using X-ray micro tomography based equipment such as a microCT scanner or other proven technologies such as a pendent drop method.
This experimental approach for the determining of the oil-water contact angle 60 comprises experimental techniques that are commonly used in the oil and gas industry to measure the contact angle between oil and water (see Andrew et.al, 2014; AlRatrout et.al, 2018; Khishvand et.al, 2017). The oil-water contact angle 60 is determined using micro-CT image data of the sister plug. This micro-CT image data is processed to remove artefacts from the image data. A sub-volume of the micro-CT image data is extracted and re-segmented. This sub-volume is also referred to as “slice”. Visibility of edges of phases the fluids and the solids in the sister plug in the slice are enhanced using available image filtering techniques such as a 3D Sobel filter. This enhanced slice of micro-CT image data is also referred to as a “resampled slice”. A binary filter is applied to the resampled slice. The oil-water contact angle 60 is measured manually by tracing vectors tangential to the solid surface and the liquid interface seen from the filtered resampled slice of the image data from the sister plug (see Andrew et.al, 2014). This experimental approach for the determining of the oil-water contact angle 60 further comprises estimating of the contact angle using an analytical approach based on the measured drainage capillary pressure and measured imbibition capillary pressures (see Masalmeh, 2001).
A plurality of synthetic values for the relative permeability 80 and a capillary pressure 90 of the selected core plug 50 are calculated in step S170. This calculating is done using mathematical methods such as but not limited to Brooks-Corey, LET and modified Corey models that are available in literature to define the capillary pressure and relative permeability curves. These known models can be used with a plurality of synthetic values for the relative permeability 80 and the capillary pressure 90 of the selected core plug 50. For example, the synthetic relative permeability tables are ensured by varying Corey's exponents nwd and nod, saturation end-points SWC and Kro*, and the fitting parameters cwd, cod, αWd, αod and bd as described in the modified Corey model (Masalmeh et al, 2007).
The rock heterogeneity state 40 for the selected core plug 50 is determined in step S125 using established approaches, such as determining a porosity variation inside the core plug using dual energy X-ray based CT scanner equipment (see Larmagnat et.al, 2019). The rock heterogeneity state 40 describes a quality of the variation of the rock properties within the core plug 50. This rock heterogeneity state 40 is determined using well established approaches This rock heterogeneity state 40 makes petroleum system modeling, formation evaluation, and reservoir simulation critical to maximizing production from oil and gas reservoirs. For example, a CT scan of the core plug 50 is used to identify the rock heterogeneity state 40.
A simulated one-dimensional drainage phase saturation profile 20s is calculated from the simulated three-dimensional drainage phase saturation profile 25s and a simulated one-dimensional imbibition phase saturation profile 30s is calculated from the simulated three-dimensional imbibition phase saturation profile 35s in step S210 as the mathematical average of the simulated three-dimensional imbibition phase saturation profile 35s at each circular cross section along the core plug 50 using the mathematical average calculation method described in step S140 (see above).
The machine learning algorithm 130 is trained using, as a training data set, the plurality of synthetic values for the relative permeability 80 and the capillary pressure 90 of the selected core plug 50 (as determined in step S170), the determined rock heterogeneity state 40 (as determined in step S125), the simulated three-dimensional drainage phase saturation profile (25s), the simulated one-dimensional drainage phase saturation profile 20s, the simulated three-dimensional imbibition phase saturation profile 35s, and the simulated one-dimensional imbibition phase saturation profile 30s. The machine learning algorithm 130 comprises a supervised deep learning algorithm, an unsupervised deep learning algorithm, or a reinforcement deep learning algorithm. The machine learning algorithm 130 comprises proprietary frameworks such as, for example, Tensorflow, Pytorch, or MXNet.
The machine learning algorithm 130 is trained in step S220 using the training data set and a plurality of sets of data for the measured three-dimensional drainage phase saturation profile 25m, the derived one-dimensional drainage phase saturation profile 20d, the measured one-dimensional imbibition phase saturation profile 30m, and the predicted three-dimensional imbibition phase saturation profile 35p which have been classified and form the so-called “ground truth” for the training of the machine learning algorithm 130.
The training of the machine learning algorithm 130 is done by applying the method 5 to a synthetic heterogenous core plug 50 having multiple rock types. The synthetic heterogenous core plug 50 is an exemplary virtual example resembling a physical/real core plug 50. The permeability distribution is determined as the petrophysical properties of the synthetic heterogenous core plug 50, as is described above in step S100. An exemplary result of the determined permeability distribution of the synthetic heterogeneous core plug 50 are shown in
The three-dimensional drainage phase saturation profile 25m for the synthetic heterogenous core plug 50 is determined (as is described above in S130). The determining is done using, for example, ten different fraction flow rates for and calculating three-dimensional drainage phase saturation profile 25m for each of these different fraction flow rates. The one-dimensional drainage phase saturation profile 25d is calculated using the mathematical average (as is described above in Step S140) for each of the ten determined three-dimensional drainage phase saturation profiles 25m.
The steps of performing the ageing operation (see step S150), measuring the oil-water contact angle 60 in the porous medium 70 (see step S160), calculating the plurality of the synthetic values for the relative permeability 80 and the capillary pressure 90 of the synthetic heterogeneous core plug 50 (see step S170) are determined. The measured three-dimensional drainage phase saturation profile 25m, the derived one-dimensional drainage phase saturation profile 20d, the measured one-dimensional imbibition phase saturation profile 30m, the measured oil-water contact angle 60, the calculated synthetic values for the relative permeability 80, the calculated synthetic values for the capillary pressure 90, and the rock heterogeneity state 40 are then input in the porous media fluid flow simulator 120 as inputs. Using these described inputs, the simulated three-dimensional drainage phase saturation profile 25s and the simulated three-dimensional imbibition phase saturation profile 35s are simulated for the different fraction flow rates using the porous media fluid flow simulator 120 as described in step S200 (see above).
The simulated one-dimensional drainage phase saturation profiles 20s and the simulated one-dimensional imbibition phase saturation profiles 30s are calculated from the simulated three-dimensional drainage phase saturation profile 25s and the simulated three-dimensional imbibition phase saturation profile 35s for the different fraction flow rates using the mathematical average (as is described in step S210 above). The machine learning algorithm 130 is trained using the measured three-dimensional drainage phase saturation profile 25m, the derived one-dimensional drainage phase saturation profile 20d, the measured one-dimensional imbibition phase saturation profile 30m, the simulated three-dimensional drainage phase saturation profile 25s, and the simulated three-dimensional imbibition phase saturation profile 35s.
The training of the machine learning algorithm 130 comprises enabling the machine learning algorithm 130 to capture the relationship between the oil distribution during drainage and imbibition of the synthetic heterogeneous core plug 50 (see also step S220 above). During a training phase, an available data set is divided into two subsets, in which one subset is used named as a primary subset and a second subset is named as a secondary subset. The machine learning algorithm 130 provides machine learning results for an analysis request by processing available data set. The results from the machine learning algorithm 130 includes a confidence metric and a prediction made from the input data. For example, the machine learning algorithm provides phase saturation and the confidence metric for each instance of the workload data input into the machine learning algorithm.
Primary data called a “input data” or “ground truth” and secondary data called “output data” is used to train the machine learning algorithm 130. The training of the machine learning algorithm 130 comprises finding and quantifying a relationship between this input data and this output data. The training of the machine learning algorithm 130 therefore comprises comparing the secondary data output by the machine learning algorithm 130 to the primary data for the workload data. If the machine learning algorithm 130 outputs a secondary data being approximately similar to the primary data of the workload data, the machine learning algorithm 130 is rewarded. If the machine learning algorithm 130 outputs a secondary data being unsimilar to the primary data of the workload data, the machine learning algorithm 130 is penalized. Rewarding and penalizing the machine learning algorithm 130 is done for updating of parameters of a model of the machine learning algorithm 130 using a mathematical function such as a “loss function”.
The loss function is a mathematical function for evaluating a degree of similarity between the secondary data output by the machine learning algorithm 130 and the primary data. The loss function returns a numerical value (usually a real number) indicating the degree of similarity. If, for example, there is a high degree of similarity between the primary data and the output secondary data, the loss function returns a small numerical value. Similar, if, for example, there is a low degree of similarity between the primary data and the output secondary data, the loss function returns a high numerical value. The loss function therefore indicates how close the secondary data output by the machine learning algorithm 130 is to the primary data. The numerical value of the loss function determines how much the machine learning algorithm 130 should be penalized or rewarded for a current set of model parameters used in the machine learning algorithm 130 in determining the secondary data.
An efficiency of the machine learning algorithm 130 is quantified using the confidence metric, which comprises a percentage, a ratio, a rating on a scale, or another indicator of accuracy, effectiveness, and/or confidence. Examples of the machine learning algorithm 130 include, but are not limited to an artificial neural network, a XGBoost-algorithm, a recurrent neural networks or a combination of these. The trained machine learning algorithm 130 is validated using blind test data generated from different heterogeneous core plugs 50 with a ten percent (10%) error as the ground truth, as can be seen in
| Number | Date | Country | Kind |
|---|---|---|---|
| 21182332.3 | Jun 2021 | EP | regional |
