CN117892622A - Rock sample relative permeability curve prediction method and system based on mercury-pressing test data - Google Patents
Rock sample relative permeability curve prediction method and system based on mercury-pressing test data Download PDFInfo
- Publication number
- CN117892622A CN117892622A CN202410054990.0A CN202410054990A CN117892622A CN 117892622 A CN117892622 A CN 117892622A CN 202410054990 A CN202410054990 A CN 202410054990A CN 117892622 A CN117892622 A CN 117892622A
- Authority
- CN
- China
- Prior art keywords
- curve
- sample
- mercury
- data
- test data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000035699 permeability Effects 0.000 title claims abstract description 72
- 238000012360 testing method Methods 0.000 title claims abstract description 57
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000003825 pressing Methods 0.000 title claims abstract description 37
- 239000011435 rock Substances 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 claims abstract description 30
- 238000002474 experimental method Methods 0.000 claims abstract description 17
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- QSHDDOUJBYECFT-UHFFFAOYSA-N mercury Chemical compound [Hg] QSHDDOUJBYECFT-UHFFFAOYSA-N 0.000 claims description 56
- 229910052753 mercury Inorganic materials 0.000 claims description 56
- 239000011148 porous material Substances 0.000 claims description 27
- 239000011159 matrix material Substances 0.000 claims description 25
- 238000009826 distribution Methods 0.000 claims description 22
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 12
- 238000003062 neural network model Methods 0.000 claims description 11
- 239000013598 vector Substances 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 10
- 238000006243 chemical reaction Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 230000015654 memory Effects 0.000 claims description 7
- 230000001419 dependent effect Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 9
- 238000011161 development Methods 0.000 abstract description 8
- 230000008901 benefit Effects 0.000 abstract description 5
- 125000004122 cyclic group Chemical group 0.000 abstract description 4
- 238000011156 evaluation Methods 0.000 abstract description 4
- 238000011160 research Methods 0.000 abstract description 4
- 238000004088 simulation Methods 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 11
- 238000002347 injection Methods 0.000 description 7
- 239000007924 injection Substances 0.000 description 7
- 238000005070 sampling Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000000418 atomic force spectrum Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a rock sample relative permeability curve prediction method and a system based on mercury-pressing test data, which relate to the field of oilfield development and analysis and have the technical scheme that: the method comprises the steps of automatically generating a large number of training sample sets through a self-supervision learning mode according to limited mercury-pressing experiment-phase-seepage experiment sample data, inputting the training sample sets into a ConvLSTM model based on a self-supervision learning framework by means of a convolutional neural network and a cyclic neural network, and constructing a relative permeability curve calculation proxy model with high generalization capability, so that data support is provided for oil reservoir development evaluation and oil reservoir fine numerical simulation research, and important social and economic benefits are achieved.
Description
Technical Field
The invention relates to the field of oilfield development and analysis, in particular to a method and a system for predicting a rock sample relative permeability curve based on mercury intrusion test data.
Background
Relative permeability refers to the relative relationship between the permeability of different phases (e.g., oil and water) in a porous medium, which determines the rate of seepage of the different phases in the reservoir, and has an important impact on the development of the field. The relative permeability curves are obtained, typically directly and indirectly. The direct method is to measure by adopting a steady state or unsteady state method through an indoor core experiment. Limited by the number of core samples, coring contamination, measurement errors, etc., laboratory direct measurement methods are not always feasible, and therefore some indirect predictive models are widely used.
At present, partial scholars propose methods for calculating a relative permeability curve by adopting dynamic data, such as a water flooding curve method and a water content curve method, which can better reflect the overall dynamic change of an oil reservoir, but the method is based on long-time production data and cannot consider the heterogeneous condition of the oil reservoir. The mercury injection test is an experiment in which mercury is injected into microscopic pores of a porous medium under a certain pressure to obtain the relationship between the pressure and the mercury volume. Both its and relative permeability curves are governed by the complex microscopic pore structure characteristics of the porous media. Because mercury-pressing experimental data are easy to obtain, the sample size is relatively large, numerous scholars propose calculation relative permeability models based on capillary pressure experiments, including Purcell models and Burdine models, and the models are based on the assumption of capillary bundle and plate theory, so that calculation result errors are large. With the rapid development of artificial intelligence technology, the optimization and machine learning methods provide new solutions for the calculation of the relative permeability of subsurface reservoirs. According to the existing experiment or numerical result, a data set of an artificial intelligence algorithm is established, and the work has been greatly successful in reliability and efficiency of predicting the relative permeability, in particular to predicting the relative permeability by adopting a three-dimensional digital core image. However, there are few methods currently employing deep learning to predict the relative permeability curve directly from the capillary force curve of a rock sample.
Therefore, how to study and design a rock sample relative permeability curve prediction method and system based on mercury intrusion test data, which can overcome the defects, is a problem which needs to be solved in the current state.
Disclosure of Invention
In order to solve the defects in the prior art, the invention aims to provide a method and a system for predicting a relative permeability curve of a rock sample based on mercury pressing test data, which automatically generate a large number of training sample sets through a self-supervision learning mode by using limited mercury pressing experiment-phase permeability experiment sample data, input the training sample sets into an established ConvLSTM model based on a self-supervision learning framework by means of a convolutional neural network and a cyclic neural network, and construct a relative permeability curve calculation proxy model with high generalization capability, thereby providing data support for oil deposit development evaluation and oil deposit fine numerical simulation research and having very important social and economic benefits.
The technical aim of the invention is realized by the following technical scheme:
In a first aspect, a method for predicting a rock sample relative permeability curve based on mercury intrusion test data is provided, comprising the steps of:
obtaining mercury-pressing experimental test data of a plurality of core samples, wherein the mercury-pressing experimental test data comprise a mercury inlet curve, a curve and a pore-throat distribution curve;
Respectively carrying out interpolation processing on the data of the same curve in the mercury-pressing experiment test data, and converting the data into corresponding single-channel image data through a gram angle field;
combining the three single-channel image data as R, G, B channels of the RGB image respectively to obtain corresponding three-channel image data;
based on a Brooks-Corey model, fitting experimental test data of the phase permeation curves corresponding to each core sample by adopting a Gauss-Newton algorithm to obtain corresponding phase permeation curve parameters;
establishing a sequence sample representing the difference and evolution relation between the image samples based on the differential matrix between the three channel image data to obtain training sample data;
Initializing a neural network, inputting training sample data into the neural network for training until a loss function converges to obtain a neural network model with parameters of a mercury-pressure curve-mapped phase permeability curve;
and inputting the mercury-pressing experimental test data of the unknown sample and the known samples into a neural network model, and predicting to obtain the parameters of the phase permeability curve of the unknown sample.
Further, the mercury inlet curve, curve and pore throat distribution curve are all expressed by ordered binary vectors with finite lengths;
The ordered binary vectors of the mercury inlet curve and the curve comprise mercury saturation and pressure values;
the ordered binary vector of the pore throat distribution curve comprises the logarithm of pore throat radius and the corresponding pore throat volume percentage.
Further, the process of converting the glamer angle field into the corresponding single-channel image data specifically includes:
Linearly converting the interpolation processed sequence data X to [ -1,1] to obtain a sequence
Mapping sequence to polar coordinates, the sequence/>, is represented within a unit circle
Traversing l i[0,128],lj e [0,128], calculating the triangular cosine sum of any (l i,lj):
Wherein represents the elements numbered l and l ilj in/> respectively; max (X) and min (X) respectively represent the maximum value and the minimum value of the sequence ; the/> represents the angular value of/> in polar coordinates; and/> denotes an element in the GASF matrix with a row number, a column number of l ilj. Further, the Brooks-Corey model parameters are:
The Brooks-Corey equation expression of the Brooks-Corey model is as follows:
Wherein represents Brooks-Corey model parameters corresponding to the ith core sample; k ro,i is the oil relative permeability of the ith sample, a dependent variable; s o,i is the oil saturation of the ith sample and the independent variable; s w,i is the water saturation, argument of the ith sample; s or,i is the residual oil saturation of the ith sample, and the range of the residual oil saturation is in the closed range [0,1] of the parameters to be solved; s wc,i is the irreducible water saturation of the ith sample, and the parameters to be solved are in the range of a closed interval [0,1 ]; n o,i is the oil phase index term of the ith sample, and the range is in the closed interval [0,6] of the parameters to be solved; n w,i is the oil phase index term of the ith sample, and the range is in the closed interval [0,6] of the parameters to be solved; the value of the water phase maximum relative permeability of the ith sample is expressed as '', and the range of the water phase maximum relative permeability of the ith sample is in a closed range [0,1 ]; k rw,i represents the water relative permeability equation, dependent variable, for the ith sample.
Further, the neural network adopts a conv2d+LSTM2d+FC multilayer architecture.
Further, the conv2d is a 2-dimensional convolutional neural network: inputting image data with a single channel and 128 length and width, and outputting tensors with dimensions (16,16,64) through a multi-layer convolutional neural network to represent image characteristics;
The LSTM2d is a long-short-time neural network that can process two-dimensional image data: taking output tensors of conv2d at 8 moments as input, taking the last 4 moment dimensions as hidden layer tensors through a time sequence network LSTM with the length of 8, and outputting tensors with the dimension of (4,16,16,8), wherein the 1 st dimension is the moment;
The FC refers to a fully connected layer, and maps tensors of last 4 moments of output of LSTM2d into vectors with length of 4, namely permeability of last 4 samples, through a linear layer.
Further, the unknown sample is placed at the last moment of all sample data.
In a second aspect, there is provided a rock sample relative permeability curve prediction system based on mercury intrusion test data, comprising:
The data acquisition module is used for acquiring mercury-pressing experimental test data of a plurality of core samples, wherein the mercury-pressing experimental test data comprise a mercury inlet curve, a curve and a pore throat distribution curve;
The image conversion module is used for respectively carrying out interpolation processing on the data of the same curve in the mercury-pressing experiment test data and converting the data into corresponding single-channel image data through a gram angle field;
The image combination module is used for respectively combining the three single-channel image data as R, G, B channels of the RGB image to obtain corresponding three-channel image data;
The parameter fitting module is used for fitting the experimental test data of the phase permeation curves corresponding to the core samples by using a Gauss-Newton algorithm based on a Brooks-Corey model to obtain corresponding phase permeation curve parameters;
the sample generation module is used for establishing a sequence sample for representing the difference and evolution relation between the image samples based on the differential matrix among the three channel image data to obtain training sample data;
The model training module is used for initializing a neural network, inputting training sample data into the neural network for training until the loss function converges to obtain a neural network model with parameters of the mercury-pressure curve-mapped permeability curve;
and the parameter prediction module is used for inputting the mercury-pressing experimental test data of the unknown sample and the known samples into the neural network model, and predicting to obtain the parameters of the permeability curve of the unknown sample.
In a third aspect, a computer terminal is provided, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the method for predicting a rock sample relative permeability curve based on mercury intrusion test data according to any one of the first aspects when executing the program.
In a fourth aspect, a computer readable medium is provided, having stored thereon a computer program for execution by a processor to implement a method for predicting a rock sample relative permeability curve based on mercury intrusion test data according to any one of the first aspects.
Compared with the prior art, the invention has the following beneficial effects:
1. According to the rock sample relative permeability curve prediction method based on mercury pressing test data, a large number of training sample sets are automatically generated through a self-supervision learning mode according to limited mercury pressing experiment-phase permeability experiment sample data, and are input into a ConvLSTM model based on a self-supervision learning framework by means of a convolutional neural network and a cyclic neural network, so that a relative permeability curve calculation proxy model with high generalization capability is constructed, data support is provided for oil reservoir development evaluation and oil reservoir fine numerical simulation research, and very important social and economic benefits are achieved;
2. According to the invention, three curves (mercury inlet, mercury injection and pore throat distribution) of mercury injection data are uniquely and reversibly represented by using a gram angle field and an RGB image, and the interrelationship of different positions on an original curve can be better represented on the image field;
3. The invention establishes a differential conversion matrix between RGB image representations of different mercury-pressing data, the matrix can enable RGB images of appointed mercury-pressing data to be converted into RGB images of mercury-pressing data through a linear matrix, the matrix is used for representing differences and interrelationships among sheets, and the representation mode is unique and reversible;
4. According to the invention, the image conversion matrix is introduced into a sequence, the dependence and evolution relation between sample images are explicitly expressed, the initial images can be sequentially converted into other images according to the conversion matrix through the sequence, and the difference and the relation between the neural network images are directly guided at the sample level, so that the change of Brooks-Corey phase permeation model parameters caused by the difference and the evolution relation between RGB images of mercury-pressing data is focused;
5. According to the method, the fact that the data samples obtained through the mercury-pressing experiment and the phase-penetration experiment are extremely limited is considered, the difference and the connection between the images are represented, then an almost infinite training sample is formed through random sampling, the relation between the samples is fully excavated, and therefore the over-fitting phenomenon of the neural network is effectively reduced, and the generalization capability is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:
1 FIG. 1 is a graph showing the mercury intrusion curve, curve, pore throat distribution curve and interpolated curves according to example 1 of the present invention;
2 FIG. 2 is a schematic diagram of a single channel image and a combined multi-channel RGB image corresponding to a mercury intrusion curve, curve, pore throat distribution curve in example 1 of the present invention;
FIG. 3 is a schematic representation of the experimental data for the phase permeation in example 1 of the present invention, and a Brooks-Corey model fit;
FIG. 4 is a schematic diagram of an evolution sequence sample in embodiment 1 of the present invention;
FIG. 5 is a schematic diagram of the neural network structure and data flow in embodiment 1 of the present invention;
FIG. 6 is an iterative schematic of a neural network loss function in embodiment 1 of the present invention;
FIG. 7 is a schematic diagram of parameters of experimental data fitting parameters of the experimental sample predicted phase permeability model according to example 1 of the present invention;
FIG. 8 is a graph showing the prediction of the sample permeability curve of the test sample according to example 1 of the present invention;
Fig. 9 is a system block diagram in embodiment 1 of the present invention.
Detailed Description
For the purpose of making apparent the objects, technical solutions and advantages of the present invention, the present invention will be further described in detail with reference to the following examples and the accompanying drawings, wherein the exemplary embodiments of the present invention and the descriptions thereof are for illustrating the present invention only and are not to be construed as limiting the present invention.
Example 1: the method for predicting the rock sample relative permeability curve based on mercury intrusion test data is realized by the following steps.
A. Mercury intrusion data were prepared. Given that N core samples are tested by mercury-pressing experiments, namely a mercury-feeding curve curve/> pore throat distribution curve/> any/> are ordered binary vectors with finite length, namely/>
Wherein i and k respectively represent the core sample number and the data point number; Respectively representing a mercury inlet curve, a curve and a pore throat distribution curve of an ith sample; the/> represents the length of the mercury intrusion curve, the length of the curve and the length of the pore throat distribution curve of the ith sample respectively; the/> represents the kth data point of the ith sample, the mercury saturation of/> pressure value of/> represents the kth mercury entry curve data point of the ith sample, the mercury saturation of/> pressure value of/> represents the kth curve data point of the ith sample, the mercury saturation of/> pressure value of/> (ri,k,fi,k) represents the kth pore throat distribution data point of the ith sample, the logarithm of the pore throat radius (based on 10) of r i,k, and the corresponding pore throat volume percentage of f i,k.
B. And carrying out interpolation processing on the mercury-pressing data. For any two different i samples and j samples (i+.j), and and/> and/> may be different, and the abscissa distance between any two adjacent data points may not be uniform, so all mercury-in curves, curves, pore throat distribution curves need to be interpolated and extrapolated by a linear interpolation algorithm, processed to be uniform in length 128 (actually the number of long and wide pixels of the image), and the abscissa between any adjacent data points is equidistant. After conversion, let the following apply to any i: and/> wherein: interpolate the mercury intrusion curve to 128 equidistant data points on [0,100 ]; interpolate the curve to 128 equidistant data points on [0,100 ]; the pore throat distribution curve is interpolated to 128 equidistant data points on [ -3,3] as shown in fig. 1.
C. For all samples, the mercury intrusion curve, curve and pore throat distribution curve are all converted into two-dimensional matrix (single-channel image) data through a gram angle field. Note that after the step B, all kinds of data are sequence data with length 128, and for any piece of sequence data, the sequence data is abbreviated as x= { X 1,x2,,x300 }, and the specific steps of conversion are as follows:
(1) The sequence X is linearly converted to [ -1,1] by this formula to obtain
(2) Mapping the sequence to polar coordinates, the sequence is represented within a unit circle as follows:
(3) Traversing i epsilon [0,128], j epsilon [0,128], and calculating the triangular cosine sum of any (i, j):
Wherein represents the elements numbered l and l ilj in/> respectively; max (X) and min (X) respectively represent the maximum value and the minimum value of the sequence/> ; the/> represents the angular value of/> in polar coordinates; and/> denotes an element in the GASF matrix with a row number, a column number of l ilj.
D. Traversing i epsilon [0, N ], and converting any i mercury feeding curve, curve and pore throat distribution curve through the step C to obtain mercury feeding data matrix, data matrix and pore throat distribution data matrix, wherein the combination of the mercury feeding data matrix, the data matrix and the pore throat distribution data matrix is expressed as image data: the 3-dimensional matrix of size (1281283), i.e., three-channel image data, is obtained by sequentially arranging/> in order, respectively , as shown in fig. 2.
E. The experimental test data of the corresponding phase permeation curves of the N core samples in the step A are fitted with the corresponding phase permeation curve parameters based on a Brooks-Corey model by using a Gauss-Newton algorithm, and are shown in figure 3. The Brooks-Corey model parameters corresponding to the ith core sample are , and the corresponding pair Brooks-Corey equation expression is as follows:
Wherein k ro,i is the oil relative permeability of the ith sample, a dependent variable; s o,i is the oil saturation of the ith sample and the independent variable; s w,i is the water saturation, argument of the ith sample; s or,i is the residual oil saturation of the ith sample, and the range of the residual oil saturation is in the closed range [0,1] of the parameters to be solved; s wc,i is the irreducible water saturation of the ith sample, and the parameters to be solved are in the range of a closed interval [0,1 ]; n o,i is the oil phase index term of the ith sample, and the range is in the closed interval [0,6] of the parameters to be solved; n w,i is the oil phase index term of the ith sample, and the range is in the closed interval [0,6] of the parameters to be solved; The water phase maximum relative permeability of the ith sample is the parameter to be solved, wherein the parameter is in a closed interval [0,1 ]; k rw,i represents the water relative permeability equation, dependent variable, for the ith sample.
F. And constructing neural network training sample data, and constructing sequence samples for representing the difference and evolution relation between the image samples based on a differential matrix between RGB images. For any th training sample, the specific steps for generating are as follows:
(1) Random sampling is performed for 8 times from N images without repetition, and the 8 images are arranged into images according to the sampling sequence, wherein N 1,n2,,n8 is a positive integer on [1, N ], and no repetition exists.
(2) The 7 differential matrices to/> to/> ,/> to/> are calculated, taking/> to/> as an example, the linear differential matrices/> are:
Where I 1-I2 is the element-wise difference result of the RGB image (three-dimensional matrix).
(3) Arranging the data obtained in the step (2) to obtain data of an mth neural network sample, namely tensor with the size of (81281283); the output data is a matrix of size (4 x 5) consisting of Brooks-Corey model parameter vectors with a label of , i.e., the length of 5 for the last 4 core samples.
And screening out a sample pair (I, Y) with a 3-channel image obtained by the mercury injection test as I and a Corey model fitting parameter as Y by the corresponding mercury injection test and the corresponding permeability test according to the rock sample well number, depth, porosity and permeability parameters. According to step F, perm (8, 40) = 2193360 samples were constructed, the sample size reached two million levels, sufficient to train the deep neural network, and the sample construction schematic is shown in fig. 4.
G. and initializing a neural network, and adopting a conv2d+LSTM2d+FC multi-layer architecture. Wherein conv2d refers to a 2-dimensional convolutional neural network (Convolution Neutral Network); LSTM2d is a long and short time neural network (Long Short Time Memories Network) which can process two-dimensional image data after modification; FC refers to the fully connected layer (Fully Connected Layer), and the details and functions of the network structure are shown in fig. 5. The input data is a time-series image tensor of (B, T, W, H, C), wherein B is the batch size, and is set to 16; t is the time sequence length and is set to 8; w is the width of the input image, set to 128; h is the height of the input image, set to 128; c is the channel of the input image, set to 1. Note that the initial hidden layer of LSTM is set to a tensor with all elements being 1. The architecture of the neural network is divided into three sub-modules conv2d, LSTM2d and FC, and the functions and the input and output of the neural network are as follows according to the data flow direction:
(1) conv2d: inputting image data with a single channel and 128 length and width, and outputting tensors with dimensions (16,16,64) through a multi-layer convolutional neural network to represent image characteristics;
(2) LSTM2d: taking output tensors of conv2d at 8 moments as input, taking the last 4 moment dimensions as hidden layer tensors through a time sequence network LSTM with the length of 8, and outputting tensors with the dimension of (4,16,16,8), wherein the 1 st dimension is the moment;
(3) FC: the tensor of the last 4 moments of the output of LSTM2d is mapped by the linear layer to a vector of length 4, i.e. the permeability of the last 4 samples. The step F is executed 16 times, 16 training sample data are obtained, and the training sample data are combined on the 1 st dimension to form a batch with the size of 16.
H. since the step F uses random sampling, the possible training sample size is Perm (8, N), i.e. the number of rows of 8 ordered samples in N. The 16 training sample data thus guarantee no repetition with a very high probability. The probability of repetition between samples is: When n=50, the repetition probability is lower than 10 -13.
I. The training sample batch is input into the neural network for training, as shown in figure 5,
J. The H, I steps are repeated until the loss function converges, resulting in a final proxy model of the parameters of the mercury intrusion curve mapped to the phase intrusion curve, as shown in fig. 6.
K. and F, for the mercury injection curve of the unknown sample, executing the step F together with mercury injection curve data of 7 other known samples, taking attention to the fact that the unknown sample is placed at the last moment of sample data, and inputting the neural network model established in the step K, so that the corresponding parameters of the permeability curve can be predicted.
After training convergence, according to the step K, relative permeability of the test sample is predicted. The relative permeability calculation proxy model was examined for calculation on the test set and the predicted profile of parameters of the permeability model for the test sample is shown in figure 7. Fig. 8 is a graph comparing the predicted relative permeability curve of the test sample with the relative permeability curve of the actual experiment, and the prediction accuracy is better.
Example 2: the rock sample relative permeability curve prediction system based on mercury intrusion test data comprises a data acquisition module, an image conversion module, an image combination module, a parameter fitting module, a sample generation module, a model training module and a parameter prediction module as shown in fig. 9.
The data acquisition module is used for acquiring mercury-pressing experimental test data of a plurality of core samples, wherein the mercury-pressing experimental test data comprise a mercury inlet curve, a curve and a pore-throat distribution curve; the image conversion module is used for respectively carrying out interpolation processing on the data of the same curve in the mercury-pressing experiment test data and converting the data into corresponding single-channel image data through a gram angle field; the image combination module is used for respectively combining the three single-channel image data as R, G, B channels of the RGB image to obtain corresponding three-channel image data; the parameter fitting module is used for fitting the experimental test data of the phase permeation curves corresponding to the core samples by using a Gauss-Newton algorithm based on a Brooks-Corey model to obtain corresponding phase permeation curve parameters; the sample generation module is used for establishing a sequence sample for representing the difference and evolution relation between the image samples based on the differential matrix among the three channel image data to obtain training sample data; the model training module is used for initializing a neural network, inputting training sample data into the neural network for training until the loss function converges to obtain a neural network model with parameters of the mercury-pressure curve-mapped permeability curve; and the parameter prediction module is used for inputting the mercury-pressing experimental test data of the unknown sample and the known samples into the neural network model, and predicting to obtain the parameters of the permeability curve of the unknown sample.
Working principle: according to the invention, a large number of training sample sets are automatically generated through a self-supervision learning mode by using a limited mercury-pressing experiment-phase permeation experiment sample data, and are input into a ConvLSTM model based on a self-supervision learning framework by means of a convolutional neural network and a cyclic neural network, so that a relative permeability curve calculation proxy model with high generalization capability is constructed, and data support is provided for oil reservoir development evaluation and oil reservoir fine numerical simulation research, so that the method has very important social and economic benefits.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing detailed description of the invention has been presented for purposes of illustration and description, and it should be understood that the invention is not limited to the particular embodiments disclosed, but is intended to cover all modifications, equivalents, alternatives, and improvements within the spirit and principles of the invention.
Claims (10)
1. The rock sample relative permeability curve prediction method based on mercury intrusion test data is characterized by comprising the following steps of:
obtaining mercury-pressing experimental test data of a plurality of core samples, wherein the mercury-pressing experimental test data comprise a mercury inlet curve, a curve and a pore-throat distribution curve;
Respectively carrying out interpolation processing on the data of the same curve in the mercury-pressing experiment test data, and converting the data into corresponding single-channel image data through a gram angle field;
combining the three single-channel image data as R, G, B channels of the RGB image respectively to obtain corresponding three-channel image data;
based on a Brooks-Corey model, fitting experimental test data of the phase permeation curves corresponding to each core sample by adopting a Gauss-Newton algorithm to obtain corresponding phase permeation curve parameters;
establishing a sequence sample representing the difference and evolution relation between the image samples based on the differential matrix between the three channel image data to obtain training sample data;
Initializing a neural network, inputting training sample data into the neural network for training until a loss function converges to obtain a neural network model with parameters of a mercury-pressure curve-mapped phase permeability curve;
and inputting the mercury-pressing experimental test data of the unknown sample and the known samples into a neural network model, and predicting to obtain the parameters of the phase permeability curve of the unknown sample.
2. The method for predicting a relative permeability curve of a rock sample based on mercury intrusion test data according to claim 1, wherein the mercury intrusion curve, curve and pore throat distribution curve are each represented by ordered binary vectors of finite length;
The ordered binary vectors of the mercury inlet curve and the curve comprise mercury saturation and pressure values;
the ordered binary vector of the pore throat distribution curve comprises the logarithm of pore throat radius and the corresponding pore throat volume percentage.
3. The method for predicting a rock sample relative permeability curve based on mercury intrusion test data according to claim 1, wherein the process of converting the angular field into corresponding single-channel image data by using a gladhand is specifically as follows:
Linearly converting the interpolation processed sequence data X to [ -1,1] to obtain a sequence
Mapping sequence to polar coordinates, the sequence/>, is represented within a unit circle
Traversing l i[0,128],lj e [0,128], calculating the triangular cosine sum of any (l i,lj):
Wherein represents the elements numbered l and l ilj in/> respectively; max (X) and min (X) respectively represent the maximum value and the minimum value of the sequence/> ; the/> represents the angular value of/> in polar coordinates; and/> denotes an element in the GASF matrix with a row number, a column number of l ilj.
4. The method for predicting a relative permeability curve of a rock sample based on mercury intrusion test data according to claim 1, wherein the Brooks-Corey model parameters are:
The Brooks-Corey equation expression of the Brooks-Corey model is as follows:
Wherein represents Brooks-Corey model parameters corresponding to the ith core sample; k ro,i is the oil relative permeability of the ith sample, a dependent variable; s o,i is the oil saturation of the ith sample and the independent variable; s w,i is the water saturation, argument of the ith sample; s or,i is the residual oil saturation of the ith sample, and the range of the residual oil saturation is in the closed range [0,1] of the parameters to be solved; s wc,i is the irreducible water saturation of the ith sample, and the parameters to be solved are in the range of a closed interval [0,1 ]; n o,i is the oil phase index term of the ith sample, and the range is in the closed interval [0,6] of the parameters to be solved; n w,i is the oil phase index term of the ith sample, and the range is in the closed interval [0,6] of the parameters to be solved; the value of the water phase maximum relative permeability of the ith sample is expressed as '', and the range of the water phase maximum relative permeability of the ith sample is in a closed range [0,1 ]; k rw,i represents the water relative permeability equation, dependent variable, for the ith sample.
5. The method for predicting a relative permeability curve of a rock sample based on mercury intrusion test data according to claim 1, wherein the neural network adopts a conv2d+lstm2d+fc multi-layer architecture.
6. The method for predicting a relative permeability curve of a rock sample based on mercury intrusion test data according to claim 1, wherein conv2d is a 2-dimensional convolutional neural network: inputting image data with a single channel and 128 length and width, and outputting tensors with dimensions (16,16,64) through a multi-layer convolutional neural network to represent image characteristics;
The LSTM2d is a long-short-time neural network that can process two-dimensional image data: taking output tensors of conv2d at 8 moments as input, taking the last 4 moment dimensions as hidden layer tensors through a time sequence network LSTM with the length of 8, and outputting tensors with the dimension of (4,16,16,8), wherein the 1 st dimension is the moment;
The FC refers to a fully connected layer, and maps tensors of last 4 moments of output of LSTM2d into vectors with length of 4, namely permeability of last 4 samples, through a linear layer.
7. The method of claim 1, wherein the unknown sample is placed at the last moment of all sample data.
8. Rock sample relative permeability curve prediction system based on mercury intrusion test data, characterized by comprising:
The data acquisition module is used for acquiring mercury-pressing experimental test data of a plurality of core samples, wherein the mercury-pressing experimental test data comprise a mercury inlet curve, a curve and a pore throat distribution curve;
The image conversion module is used for respectively carrying out interpolation processing on the data of the same curve in the mercury-pressing experiment test data and converting the data into corresponding single-channel image data through a gram angle field;
The image combination module is used for respectively combining the three single-channel image data as R, G, B channels of the RGB image to obtain corresponding three-channel image data;
The parameter fitting module is used for fitting the experimental test data of the phase permeation curves corresponding to the core samples by using a Gauss-Newton algorithm based on a Brooks-Corey model to obtain corresponding phase permeation curve parameters;
the sample generation module is used for establishing a sequence sample for representing the difference and evolution relation between the image samples based on the differential matrix among the three channel image data to obtain training sample data;
The model training module is used for initializing a neural network, inputting training sample data into the neural network for training until the loss function converges to obtain a neural network model with parameters of the mercury-pressure curve-mapped permeability curve;
and the parameter prediction module is used for inputting the mercury-pressing experimental test data of the unknown sample and the known samples into the neural network model, and predicting to obtain the parameters of the permeability curve of the unknown sample.
9. A computer terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for predicting a rock sample relative permeability curve based on mercury intrusion test data according to any one of claims 1 to 7 when executing the program.
10. A computer readable medium having stored thereon a computer program, wherein the computer program is executable by a processor to implement a method of rock sample relative permeability curve prediction based on mercury intrusion test data according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410054990.0A CN117892622A (en) | 2024-01-15 | 2024-01-15 | Rock sample relative permeability curve prediction method and system based on mercury-pressing test data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410054990.0A CN117892622A (en) | 2024-01-15 | 2024-01-15 | Rock sample relative permeability curve prediction method and system based on mercury-pressing test data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117892622A true CN117892622A (en) | 2024-04-16 |
Family
ID=90646383
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410054990.0A Pending CN117892622A (en) | 2024-01-15 | 2024-01-15 | Rock sample relative permeability curve prediction method and system based on mercury-pressing test data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117892622A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112612997A (en) * | 2020-12-29 | 2021-04-06 | 中国石油大学(华东) | Oil-water relative permeability curve calculation method based on deep learning |
| US11513088B1 (en) * | 2021-11-05 | 2022-11-29 | Institute Of Geology And Geophysics, Chinese Academy Of Sciences | Method for comprehensively characterizing lower limit of oil accumulation of deep marine carbonate reservoir |
| CN115935834A (en) * | 2023-02-23 | 2023-04-07 | 中国石油大学(华东) | History fitting method based on deep autoregressive network and continuous learning strategy |
| CN116310399A (en) * | 2023-03-22 | 2023-06-23 | 中南大学 | AE-CNN-based high-dimensional feature map target identification method and system |
-
2024
- 2024-01-15 CN CN202410054990.0A patent/CN117892622A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112612997A (en) * | 2020-12-29 | 2021-04-06 | 中国石油大学(华东) | Oil-water relative permeability curve calculation method based on deep learning |
| US11513088B1 (en) * | 2021-11-05 | 2022-11-29 | Institute Of Geology And Geophysics, Chinese Academy Of Sciences | Method for comprehensively characterizing lower limit of oil accumulation of deep marine carbonate reservoir |
| CN115935834A (en) * | 2023-02-23 | 2023-04-07 | 中国石油大学(华东) | History fitting method based on deep autoregressive network and continuous learning strategy |
| CN116310399A (en) * | 2023-03-22 | 2023-06-23 | 中南大学 | AE-CNN-based high-dimensional feature map target identification method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112597985B (en) | Crowd counting method based on multi-scale feature fusion | |
| CN110580727B (en) | Depth V-shaped dense network imaging method with increased information flow and gradient flow | |
| CN114693005B (en) | Three-dimensional underground oil reservoir dynamic prediction method based on convolution Fourier neural network | |
| CN115345238B (en) | Method and device for generating seawater transparency fusion data | |
| Hajizadeh et al. | An algorithm for 3D pore space reconstruction from a 2D image using sequential simulation and gradual deformation with the probability perturbation sampler | |
| CN110826764B (en) | Random forest algorithm-based vegetation index prediction method, system and equipment | |
| CN113743301B (en) | Solid-state nanopore sequencing electric signal noise reduction processing method based on residual self-encoder convolutional neural network | |
| CN116067842A (en) | Method, device and storage medium for monitoring wettability of rock core | |
| CN113435662B (en) | Water-drive reservoir yield prediction method and device and storage medium | |
| CN110671092A (en) | Oil gas productivity detection method and system | |
| Regazzoni et al. | A physics-informed multi-fidelity approach for the estimation of differential equations parameters in low-data or large-noise regimes | |
| CN117892622A (en) | Rock sample relative permeability curve prediction method and system based on mercury-pressing test data | |
| Zhang et al. | Efficient surrogate modeling based on improved vision transformer neural network for history matching | |
| CN112949944A (en) | Underground water level intelligent prediction method and system based on space-time characteristics | |
| CN113222209A (en) | Regional tail gas migration prediction method and system based on domain adaptation and storage medium | |
| Zhang et al. | 3D stochastic reconstruction of porous media based on attention mechanisms and residual networks | |
| CN105787507B (en) | LS SVMs on-line study methods based on budget supporting vector collection | |
| CN116859478A (en) | Groundwater simulation method and system based on transient electromagnetic imaging | |
| CN110782023A (en) | Reduction residual module porous convolution architecture network and rapid semantic segmentation method | |
| CN114612751B (en) | Whole machine point cloud data down-sampling method based on semantic learning | |
| CN113869350B (en) | Ocean current prediction method and system based on space feature difference | |
| CN115688583A (en) | Seepage simulation method for rock core, processor and machine-readable storage medium | |
| CN115540832A (en) | Satellite altimetry submarine topography correction method and system based on VGGNet | |
| CN116266377A (en) | Three-dimensional reconstruction method for microstructure of core image based on LSTM | |
| Yao et al. | Hybrid algorithm for parameter estimation of the groundwater flow model with an improved genetic algorithm and gauss-newton method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |