CN118313265A - Oil reservoir history fitting method, system, equipment and medium based on error learning double-agent model - Google Patents
Oil reservoir history fitting method, system, equipment and medium based on error learning double-agent model Download PDFInfo
- Publication number
- CN118313265A CN118313265A CN202410528088.8A CN202410528088A CN118313265A CN 118313265 A CN118313265 A CN 118313265A CN 202410528088 A CN202410528088 A CN 202410528088A CN 118313265 A CN118313265 A CN 118313265A
- Authority
- CN
- China
- Prior art keywords
- model
- oil reservoir
- agent
- dual
- learning
- 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
- 238000000034 method Methods 0.000 title claims abstract description 58
- 239000002245 particle Substances 0.000 claims abstract description 90
- 230000006870 function Effects 0.000 claims abstract description 47
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 28
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000004088 simulation Methods 0.000 claims abstract description 17
- 230000009467 reduction Effects 0.000 claims abstract description 11
- 238000013135 deep learning Methods 0.000 claims abstract description 10
- 238000004519 manufacturing process Methods 0.000 claims description 52
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 16
- 230000009977 dual effect Effects 0.000 claims description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 10
- 230000015654 memory Effects 0.000 claims description 10
- 230000000306 recurrent effect Effects 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 230000007704 transition Effects 0.000 claims description 6
- 230000006403 short-term memory Effects 0.000 claims description 5
- 230000007787 long-term memory Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 abstract description 7
- 238000011161 development Methods 0.000 abstract description 3
- 239000003795 chemical substances by application Substances 0.000 description 92
- 230000008569 process Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 6
- 230000035699 permeability Effects 0.000 description 6
- 238000010276 construction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000000513 principal component analysis Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
- 235000020681 well water Nutrition 0.000 description 1
- 239000002349 well water Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of oil and gas field development, and discloses an oil reservoir history fitting method, system, equipment and medium based on an error learning double-agent model, which comprises the following steps: reducing the dimension of the oil reservoir model parameters; carrying out oil deposit numerical simulation on the oil deposit model parameters subjected to the dimension reduction to obtain learning sample data of the oil deposit model parameters; according to learning sample data of oil reservoir model parameters, a deep learning neural network is utilized to train to obtain a double-agent model of the oil reservoir model parameters, and then a particle swarm algorithm is combined to conduct overall optimization solution on a history fit objective function, so that prediction accuracy and efficiency of the double-agent model can be improved to a great extent, difficulty in setting super parameters is reduced, and reliable and efficient oil reservoir history fit is facilitated.
Description
Technical Field
The invention relates to the technical field of oil and gas field development, in particular to an oil reservoir history fitting method, system, equipment and medium based on an error learning double-agent model.
Background
The numerical simulation of the oil reservoir is an effective technical means, the characteristics of the oil reservoir can be deeply known through mathematical modeling and calculation solution, the productivity and the oil-water distribution condition of the oil reservoir are predicted, and more accurate and scientific decisions can be made in the development process of the oil field. For complex large-scale reservoirs, the forward process involved in reservoir numerical simulation is very time consuming, and data assimilation-based history fits typically require hundreds or thousands of forward simulations to update the state parameters of the system, resulting in very low computational efficiency. The use of proxy models instead of reservoir numerical simulation is one way to solve this problem. However, the difference of the model structure and the selection of the super parameters affect the prediction accuracy of the proxy model, and the prediction accuracy of a single proxy model is still limited for complex problems such as a production curve changing along with time and a saturated front evolving rapidly.
Therefore, a new method for fitting the oil reservoir history is needed to realize efficient and accurate prediction of the dynamic production data of the oil reservoir.
Disclosure of Invention
The invention provides an oil reservoir history fitting method, system, equipment and medium based on an error learning double-agent model, which are used for solving the defects of poor accuracy and low efficiency of predicting oil reservoir production data in the prior art.
The invention provides a method for constructing a double-agent model based on error learning, which comprises the following steps:
reducing the dimension of the oil reservoir model parameters to obtain the dimension-reduced oil reservoir model parameters;
Carrying out oil deposit numerical simulation on the oil deposit model parameters subjected to the dimension reduction to obtain learning sample data of the oil deposit model parameters, wherein the learning sample data comprises oil deposit parameter sample data and oil deposit production sample data;
According to learning sample data of oil reservoir model parameters, training to obtain a double-agent model of the oil reservoir model parameters by using a deep learning neural network, wherein the double-agent model comprises a first agent sub-model and a second agent sub-model, the first agent sub-model is used for learning the relation between the oil reservoir parameter sample data and oil reservoir production sample data, and the second agent sub-model is used for learning the prediction error of the first agent sub-model.
In one embodiment, the dimension reduction is performed on the oil reservoir model parameters to obtain the dimension reduced oil reservoir model parameters, which includes:
And (3) reducing the dimension of the oil reservoir model parameters by using a principal component analysis (PRINCIPAL COMPONENT ANALYSIS, PCA) to obtain the dimension-reduced oil reservoir model parameters.
In one embodiment, the performing the reservoir numerical simulation on the reservoir model parameters after the dimension reduction to obtain learning sample data of the reservoir model parameters includes:
generating a plurality of oil reservoir model parameter samples by using a first expression according to the oil reservoir model parameters after the dimension reduction;
And respectively carrying out oil reservoir numerical simulation on the plurality of oil reservoir model parameter samples to obtain learning sample data of oil reservoir model parameters.
In one embodiment, the first expression is:
In the first expression of the present invention, Representing the generated reservoir model parameter samples,Representing the mean of the reservoir model parameters for all prior models, U L represents the orthogonal matrix and ζ L represents random numbers conforming to a Gaussian distribution.
In one embodiment, the expression of the dual agent model is:
FDF＝F1+F2，
in the expression of the dual-proxy model, F Df denotes a model framework of the dual-proxy model, F 1 denotes a prediction result of the first proxy sub-model, and F 2 denotes a prediction result of the second proxy sub-model.
In one embodiment, the network structure of the first proxy sub-model includes a convolutional neural network (Convolutional Neural Networks, CNN) and a long-short term memory recurrent neural network (Long Short Term Memory, LSTM), the training is performed to obtain a dual-proxy model of reservoir model parameters according to learning sample data of the reservoir model parameters by using the deep learning neural network, and the method comprises the following steps:
Extracting spatial features of oil reservoir parameter sample data through a convolutional neural network and flattening the spatial features;
And obtaining time sequence information of oil reservoir production sample data through the long-term and short-term memory recurrent neural network, and learning a time sequence mode between the oil reservoir parameter sample data and the oil reservoir production sample data by combining the flattened spatial characteristics of the recurrent neural network.
In one embodiment, the second agent sub-model includes an encoder, a decoder, and a transition module, and the training is performed to obtain a dual agent model of the reservoir model parameters by using a deep learning neural network according to learning sample data of the reservoir model parameters, including:
Extracting oil reservoir parameter characteristics of oil reservoir parameter sample data through an encoder;
Mapping the oil reservoir parameter characteristics to a data space for learning the prediction error of the first proxy sub-model through a decoder;
and converting the prediction result of the second agent sub-model into an image form through a transition module.
The invention also provides an oil reservoir history fitting method based on the error learning double-agent model, which comprises the following steps:
Setting a history fitting objective function of the double-agent model obtained by the construction method of the double-agent model based on error learning;
Carrying out oil reservoir history fitting through the double-agent model, and solving a history fitting objective function of the double-agent model by utilizing a particle swarm algorithm so as to update oil reservoir model parameters;
predicting to obtain oil reservoir production data according to the updated oil reservoir model parameters by using the double-agent model;
And solving a history fit objective function of the double-agent model by using a particle swarm algorithm again according to the updated oil deposit model parameters and the predicted oil deposit production data, and predicting the oil deposit production data by using the double-agent model until fitting termination conditions are met, so as to complete oil deposit history fitting.
In one embodiment, the history-fit objective function of the dual-proxy model is:
In the history fit objective function of the dual-proxy model, m represents the reservoir model parameters obtained by the particle swarm algorithm, F DF (m) represents the reservoir production data predicted by the dual-proxy model, d obs represents the observation data, An inverse matrix representing the covariance matrix of the observed data, m pr representing the a priori model parameters,An inverse of the covariance matrix representing the a priori model parameters.
In one embodiment, the solving the history-fit objective function of the dual proxy model using a particle swarm algorithm to update the reservoir model parameters comprises:
initializing population and particle swarm algorithm parameters;
Calculating the fitness value of the particles, and updating the optimal position of the particles and the optimal position of the population according to the fitness value of the particles;
And updating the position and the speed of the particles according to the updated optimal position of the particles and the population optimal position.
In one embodiment, the calculating the fitness value of the particle and updating the particle optimal position and the population optimal position according to the fitness value of the particle comprises:
and updating the particle optimal position by using an updating expression of the particle optimal position according to the fitness value of the particle, wherein the updating expression of the particle optimal position is as follows:
in the updated expression of the optimal position of the particle, Represents the optimal position of the particles for the k+1st iteration, f () represents the fitness value function,Representing the particle position of the k +1 iteration,Indicating the optimal position of the particles for the kth iteration.
In one embodiment, the calculating the fitness value of the particle and updating the particle optimal position and the population optimal position according to the fitness value of the particle comprises:
Updating the optimal position of the population by using an updating expression of the optimal position of the population according to the fitness value of the particles, wherein the updating expression of the optimal position of the population is as follows:
In the updated expression of the optimal position of the population, Represents the optimal position of the population for the (k+1) th iteration, f () represents the fitness value function,Indicating the optimal position of the particles for the k +1 iteration,Representing the optimal position of the population for the kth iteration.
The invention also provides an oil reservoir history fitting system based on the error learning double-agent model, which comprises:
A setting module, configured to: setting a history fitting objective function of the double-agent model obtained by the construction method of the double-agent model based on error learning;
A solving module for: carrying out oil reservoir history fitting through the double-agent model, and solving a history fitting objective function of the double-agent model by utilizing a particle swarm algorithm so as to update oil reservoir model parameters;
a prediction module for: predicting to obtain oil reservoir production data according to the updated oil reservoir model parameters by using the double-agent model;
An iteration module for: and solving a history fit objective function of the double-agent model by using a particle swarm algorithm again according to the updated oil deposit model parameters and the predicted oil deposit production data, and predicting the oil deposit production data by using the double-agent model until fitting termination conditions are met, so as to complete oil deposit history fitting.
The invention also provides electronic equipment, which comprises a processor and a memory storing a computer program, wherein the processor realizes the construction method of the double-agent model based on error learning and/or the reservoir history fitting method based on the error learning double-agent model when executing the computer program.
The invention also provides a non-transitory computer readable storage medium, on which a computer program is stored, which when executed by a processor implements the method for constructing the dual-proxy model based on error learning and/or the method for fitting the reservoir history based on the dual-proxy model based on error learning described in any one of the above.
The present invention also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and when the computer program is executed by a processor, the computer can execute the method for constructing the dual-agent model based on error learning and/or the method for fitting reservoir history based on the dual-agent model based on error learning.
According to the oil reservoir history fitting method, system, equipment and medium based on the error learning double-agent model, the high-dimensional problem is converted into the low-dimensional problem by reducing the dimension of the parameters of the oil reservoir model, the first agent sub-model is used for replacing the oil reservoir numerical simulation process in the double-agent model, the second sub-model is used for learning the prediction error of the first sub-model, the prediction stability of the double-agent model can be better ensured, the history fitting objective function is subjected to global optimization solution by combining the particle swarm algorithm, the prediction accuracy and efficiency of the double-agent model can be greatly improved, meanwhile, the difficulty of setting the super-parameter can be reduced by the framework of the double-agent model, the reliable and efficient oil reservoir history fitting is facilitated, and the efficient and accurate prediction of the dynamic oil reservoir production data is realized.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following brief description will be given of the drawings used in the embodiments or the description of the prior art, it being obvious that the drawings in the following description are some embodiments of the invention and that other drawings can be obtained from them without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a reservoir history fitting method based on an error learning dual-agent model.
FIG. 2 is a graph of an example permeability field for a two-dimensional reservoir.
FIG. 3 is a graph of a two-dimensional reservoir permeability field generated from low-dimensional variables.
FIG. 4 is a graph comparing single well production data obtained using the dual proxy model, the existing single proxy model, and the reservoir numerical simulator provided by the present invention.
Fig. 5 is a schematic flow chart of a particle swarm algorithm.
FIG. 6 is a diagram of a history-fit objective function change for a dual-proxy model.
FIG. 7 is a graphical representation of the history fit results of a dual agent model.
FIG. 8 is a schematic diagram of a structure of a reservoir history fitting system based on an error learning dual-agent model.
Fig. 9 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions thereof will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments, which should not be construed as limiting the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention. In the description of the present invention, it is to be understood that the terminology used is for the purpose of description only and is not to be interpreted as indicating or implying relative importance.
The method, system, equipment and medium for oil reservoir history fitting based on the error learning double-agent model provided by the invention are described below with reference to fig. 1 to 9.
FIG. 1 is a flow chart of a reservoir history fitting method based on an error learning dual-agent model provided by the invention. Referring to fig. 1, the oil reservoir history fitting method based on the error learning dual-agent model provided by the invention may include:
Step S110, dimension reduction is carried out on the oil reservoir model parameters to obtain the oil reservoir model parameters after dimension reduction, wherein the oil reservoir model parameters can comprise static parameters, distribution and property data of fluid, initial oil reservoir conditions, production dynamic parameters and the like, and the oil reservoir model parameters in the embodiment are permeability field parameters belonging to the static parameters;
Step S120, carrying out oil deposit numerical simulation on the oil deposit model parameters subjected to the dimension reduction to obtain learning sample data of the oil deposit model parameters, wherein the learning sample data comprise oil deposit parameter sample data and oil deposit production sample data;
Step S130, training to obtain a double-agent model of the oil reservoir model parameters by utilizing a deep learning neural network according to learning sample data of the oil reservoir model parameters, wherein the double-agent model is used for replacing an oil reservoir numerical simulation process of the oil reservoir model parameters, and comprises a first agent sub-model and a second agent sub-model, the first agent sub-model is used for learning the relation between the oil reservoir parameter sample data and the oil reservoir production sample data, and the second agent sub-model is used for learning a prediction error of the first agent sub-model;
step S140, setting a history fit objective function of a double-agent model;
step S150, carrying out oil reservoir history fitting through the double-agent model, and solving a history fitting objective function of the double-agent model by utilizing a particle swarm algorithm so as to update oil reservoir model parameters;
step S160, predicting and obtaining oil reservoir production data by utilizing the double-agent model according to the updated oil reservoir model parameters;
And S170, solving a history fit objective function of the double-agent model by using a particle swarm algorithm again according to the updated oil reservoir model parameters and the predicted oil reservoir production data, and predicting the oil reservoir production data by using the double-agent model until fitting termination conditions are met, so as to complete oil reservoir history fitting.
It should be noted that, the execution main body of the oil deposit history fitting method based on the error learning dual-agent model provided by the invention can be any terminal side equipment meeting technical requirements, such as an oil deposit history fitting device based on the error learning dual-agent model.
In one embodiment, step S110 may use principal component analysis (PRINCIPAL COMPONENT ANALYSIS, PCA) to dimensionality reduce the reservoir model parameters to obtain the dimensionality reduced reservoir model parameters. For example, as shown in fig. 2, step S110 may reduce the dimension of 10000-dimension oil reservoir model parameters to 100-dimension oil reservoir model parameters, so as to shorten the calculation time and improve the calculation efficiency.
Specifically, assuming that the prior model (i.e., the target reservoir model, which may be obtained from logging, core, etc.) has N r independently generated prior model parameters m 1,m2,…,mNr, these prior model parameters may be considered as reservoir model parameters, i.e., the data to be fitted, the means of these prior model parametersThe method comprises the following steps:
For example, the N r a priori model parameters are different permeability fields, and the average value is the average value of the N r permeability fields;
and (3) removing the average value of the prior model parameter m to obtain a matrix X:
For covariance matrix Singular value decomposition is performed:
In the method, in the process of the invention, U represents a left singular matrix, R represents a domain, and N c represents the prior model grid number; Sigma represents a singular value matrix, other elements except a diagonal are zero, and the elements on the diagonal are singular values; V denotes a right singular matrix.
The singular values are ordered from large to small, and the number of only N L singular values lambda i.NL is kept as follows:
The dimension-reduced variable ζ L is:
in one embodiment, step S120 may generate thousands of reservoir model parameter samples from the reduced-dimension reservoir model parameters (i.e., low-dimensional variables of the reservoir model parameters) using the first expression And then, respectively carrying out oil deposit numerical simulation on a plurality of oil deposit model parameter samples by using an Eclipse oil deposit numerical simulator to obtain learning sample data of oil deposit model parameters, and dividing the learning sample data into a training set (800 groups of data) and a testing set (200 groups of data) for training and testing the double-agent model. For example, step S120 may generate 1000 oil reservoir model parameter samples from the 100-dimensional oil reservoir model parameters (100-dimensional variables) obtained in step S110, and a part of the oil reservoir model parameter samples are shown in fig. 3.
Wherein the first expression is:
In the first expression of the present invention, Representing a sample of reservoir model parameters generated using PCA,Representing the mean of all prior model parameters, U L represents the orthogonal matrix and ζ L represents the random number that fits the gaussian distribution. In the embodiment, a mapping relation between an oil reservoir parameter field and production data is established by using a double-proxy model, a second proxy model for learning a prediction error of the first proxy model is additionally arranged on the basis of the first proxy model for replacing an oil reservoir numerical simulation process, and the output of the double-proxy model is the sum of the output of the first proxy model and the output of the second proxy model.
In one embodiment, the expression of the dual agent model is:
FDF＝F1+F2，
in the expression of the dual-proxy model, F DF denotes a model framework of the dual-proxy model, F 1 denotes a prediction result of the first proxy sub-model, and F 2 denotes a prediction result of the second proxy sub-model.
In this embodiment, the first proxy model and the second proxy model are trained by using learning sample data, the inputs of the first proxy model and the second proxy model are all oil reservoir parameter sample data, such as permeability distribution, porosity distribution, net hair thickness ratio, and the like, the output of the first proxy model is oil reservoir production sample data, such as single well oil production rate, single well water production rate, bottom hole pressure, and the like, and the output of the second proxy model is a prediction error of the first proxy model.
The prior model may be predicted using a trained dual-proxy model to obtain prior production data. As shown in fig. 4, the prior production data obtained using the dual-proxy model substantially matches the production data obtained using the numerical simulator, and the predictions of the dual-proxy model are more accurate than those of the single-proxy model.
In one embodiment, the network structure of the first proxy sub-model includes a convolutional neural network (Convolutional Neural Networks, CNN) and a long-short-term memory recurrent neural network (Long Short Term Memory, LSTM), and the training is performed to obtain a dual-proxy model of the reservoir model parameters according to learning sample data of the reservoir model parameters by using the deep learning neural network, including:
Extracting spatial features of oil reservoir parameter sample data through a convolutional neural network and flattening the spatial features;
And obtaining time sequence information of oil reservoir production sample data through the long-term and short-term memory recurrent neural network, and learning a time sequence mode between the oil reservoir parameter sample data and the oil reservoir production sample data by combining the flattened spatial characteristics of the recurrent neural network.
In one embodiment, the second agent sub-model includes an encoder, a decoder, and a transition module, and the training is performed to obtain a dual agent model of the reservoir model parameters by using a deep learning neural network according to learning sample data of the reservoir model parameters, including:
Extracting oil reservoir parameter characteristics of oil reservoir parameter sample data through an encoder;
Mapping the oil reservoir parameter characteristics to a data space for learning the prediction error of the first proxy sub-model through a decoder;
and converting the prediction result of the second agent sub-model into an image form through a transition module.
In one embodiment, step S130 may include:
Setting a history fitting objective function of a double-agent model;
And solving a history fit objective function of the double-agent model by using a particle swarm algorithm so as to train and obtain a final double-agent model.
In one embodiment, the prior distribution of the prior model parameters and the observation error (the observation error epsilon is the error between the actual production data value and the production data d obs obtained by observing the oil field) are assumed to be gaussian distributions, and under the condition of given observation data, the posterior probability is known by the bayesian theory:
p(m|dobs)∝p(dobs|m)p(m)，
Wherein m represents a priori model parameter to be fitted; d obs represents historical observations, p (m|d obs) is the posterior probability of m; p (d obs |m) is the posterior probability of d obs; p (m) is the a priori probability of m.
Using the dual proxy model framework established in step S3, the relationship of the observed data d obs with the a priori model parameters m to be fitted can be expressed as:
dobs＝FDF(m)+ε，
Where ε corresponds to a Gaussian distribution with a mean of 0 and a covariance of C D, and represents the error of the observed data.
The posterior probability of the observed data d obs is:
The prior probability of the prior model parameter m to be fitted is as follows:
From the posterior probability of the observation data d obs and the prior probability of the prior model parameter m to be fitted, the posterior probability p (m|d obs) can be expressed as:
Wherein, C D is covariance matrix of the observed data; m pr is a priori model parameters; c M is the covariance matrix of the a priori model parameters.
The history-fit objective function of the dual-proxy model can be expressed as:
In the history fit objective function of the dual-proxy model, m represents the reservoir model parameters obtained by the particle swarm algorithm, F DF (m) represents the reservoir production data predicted by the dual-proxy model, d obs represents the observation data, An inverse matrix representing the covariance matrix of the observed data, m pr representing the a priori model parameters,An inverse of the covariance matrix representing the a priori model parameters.
The particle swarm algorithm has simple structure and high convergence speed, and is suitable for global optimization solution. In one embodiment, referring to fig. 5, step S150 may include:
A. initializing population and particle swarm algorithm parameters.
Specifically, the population size may be set to N p, and then the position x i and velocity v i of each particle in the D-dimensional space may be expressed as:
xi＝[xi,1,xi,2,...,xi,D],i＝1,2,...,NP，
vi＝[vi,1,vi,2,...,vi,D],i＝1,2,...,NP，
the optimal positions of the particles and populations, respectively, can be expressed as:
pi＝[pi,1,pi,2,...,pi,D],i＝1,2,...,NP，
pg＝[pg,1,pg,2,...,pg,D],i＝1,2,...,NP，
Where p i and p g represent the particle and population optimal positions, respectively.
B. And calculating the fitness value of the particles, and updating the optimal position of the particles and the optimal position of the population according to the fitness value of the particles.
The updating expression of the optimal position of the particle is as follows:
in the updated expression of the optimal position of the particle, Represents the optimal position of the particles for the k+1st iteration, f () represents the fitness value function,Representing the particle position of the k +1 iteration,Indicating the optimal position of the particles for the kth iteration.
And, the update expression of the population optimal position is:
In the updated expression of the optimal position of the population, Represents the optimal position of the population for the (k+1) th iteration, f () represents the fitness value function,Indicating the optimal position of the particles for the k +1 iteration,Representing the optimal position of the population for the kth iteration.
C. Updating the position and speed of the particles according to the updated particle optimal position and population optimal position;
In the method, in the process of the invention, Representing the particle velocity for the k +1 iteration,Representing the particle velocity of the kth iteration,Indicating the optimal position of the particles for the kth iteration,Representing the particle position of the k +1 iteration,Representing the particle position of the kth iteration,Representing the optimal position of the population for the kth iteration, ω represents the inertial weight, c 1 and c 2 represent the learning factors, and r 1 and r 2 represent the random numbers between [0,1 ].
After the history fit objective function of the dual-proxy model is solved in step S150 to update the oil reservoir model parameters, step S160 may utilize the dual-proxy model to replace the numerical simulator to predict and obtain oil reservoir production data according to the updated oil reservoir model parameters, step S170 then utilizes the particle swarm algorithm to solve the history fit objective function of the dual-proxy model and utilize the dual-proxy model to predict the oil reservoir production data again according to the updated oil reservoir model parameters and the predicted oil reservoir production data until the fit termination condition is met (e.g., the preset iteration number is reached, the difference between the result of the last history fit objective function and the result of the next history fit objective function is within a preset negligible range, etc.), and the oil reservoir history fit is completed.
In this embodiment, the population size is set to 100, the iteration number is 5, the inertia weight ω=1, the learning factor c 1＝c2 =1.5, and the particle swarm optimization is performed. FIG. 6 is a graph of a history fit objective function versus iteration number for a dual agent model. As shown in fig. 6, as the number of iterations increases, the population evolves and the history-fit objective function decreases and gradually converges. Fig. 7 shows the history fitting result of the dual-proxy model, and it can be seen from analysis of fig. 7 that the prior data is the reservoir data predicted by the dual-proxy model which is trained before optimization by using the objective function, the observed data is the actual reservoir data, the history fitting data of the dual-proxy model is the reservoir data predicted after optimization by using the objective function and the particle swarm algorithm, and the prior data and the observed data have larger errors, and the history fitting data of the dual-proxy model and the observed data are well matched.
Compared with the prediction result using a single agent model, the dual agent model framework provided by the invention can improve the accuracy of predicting the oil reservoir production data, and the prediction uncertainty of the posterior model can be better measured by considering the agent prediction error in the objective function of the history fitting.
The double-agent model framework provided by the invention can reduce the difficulty of super-parameter setting. In practice, the hyper-parameter adjustment is awkward and time consuming for the engineer. The framework facilitates reliable and efficient history matching by engineers.
According to the oil reservoir history fitting method based on the error learning double-agent model, the dimension of the oil reservoir model parameters is reduced, the high-dimensional problem is converted into the low-dimensional problem, the first agent sub-model in the double-agent model is used for replacing the oil reservoir numerical simulation process, the second sub-model is used for learning the prediction error of the first sub-model, the prediction stability of the double-agent model can be better ensured, the particle swarm algorithm is combined for carrying out global optimization solution on the history fitting objective function, the prediction accuracy and efficiency of the double-agent model can be greatly improved, meanwhile, the difficulty of setting the super-parameter can be reduced by the framework of the double-agent model, and the reliable and efficient oil reservoir history fitting is facilitated.
The oil deposit history fitting system based on the error learning double-agent model provided by the invention is described below, and the oil deposit history fitting system based on the error learning double-agent model described below and the oil deposit history fitting method based on the error learning double-agent model described above can be correspondingly referred to each other.
Referring to fig. 8, an oil reservoir history fitting system based on an error learning dual-agent model provided by the present invention may include:
A setting module, configured to: setting a history fitting objective function of the double-agent model obtained by the construction method of the double-agent model based on error learning;
A solving module for: carrying out oil reservoir history fitting through the double-agent model, and solving a history fitting objective function of the double-agent model by utilizing a particle swarm algorithm so as to update oil reservoir model parameters;
a prediction module for: predicting to obtain oil reservoir production data according to the updated oil reservoir model parameters by using the double-agent model;
An iteration module for: and solving a history fit objective function of the double-agent model by using a particle swarm algorithm again according to the updated oil deposit model parameters and the predicted oil deposit production data, and predicting the oil deposit production data by using the double-agent model until fitting termination conditions are met, so as to complete oil deposit history fitting.
Fig. 9 illustrates a physical schematic diagram of an electronic device, as shown in fig. 9, which may include: processor 810, communication interface (Communications Interface) 820, memory 830, and communication bus 840, wherein processor 810, communication interface 820, memory 830 accomplish communication with each other through communication bus 840. The processor 810 may invoke logic instructions in the memory 830 to perform the method of constructing the error learning-based dual-proxy model and/or the method of reservoir history fitting based on the error learning dual-proxy model provided in any of the above.
Further, the logic instructions in the memory 830 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product, where the computer program product includes a computer program, where the computer program can be stored on a non-transitory computer readable storage medium, and when the computer program is executed by a processor, the computer can execute the method for constructing a dual-proxy model based on error learning and/or the method for fitting a reservoir history based on the dual-proxy model based on error learning provided by the above methods.
In yet another aspect, the present invention further provides a non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor, is implemented to perform the method for building a dual proxy model based on error learning and/or the method for fitting a reservoir history based on error learning dual proxy model provided by the above methods.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. The method for constructing the double-agent model based on error learning is characterized by comprising the following steps of:
reducing the dimension of the oil reservoir model parameters to obtain the dimension-reduced oil reservoir model parameters;
Carrying out oil deposit numerical simulation on the oil deposit model parameters subjected to the dimension reduction to obtain learning sample data of the oil deposit model parameters, wherein the learning sample data comprises oil deposit parameter sample data and oil deposit production sample data;
According to learning sample data of oil reservoir model parameters, training to obtain a double-agent model of the oil reservoir model parameters by using a deep learning neural network, wherein the double-agent model comprises a first agent sub-model and a second agent sub-model, the first agent sub-model is used for learning the relation between the oil reservoir parameter sample data and oil reservoir production sample data, and the second agent sub-model is used for learning the prediction error of the first agent sub-model.
2. The method for constructing a dual-proxy model based on error learning according to claim 1, wherein the performing the reservoir numerical simulation on the reservoir model parameters after the dimension reduction to obtain learning sample data of the reservoir model parameters comprises:
generating a plurality of oil reservoir model parameter samples by using a first expression according to the oil reservoir model parameters after the dimension reduction;
respectively carrying out oil reservoir numerical simulation on a plurality of oil reservoir model parameter samples to obtain learning sample data of oil reservoir model parameters;
Wherein the first expression is:
In the first expression of the present invention, Representing the generated reservoir model parameter samples,Representing the mean of the reservoir model parameters for all prior models, U L represents the orthogonal matrix and ζ L represents random numbers conforming to a Gaussian distribution.
3. The method for constructing a dual-proxy model based on error learning as claimed in claim 1, wherein the expression of the dual-proxy model is:
FDF＝F1+F2,
in the expression of the dual-proxy model, F DF denotes a model framework of the dual-proxy model, F 1 denotes a prediction result of the first proxy sub-model, and F 2 denotes a prediction result of the second proxy sub-model.
4. The method for constructing a dual-proxy model based on error learning according to claim 3, wherein the network structure of the first proxy sub-model includes a convolutional neural network and a long-term memory recurrent neural network, the dual-proxy model for obtaining the reservoir model parameters is trained by using the deep learning neural network according to the learning sample data of the reservoir model parameters, and the method comprises the following steps:
Extracting spatial features of oil reservoir parameter sample data through a convolutional neural network and flattening the spatial features;
And obtaining time sequence information of oil reservoir production sample data through the long-term and short-term memory recurrent neural network, and learning a time sequence mode between the oil reservoir parameter sample data and the oil reservoir production sample data by combining the flattened spatial characteristics of the recurrent neural network.
5. The method for constructing a dual-proxy model based on error learning according to claim 4, wherein the second proxy sub-model comprises an encoder, a decoder and a transition module, the dual-proxy model for obtaining the reservoir model parameters is trained by using a deep learning neural network according to learning sample data of the reservoir model parameters, and the method comprises the following steps:
Extracting oil reservoir parameter characteristics of oil reservoir parameter sample data through an encoder;
Mapping the oil reservoir parameter characteristics to a data space for learning the prediction error of the first proxy sub-model through a decoder;
and converting the prediction result of the second agent sub-model into an image form through a transition module.
6. An oil reservoir history fitting method based on an error learning double-agent model is characterized by comprising the following steps of:
Setting a history fit objective function of the dual-agent model obtained by the method for constructing the dual-agent model based on error learning as claimed in any one of claims 1 to 5;
Carrying out oil reservoir history fitting through the double-agent model, and solving a history fitting objective function of the double-agent model by utilizing a particle swarm algorithm so as to update oil reservoir model parameters;
predicting to obtain oil reservoir production data according to the updated oil reservoir model parameters by using the double-agent model;
according to the updated oil reservoir model parameters and the predicted oil reservoir production data, solving a history fit objective function of the double-agent model by using a particle swarm algorithm again and predicting the oil reservoir production data by using the double-agent model until fitting termination conditions are met, and completing oil reservoir history fitting;
the history fit objective function of the dual-agent model is as follows:
In the history fit objective function of the dual-proxy model, m represents the reservoir model parameters obtained by the particle swarm algorithm, F DF (m) represents the reservoir production data predicted by the dual-proxy model, d obs represents the observation data, An inverse matrix representing the covariance matrix of the observed data, m pr representing the a priori model parameters,An inverse of the covariance matrix representing the a priori model parameters.
7. The method of claim 6, wherein solving the history-fit objective function of the dual-proxy model using a particle swarm algorithm to update the reservoir model parameters comprises:
initializing population and particle swarm algorithm parameters;
Calculating the fitness value of the particles, and updating the optimal position of the particles and the optimal position of the population according to the fitness value of the particles;
updating the position and speed of the particles according to the updated particle optimal position and population optimal position;
preferably, the calculating the fitness value of the particle, and updating the particle optimal position and the population optimal position according to the fitness value of the particle includes:
and updating the particle optimal position by using an updating expression of the particle optimal position according to the fitness value of the particle, wherein the updating expression of the particle optimal position is as follows:
in the updated expression of the optimal position of the particle, Represents the optimal position of the particles for the k+1st iteration, f () represents the fitness value function,Representing the particle position of the k +1 iteration,Representing the optimal position of the particles of the kth iteration;
preferably, the calculating the fitness value of the particle, and updating the particle optimal position and the population optimal position according to the fitness value of the particle includes:
Updating the optimal position of the population by using an updating expression of the optimal position of the population according to the fitness value of the particles, wherein the updating expression of the optimal position of the population is as follows:
In the updated expression of the optimal position of the population, Represents the optimal position of the population for the (k+1) th iteration, f () represents the fitness value function,Indicating the optimal position of the particles for the k +1 iteration,Representing the optimal position of the population for the kth iteration.
8. An oil reservoir history fitting system based on an error learning dual-agent model, comprising:
a setting module, configured to: setting a history fit objective function of the dual-agent model obtained by the method for constructing the dual-agent model based on error learning as claimed in any one of claims 1 to 5;
A solving module for: carrying out oil reservoir history fitting through the double-agent model, and solving a history fitting objective function of the double-agent model by utilizing a particle swarm algorithm so as to update oil reservoir model parameters;
a prediction module for: predicting to obtain oil reservoir production data according to the updated oil reservoir model parameters by using the double-agent model;
An iteration module for: and solving a history fit objective function of the double-agent model by using a particle swarm algorithm again according to the updated oil deposit model parameters and the predicted oil deposit production data, and predicting the oil deposit production data by using the double-agent model until fitting termination conditions are met, so as to complete oil deposit history fitting.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of building an error learning based dual-agent model according to any one of claims 1-5 and/or the method of reservoir history fitting based on an error learning dual-agent model according to any one of claims 6 or 7 when executing the program.
10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the method of constructing a dual proxy model based on error learning as claimed in any one of claims 1 to 5 and/or the method of reservoir history fitting based on an error learning dual proxy model as claimed in any one of claims 6 or 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410528088.8A CN118313265A (en) | 2024-04-29 | 2024-04-29 | Oil reservoir history fitting method, system, equipment and medium based on error learning double-agent model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410528088.8A CN118313265A (en) | 2024-04-29 | 2024-04-29 | Oil reservoir history fitting method, system, equipment and medium based on error learning double-agent model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118313265A true CN118313265A (en) | 2024-07-09 |
Family
ID=91733615
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410528088.8A Pending CN118313265A (en) | 2024-04-29 | 2024-04-29 | Oil reservoir history fitting method, system, equipment and medium based on error learning double-agent model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118313265A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118552338A (en) * | 2024-07-30 | 2024-08-27 | 中国石油大学(华东) | Fast update history fitting paradigm based on multi-stage history experience migration |
-
2024
- 2024-04-29 CN CN202410528088.8A patent/CN118313265A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118552338A (en) * | 2024-07-30 | 2024-08-27 | 中国石油大学(华东) | Fast update history fitting paradigm based on multi-stage history experience migration |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112308961B (en) | Robot rapid robust three-dimensional reconstruction method based on layered Gaussian mixture model | |
| CN108764540B (en) | Water supply network pressure prediction method based on parallel LSTM series DNN | |
| CN115935834B (en) | Historical fitting method based on deep autoregressive network and continuous learning strategy | |
| CN111783209B (en) | Self-adaptive structure reliability analysis method combining learning function and kriging model | |
| CN108804334B (en) | Discrete software reliability increase testing and evaluating method based on self-adaptive sampling | |
| CN112307536B (en) | Dam seepage parameter inversion method | |
| CN118313265A (en) | Oil reservoir history fitting method, system, equipment and medium based on error learning double-agent model | |
| CN111967183A (en) | Method and system for calculating line loss of distribution network area | |
| CN113777931B (en) | Icing wing type pneumatic model construction method, device, equipment and medium | |
| CN112163671A (en) | New energy scene generation 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 | |
| CN116775687A (en) | Time sequence prediction closed-loop framework based on digital-analog symbiosis | |
| CN115982141A (en) | Characteristic optimization method for time series data prediction | |
| CN111105005A (en) | Wind power prediction method | |
| CN117010260A (en) | Automatic history fit model prediction method, system and equipment for fractured reservoir | |
| CN112541254A (en) | Automatic oil reservoir history fitting method for optimizing deep learning dimension reduction reconstruction parameters | |
| CN116050300A (en) | Method and device for predicting aerodynamic thermal flow field of surface of high-speed moving body | |
| CN115712977A (en) | Gear reducer robust optimization design method based on assistance of Kriging surrogate model | |
| CN115293334B (en) | Model-based unmanned equipment control method for high-sample-rate deep reinforcement learning | |
| CN118643746B (en) | Automatic history fitting method integrating geologic modeling and numerical simulation agent model | |
| CN111914865A (en) | Probability main component analysis method based on random core | |
| CN116595883B (en) | Real-time online system state correction method for numerical reactor | |
| CN116127872A (en) | Method and device for predicting high-speed moving body surface aerodynamic pressure field | |
| CN118350473A (en) | Process tensor chromatography model training and process tensor chromatography method and device | |
| Kimura et al. | Bayesian Estimation of Deer Population Dynamics Using Hamiltonian Monte Carlo Algorithm |
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 |