CN114818548A - Aquifer parameter field inversion method based on convolution generated confrontation network - Google Patents
Aquifer parameter field inversion method based on convolution generated confrontation network Download PDFInfo
- Publication number
- CN114818548A CN114818548A CN202210738233.6A CN202210738233A CN114818548A CN 114818548 A CN114818548 A CN 114818548A CN 202210738233 A CN202210738233 A CN 202210738233A CN 114818548 A CN114818548 A CN 114818548A
- Authority
- CN
- China
- Prior art keywords
- parameter field
- aquifer
- convolution
- normal distribution
- sample
- 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.)
- Granted
Links
Images
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
- G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
-
- 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
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A20/00—Water conservation; Efficient water supply; Efficient water use
- Y02A20/152—Water filtration
Abstract
The invention discloses an aquifer parameter field inversion method based on convolution generation confrontation network, which utilizes the convolution generation confrontation network to construct and learn the mapping relation from heterogeneous characteristics of an aquifer parameter field to low-dimensional standard normal distribution vectors, realizes simple parameterized representation of complex heterogeneous characteristics, fully considers and reserves the complex heterogeneous characteristics of the aquifer parameter field, and obviously reduces the number of parameters to be estimated. According to the invention, the antagonism network generated by convolution is combined with the multi-step data assimilation set smoothing algorithm, and based on the underground water flow-solute transport numerical model, the water head and the concentration observation data, the inversion of the parameter field of the complex heterogeneous aquifer can be efficiently and reliably realized only by estimating the low-dimensional normal distribution variable, so that the simulation and the reliability of the underground water flow numerical model are effectively improved.
Description
Technical Field
The invention relates to an aquifer parameter field inversion method based on convolution generation confrontation network, and belongs to the technical field of groundwater numerical simulation inversion in the hydrogeology field.
Background
The underground water flow-solute transport numerical simulation technology is an important tool for reproducing and predicting the water flow and solute transport process in an aquifer, and is widely applied to the fields of underground water pollution process research, underground water pollution source tracing, risk assessment and the like.
In order to obtain a reliable numerical simulation result of underground water flow-solute transport, a parameter field of an aquifer needs to be accurately described. Due to the heterogeneity of underground media and the lack of observation data, the distribution of the characterization parameter field directly based on the drilling data has great uncertainty, which affects the reliability of the simulation result. Therefore, further estimation and correction of the parameter field by inversion simulation are usually required based on observed data such as water head and concentration.
However, directly inverting the inhomogeneous parametric field results in huge computational effort (i.e., the "dimensionality disaster" problem) due to the excessive number of parameters to be estimated. In order to reduce the number of estimated parameters, the heterogeneity of the parameter field is generally required to be simplified so as to reduce the calculation amount for solving the inversion problem, but the computation amount can reduce the fidelity and the simulation of the model, thereby limiting the reliability and the application range of the underground water flow-solute transport numerical simulation technology. In order to improve the efficiency and reliability of aquifer parameter field inversion, a scheme which is worth exploring is to combine the deep learning technology with the traditional inversion method.
Disclosure of Invention
The invention aims to overcome the technical defects in the prior art and provide an aquifer parameter field inversion method based on a convolution generated confrontation network, the method generates the confrontation network by developing and applying the convolution, realizes simple parametric representation of a complex heterogeneous aquifer parameter field, and can efficiently and reliably realize inversion of the complex heterogeneous aquifer parameter field only by estimating a low-dimensional standard normal distribution variable in inversion calculation, thereby improving the reliability of the underground water flow-solute transport simulation technology.
The invention adopts the following technical scheme: an aquifer parameter field inversion method based on convolution generation countermeasure network comprises the following steps:
step SS 1: collecting aquifer geological parameter prior data, and randomly generating an aquifer parameter field sample set meeting prior information by geological statistical software;
step SS 2: inputting the sample set of the aquifer parameter field into a convolution generation countermeasure network, learning the geological statistical characteristics of the confrontation network, constructing a mapping relation between low-dimensional standard normal distribution and the geological statistical characteristics of the confrontation network based on the sample set of the aquifer parameter field, and establishing an aquifer parameter field generation model;
step SS 3: constructing an initial numerical model of underground water flow-solute transport, and simulating the water flow and solute transport process;
step SS 4: randomly generating a series of low-dimensional standard normal distribution variable samples, inputting the samples into a trained convolution to generate a confrontation network, and obtaining corresponding aquifer parameter field samples;
step SS 5: inputting the aquifer parameter field sample generated in the step SS4 into the underground water flow-solute transport numerical model in the step SS3 to obtain a water head and concentration simulation value at an observation point;
step SS 6: updating a low-dimensional standard normal distribution variable sample by using a multi-step data assimilation set smoothing algorithm based on the analog value and the observed value of the water head and the concentration;
step SS 7: inputting the updated low-dimensional standard normal distribution variable sample in the step SS6 into a trained convolution generation confrontation network to obtain an updated aquifer parameter field sample, and inputting the underground water flow-solute transport numerical model to obtain a water head and concentration simulation value at an observation point corresponding to the updated aquifer parameter field;
step SS 8: repeating the steps SS6 and SS7 until a preset convergence criterion is reached; and inputting the low-dimensional standard normal distribution variable sample obtained by the last iteration into a convolution to generate a confrontation network, and obtaining a corresponding posterior aquifer parameter field.
As a preferred embodiment, step SS1 includes: and generating a random parameter field sample meeting the aquifer geological parameter prior data.
As a preferred embodiment, step SS2 specifically includes: the convolution generation countermeasure network includes a generatorGSum discriminatorD(ii) a The generatorGFor: learning the geological statistical characteristics of the aquifer parameter field, constructing a mapping relation between low-dimensional standard normal distribution and the geological statistical characteristics of the low-dimensional standard normal distribution, and forming an aquifer parameter field generation model, namely giving input of a low-dimensional standard normal distribution variable to generate an aquifer parameter field sample; the discriminatorDFor judging generatorsGAnd whether the generated aquifer parameter field sample accords with the geological statistical characteristics.
As a preferred embodiment, the convolution in step SS2 generates a generator of the counterpoise networkGSum discriminatorDPerforming alternate confrontation training, generatorGThe loss function used for training is:
distinguishing deviceDThe loss function used for training is:
in the formulas (1) and (2),Nwhich is indicative of the number of training samples,
represents a low-dimensional standard normal distribution variable,
representing a data stream by a generatorGThe resulting samples of the parameter field of the aquifer,
representing aquifer parameter field samples generated by geostatistical software.
As a preferred embodiment, the convolution generates a generator of the countermeasure networkGUsing convolutionThe multiple residual dense blocks replace the conventional convolutional layer as the basic architecture of the network.
As a preferred embodiment, the convolution generates a generator of the countermeasure networkGFor full convolution networks, i.e. generatorsGThe middle layer is a convolution layer and does not contain a full connection layer.
As a preferred embodiment, the multi-step data assimilation set smoothing algorithm in step SS6 updates the parameter samples to be estimated by using the following formula:
in the formula (3), the first and second groups,i=1, …, N iter as an iterative factor, in totalN iter Performing secondary iteration;j=1, …, N e is a sample factor, collectively included in the setN e A sample is obtained;
and
respectively represent the parameters before and after the update,
an interactive covariance matrix representing the parameters and observations,
in order to observe the covariance matrix of the errors,
in order to be a factor of the disturbance,
for the observed value after adding disturbance, the disturbance error is normally distributed, and the covariance matrix of the disturbance error is
，
For given input parameters
Model simulation values under the conditions.
As a preferred embodiment, the multi-step data assimilation set smoothing algorithm adopted in step SS6 only needs to invert a low-dimensional standard normal distribution variable, and then inputs the inverted low-dimensional standard normal distribution variable into a convolution countermeasure generation network to obtain a corresponding aquifer parameter field.
In a preferred embodiment, the steps SS6 to SS8 loop automatically implement the following process by writing Python scripts: the method comprises the steps of inputting a given low-dimensional standard normal distribution sample to generate an aquifer parameter field to realize, operating an underground water flow-solute transport model to obtain a water head and solute concentration simulation value, calling a multi-step data assimilation set smoothing algorithm to update the low-dimensional standard normal distribution sample, judging whether the convergence standard is reached, and outputting a posterior sample.
The invention achieves the following beneficial effects: the method utilizes the convolution to generate the confrontation network to map the complex aquifer heterogeneous parameter field to the low-dimensional standard normal distribution space, and can greatly reduce the number of inversion parameters compared with the prior art on the premise of keeping the heterogeneous characteristics of the parameter field, thereby obviously improving the efficiency of aquifer parameter field inversion and the reliability of the underground water flow-solute migration simulation technology.
Drawings
FIG. 1 is a schematic diagram of an aquifer parameter field inversion process based on convolution generation of a confrontation network according to the invention.
Fig. 2 is a schematic structural diagram of a convolutional encoder, which includes 3 parts: convolution/transposition convolution, batch normalization and activation functions, wherein the generator adopts a ReLU activation function, the discriminator adopts a LeakyReLU activation function, and the last convolution layer of the generator only contains convolution operation.
Fig. 3 is a schematic diagram of a convolution multiple residual dense block structure adopted by a convolution generation countermeasure network, and a double residual connection structure is adopted.
FIG. 4 is a comparison of geostatistical software GSLIB and random log-permeability coefficient field samples generated by a convolution-generated countermeasure network.
Fig. 5 is a logarithmic permeability coefficient field of a target aquifer, a prior logarithmic permeability coefficient field generated randomly, a posterior logarithmic permeability coefficient field obtained by inversion and a standard deviation field of a posterior sample, wherein black diamond points are tracer release positions, and white dots are positions of a waterhead and a concentration observation point.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
Example 1: as shown in fig. 1, the invention provides an aquifer parameter field inversion method for generating a confrontation network based on convolution, which is characterized by comprising the following steps:
step SS 1: collecting aquifer geological parameter prior data, and randomly generating an aquifer parameter field sample set meeting prior information by geological statistical software;
step SS 2: inputting the sample set of the aquifer parameter field into a convolution generation countermeasure network, learning the geological statistical characteristics of the confrontation network, constructing a mapping relation between low-dimensional standard normal distribution and the geological statistical characteristics of the confrontation network based on the sample set of the aquifer parameter field, and establishing an aquifer parameter field generation model;
step SS 3: constructing an initial numerical model of underground water flow-solute transport, and simulating the water flow and solute transport process;
step SS 4: randomly generating a series of low-dimensional standard normal distribution variable samples, inputting the samples into a trained convolution to generate a confrontation network, and obtaining corresponding aquifer parameter field samples;
step SS 5: inputting the aquifer parameter field sample generated in the step SS4 into the underground water flow-solute transport numerical model in the step SS3 to obtain a water head and concentration simulation value at an observation point;
step SS 6: updating a low-dimensional standard normal distribution variable sample by using a multi-step data assimilation set smoothing algorithm based on the analog value and the observed value of the water head and the concentration;
step SS 7: inputting the updated low-dimensional standard normal distribution variable sample in the step SS6 into a trained convolution generation confrontation network to obtain an updated aquifer parameter field sample, and inputting the underground water flow-solute transport numerical model to obtain a water head and concentration simulation value at an observation point corresponding to the updated aquifer parameter field;
step SS 8: repeating the steps SS6 and SS7 until a preset convergence criterion is reached; and inputting the low-dimensional standard normal distribution variable sample obtained by the last iteration into a convolution generation confrontation network to obtain a corresponding posterior aquifer parameter field.
As a preferred embodiment, step SS1 includes: and generating random parameter field samples meeting the aquifer geological parameter prior information.
As a preferred embodiment, step SS2 specifically includes: the convolution generation countermeasure network includes a generatorGSum discriminatorD(ii) a The generatorGFor: learning the geological statistical characteristics of the aquifer parameter field, constructing a mapping relation between low-dimensional standard normal distribution and the geological statistical characteristics of the low-dimensional standard normal distribution, and forming an aquifer parameter field generation model, namely giving input of a low-dimensional standard normal distribution variable to generate an aquifer parameter field sample; the discriminatorDFor judging generatorsGAnd whether the generated aquifer parameter field sample accords with the geological statistical characteristics.
As a preferred embodiment, the convolution in step SS2 generates a generator of the counterpoise networkGSum discriminatorDPerforming alternate confrontation training, generatorGThe loss function used for training is:
distinguishing deviceDThe loss function used for training is:
in the formulas (1) and (2),Nwhich is indicative of the number of training samples,
represents a low-dimensional standard normal distribution variable,
representing a data stream by a generatorGThe resulting samples of the parameter field of the aquifer,
representing aquifer parameter field samples generated by geostatistical software.
As a preferred embodiment, the convolution generates a generator of the countermeasure networkGThe convolution multiple residual error dense block is adopted to replace the traditional convolution layer to serve as a basic framework of the network, so that the network training is more stable, and the capability of learning the complex mapping from the low-dimensional standard normal distribution to the geological statistical characteristics is stronger.
In a preferred embodiment, the generator G of the convolution generation countermeasure network is a full convolution network, i.e. the generator G is a convolution layer and does not contain a full connection layer. The full convolution network structure not only can fully learn the spatial statistical characteristics of the aquifer parameter field, but also can obviously reduce the number of network parameters.
As a preferred embodiment, the multi-step data assimilation set smoothing algorithm in step SS6 updates the parameter samples to be estimated by using the following formula:
in the formula (3), the first and second groups,i=1, …, N iter is an iteration factor, coN iter Performing secondary iteration;j=1, …, N e is a sample factor, collectively included in the setN e A sample is obtained;
and
respectively represent the parameters before and after the update,
an interactive covariance matrix representing the parameters and observations,
in order to observe the error covariance matrix,
is a perturbation factor (usually set to
）,
For the observation value after adding disturbance, the disturbance error is in normal distribution, and the covariance matrix of the disturbance error is
，
For given input parameters
Model simulation values under the conditions.
As a preferred embodiment, the multi-step data assimilation set smoothing algorithm adopted in step SS6 only needs to invert a low-dimensional standard normal distribution variable, and then inputs the inverted low-dimensional standard normal distribution variable into a convolution countermeasure generation network to obtain a corresponding aquifer parameter field. Therefore, the number of inversion parameters is obviously reduced, and the calculation efficiency is improved.
As a preferred embodiment, the steps SS 6-SS 8 loop automatically implement the following process by writing Python script: the method comprises the steps of inputting a given low-dimensional standard normal distribution sample to generate an aquifer parameter field to realize, operating an underground water flow-solute transport model to obtain a water head and solute concentration simulation value, calling a multi-step data assimilation set smoothing algorithm to update the low-dimensional standard normal distribution sample, judging whether the convergence standard is reached, and outputting a posterior sample.
According to the method, a permeability coefficient field generated randomly by geological statistical software GSLIB according to geological parameter prior information of an actual aquifer is used as a target parameter field, as shown in figure 5, underground water flows from left to right, the hydraulic gradient is set to be 0.04, tracers are released in 3 wells at the upstream, tracer concentration data of water heads and tracer concentration data at 8 moments are collected at 10 different depths of 24 observation wells at the downstream, and 2160 observation data are counted. To simulate the actual observation error, a 5% noise perturbation was added to the 2160 collected observations, and the permeability coefficient field was inverted based on the collected observations. The specific implementation steps include 7 stages:
stage 1: based on prior information, 40000 samples of permeability coefficient field random fields with different geostatistical features (mean, variance, correlation length in three directions) were generated by GSLIB.
Stage 2: training convolution based on 40000 samples generated in the stage 1 to generate a confrontation network, and learning a function mapping relation from a 256-dimensional standard normal distribution to a penetration coefficient random field containing 10 × 32 × 64= 20480 spatial grids as shown in formula (4)
:
The structure of the convolution generation countermeasure network is shown in fig. 2 and 3, network training is performed by using the loss functions shown in formulas (1) and (2), network parameters are updated by using an Adam optimizer carried in a Pytorch library, the learning rates of the generator and the discriminator are set to be 0.0002, the sample batch size (batch size) is set to be 100, the number of training rounds (epoch) is set to be 50, and other parameters of the Adam optimizer use system default values.
Stage 3: and randomly generating 400 groups of 256-dimensional standard normal distribution variable samples, inputting the samples into a generator for generating a countermeasure network by convolution, and obtaining 400 corresponding groups of permeability coefficient fields for realization. Fig. 4 compares the implementation of the random permeability coefficient field generated by the geostatistical software and the convolution-generated countermeasure network, and it can be seen from the figure that the permeability coefficient field generated by the convolution-generated countermeasure network and the permeability coefficient field generated by the geostatistical software have similar heterogeneous characteristics.
And 4, stage: and (3) giving 400 groups of generated permeability coefficient fields to realize, and taking the obtained permeability coefficient fields as input of an underground water flow-solute transport model to obtain analog values of sampling positions and time corresponding to observed data of water head and tracer concentration.
Stage 5: based on 2160 observed water head and tracer concentration data and the simulated values of the model under 400 sets of permeability coefficient field conditions, 400 sets of 256-dimensional standard normal distribution variable samples are updated by using a multi-step data assimilation set smoothing algorithm (the iteration number is set to be 20 times) based on the equation shown in the formula (3).
Stage 6: inputting the updated 400 groups of 256-dimensional standard normal distribution variable samples into a generator for generating a countermeasure network by convolution to obtain the updated 400 groups of permeability coefficient fields.
Stage 7: judging whether the maximum iteration times is reached, if not, repeating the 4 th to 6 th stages; if the maximum iteration times are reached, the loop is terminated, and 400 groups of permeability coefficient fields obtained by the last iteration updating are used as posterior samples of the inversion permeability coefficient fields.
The steps 4-7 are circularly and automatically realized by writing Python scripts: the method comprises the steps of inputting a given low-dimensional standard normal distribution sample to generate an aquifer parameter field to realize, operating an underground water flow-solute transport model to obtain a water head and solute concentration simulation value, calling a multi-step data assimilation set smoothing algorithm to update the low-dimensional standard normal distribution sample, judging whether the convergence standard is reached, and outputting a posterior sample.
Fig. 5 shows a target aquifer parameter field, 1 prior and 1 posterior aquifer parameter fields selected randomly, and inversion standard deviations (calculated from 400 posterior samples) of the aquifer parameter field, and it can be seen from fig. 5 that the prior aquifer parameter field is significantly different from the target parameter field, and the aquifer parameter field obtained by inversion based on the waterhead and concentration observation data is very close to the target parameter field in distribution.
It is worth mentioning that the permeability coefficient values of 20480 grids can be obtained by inversion only by estimating 256 standard normal distribution variables in the inversion process, and the number of inversion parameters is reduced by 98.75% on the premise of no loss of heterogeneous characteristics of a aquifer parameter field. Therefore, the invention can effectively and efficiently realize inversion identification of the aquifer parameter field.
As will be appreciated by one skilled in the art, 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.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (9)
1. An aquifer parameter field inversion method based on convolution generation confrontation network is characterized by comprising the following steps:
step SS 1: collecting aquifer geological parameter prior data, and randomly generating an aquifer parameter field sample set meeting prior information by geological statistical software;
step SS 2: inputting the sample set of the aquifer parameter field into a convolution generation countermeasure network, learning the geological statistical characteristics of the confrontation network, constructing a mapping relation between low-dimensional standard normal distribution and the geological statistical characteristics of the confrontation network based on the sample set of the aquifer parameter field, and establishing an aquifer parameter field generation model;
step SS 3: constructing an initial numerical model of underground water flow-solute transport, and simulating the water flow and solute transport process;
step SS 4: randomly generating a series of low-dimensional standard normal distribution variable samples, inputting the samples into a trained convolution to generate a confrontation network, and obtaining corresponding aquifer parameter field samples;
step SS 5: inputting the aquifer parameter field sample generated in the step SS4 into the underground water flow-solute transport numerical model in the step SS3 to obtain a water head and concentration simulation value at an observation point;
step SS 6: updating a low-dimensional standard normal distribution variable sample by using a multi-step data assimilation set smoothing algorithm based on the analog value and the observed value of the water head and the concentration;
step SS 7: inputting the updated low-dimensional standard normal distribution variable sample in the step SS6 into a trained convolution generation confrontation network to obtain an updated aquifer parameter field sample, and inputting the underground water flow-solute transport numerical model to obtain a water head and concentration simulation value at an observation point corresponding to the updated aquifer parameter field;
step SS 8: repeating the steps SS6 and SS7 until a preset convergence criterion is reached; and inputting the low-dimensional standard normal distribution variable sample obtained by the last iteration into a convolution to generate a confrontation network, and obtaining a corresponding posterior aquifer parameter field.
2. The method for the aquifer parameter field inversion based on the convolution generation countermeasure network as claimed in claim 1, wherein the step SS1 includes: and generating a random parameter field sample meeting the aquifer geological parameter prior data.
3. The method for inverting the aquifer parameter field based on the convolution generation countermeasure network according to claim 1, wherein the step SS2 specifically comprises: the convolution generation countermeasure network contains the generatorGSum discriminatorD(ii) a The generatorGFor: learning the geological statistical characteristics of the aquifer parameter field, constructing a mapping relation between low-dimensional standard normal distribution and the geological statistical characteristics of the low-dimensional standard normal distribution, and forming an aquifer parameter field generation model, namely giving input of a low-dimensional standard normal distribution variable to generate an aquifer parameter field sample; the discriminatorDFor judging generatorsGThe resultant water contentWhether the layer parameter field samples meet the geostatistical characteristics.
4. The aquifer parameter field inversion method for generating a countermeasure network based on convolution of claim 3, wherein the convolution in the step SS2 is used for generating a generator of the countermeasure networkGSum discriminatorDPerforming alternate confrontation training, generatingGThe loss function used for training is:
distinguishing deviceDThe loss function used for training is:
in the formulas (1) and (2),Nwhich is indicative of the number of training samples,
represents a low-dimensional standard normal distribution variable,
representation generatorGThe samples of the parameter field of the aquifer that are generated,
representing aquifer parameter field samples generated by geostatistical software.
5. The method of claim 3, wherein the generator of the convolutional antagonistic network is a generator of the convolutional antagonistic networkGThe convolution multiple residual error dense block is adopted to replace the traditional convolution layer to serve as the basic framework of the network.
6. The method of claim 3, wherein the generator of the convolutional antagonistic network is a generator of the convolutional antagonistic networkGFor full convolutional networks, i.e. generatorsGThe middle layer is a convolution layer and does not contain a full connection layer.
7. The method for inverting the aquifer parameter field based on the convolution generation countermeasure network as claimed in claim 1, wherein the multi-step data assimilation set smoothing algorithm in the step SS6 adopts the following formula to update the parameter samples to be estimated:
in the formula (3), the first and second groups,i=1, …, N iter as an iterative factor, in totalN iter Performing secondary iteration;j=1, …, N e is a sample factor, collectively included in the setN e A sample is obtained;
and
respectively represent the parameters before and after the update,
an interactive covariance matrix representing the parameters and observations,
in order to observe the error covariance matrix,
in order to be a factor of the disturbance,
for the observed value after adding disturbance, the disturbance error is normally distributed, and the covariance matrix of the disturbance error is
，
For given input parameters
Model simulation values under the conditions.
8. The aquifer parameter field inversion method based on convolution generation confrontation network according to claim 1, characterized in that the multistep data assimilation set smoothing algorithm adopted in the step SS6 only needs to invert low-dimensional standard normal distribution variables, and then inputs the variables into convolution confrontation generation network to obtain the corresponding aquifer parameter field.
9. The method for inverting the aquifer parameter field based on the convolution generation countermeasure network according to claim 1, wherein the steps SS 6-SS 8 loop to automatically implement the following processes by writing Python scripts: the method comprises the steps of inputting a given low-dimensional standard normal distribution sample to generate an aquifer parameter field to realize, operating an underground water flow-solute transport model to obtain a water head and solute concentration simulation value, calling a multi-step data assimilation set smoothing algorithm to update the low-dimensional standard normal distribution sample, judging whether the convergence standard is reached, and outputting a posterior sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210738233.6A CN114818548B (en) | 2022-06-28 | 2022-06-28 | Aquifer parameter field inversion method for generating confrontation network based on convolution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210738233.6A CN114818548B (en) | 2022-06-28 | 2022-06-28 | Aquifer parameter field inversion method for generating confrontation network based on convolution |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114818548A true CN114818548A (en) | 2022-07-29 |
| CN114818548B CN114818548B (en) | 2022-09-06 |
Family
ID=82522287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210738233.6A Active CN114818548B (en) | 2022-06-28 | 2022-06-28 | Aquifer parameter field inversion method for generating confrontation network based on convolution |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114818548B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116070549A (en) * | 2023-03-06 | 2023-05-05 | 西南交通大学 | Underground space flooding situation rapid deduction method, device, equipment and medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190280A (en) * | 2018-09-18 | 2019-01-11 | 东北农业大学 | A kind of pollution source of groundwater inverting recognition methods based on core extreme learning machine alternative model |
| US20190302290A1 (en) * | 2018-03-27 | 2019-10-03 | Westerngeco Llc | Generative adversarial network seismic data processor |
| CN111711820A (en) * | 2020-05-13 | 2020-09-25 | 哈尔滨工业大学 | Crack image compression sampling method based on generation countermeasure network |
| CN112149353A (en) * | 2020-09-24 | 2020-12-29 | 南京大学 | Method for identifying DNAPL pollutant distribution in underground aquifer based on convolutional neural network |
| CN112949089A (en) * | 2021-04-01 | 2021-06-11 | 吉林大学 | Aquifer structure inversion identification method based on discrete convolution residual error network |
| CN113537354A (en) * | 2021-07-19 | 2021-10-22 | 吉林大学 | Aquifer structure stage type stochastic inversion identification method based on deep learning |
-
2022
- 2022-06-28 CN CN202210738233.6A patent/CN114818548B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190302290A1 (en) * | 2018-03-27 | 2019-10-03 | Westerngeco Llc | Generative adversarial network seismic data processor |
| CN109190280A (en) * | 2018-09-18 | 2019-01-11 | 东北农业大学 | A kind of pollution source of groundwater inverting recognition methods based on core extreme learning machine alternative model |
| CN111711820A (en) * | 2020-05-13 | 2020-09-25 | 哈尔滨工业大学 | Crack image compression sampling method based on generation countermeasure network |
| CN112149353A (en) * | 2020-09-24 | 2020-12-29 | 南京大学 | Method for identifying DNAPL pollutant distribution in underground aquifer based on convolutional neural network |
| CN112949089A (en) * | 2021-04-01 | 2021-06-11 | 吉林大学 | Aquifer structure inversion identification method based on discrete convolution residual error network |
| CN113537354A (en) * | 2021-07-19 | 2021-10-22 | 吉林大学 | Aquifer structure stage type stochastic inversion identification method based on deep learning |
Non-Patent Citations (5)
| Title |
|---|
| JICHAO BAO等: "Coupling ensemble smoother and deep learning with generative adversarial networks to deal with non-Gaussianity in flow and transport data assimilation", 《JOURNAL OF HYDROLOGY》 * |
| SHAOXING MO等: "Deep Convolutional Encoder-Decoder Networks for Uncertainty Quantification of Dynamic Multiphase Flow in Heterogeneous Media", 《WATER RESOURCES RESEARCH》 * |
| 康文杰等: "深度卷积生成对抗网络对神经影像通用数据特征的学习", 《智慧健康》 * |
| 李家兴: "基于深度学习的含水层参数反演研究", 《中国优秀博硕士学位论文全文数据库(硕士)（基础科学辑）》 * |
| 莫绍星: "基于深度学习的地下水模拟高维不确定性分析和反演", 《中国优秀博硕士学位论文全文数据库(博士)(工程科技Ⅰ辑)》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116070549A (en) * | 2023-03-06 | 2023-05-05 | 西南交通大学 | Underground space flooding situation rapid deduction method, device, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114818548B (en) | 2022-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104866578B (en) | A kind of imperfect Internet of Things data mixing fill method | |
| US8805659B2 (en) | System and method for uncertainty quantification in reservoir simulation | |
| CN113052371A (en) | Residual oil distribution prediction method and device based on deep convolutional neural network | |
| CN106951512A (en) | A kind of end-to-end session control method based on hybrid coding network | |
| CN109300128B (en) | Transfer learning image processing method based on convolution neural network hidden structure | |
| CN110162972B (en) | UAF vulnerability detection method based on statement joint coding deep neural network | |
| Tianqi et al. | Development and application of a new algorithm for automated pit removal for grid DEMs | |
| CN114818548B (en) | Aquifer parameter field inversion method for generating confrontation network based on convolution | |
| CN114723095A (en) | Missing well logging curve prediction method and device | |
| CN111027249B (en) | Machine learning-based inter-well connectivity evaluation method | |
| CN111058840A (en) | Organic carbon content (TOC) evaluation method based on high-order neural network | |
| CN115169543A (en) | Short-term photovoltaic power prediction method and system based on transfer learning | |
| CN115049124A (en) | Deep and long tunnel water inrush prediction method based on Bayesian network | |
| CN112949089B (en) | Aquifer structure inversion identification method based on discrete convolution residual error network | |
| CN113194493B (en) | Wireless network data missing attribute recovery method and device based on graph neural network | |
| CN110633868A (en) | Method for predicting properties of oil and gas in exploration well oil testing layer by optimizing neural network through genetic algorithm | |
| Wan et al. | SWMM calibration using genetic algorithms | |
| Juda et al. | Comparison of three recent discrete stochastic inversion methods and influence of the prior choice | |
| CN113537354B (en) | Aquifer structure staged stochastic inversion identification method based on deep learning | |
| CN110879778A (en) | Novel dynamic feedback and improved patch evaluation software automatic restoration method | |
| CN113641733B (en) | Real-time intelligent estimation method for river cross section flow | |
| CN113392958B (en) | Parameter optimization and application method and system of fuzzy neural network FNN | |
| Juda et al. | An attempt to boost posterior population expansion using fast machine learning algorithms | |
| CN113313700A (en) | X-ray image interactive segmentation method based on deep learning | |
| CN110866607A (en) | Machine learning-based penetration behavior prediction 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |