An intelligent identification method and system for non-homogeneous aquifer structure by combining resistivity tomography and hydrological observation data

Abstract

The application discloses a kind of intelligent identification method and system for heterogeneous aquifer structure fusing resistivity tomography and hydrological observation data.Method includes: based on lithofacies observation condition data, using random simulation method to train aquifer structure generation model;Based on aquifer structure generation model and solute transport simulation, construct the data set of "lithofacies structure-concentration, head data";Based on lithophysical relationship equation and the forward simulation of resistivity tomography method, construct the data set of "lithofacies structure-apparent resistivity data";Based on the data set of "lithofacies structure-concentration, head data" and the data set of "lithofacies structure-apparent resistivity data", respectively train alternative model for predicting hydrological observation data and alternative model for predicting apparent resistivity data;Based on different types of actual observation data, use data fusion algorithm to optimize random variable, finally obtain a group of heterogeneous aquifer structure with minimum error between dynamic response observation data and actual observation data.

Description

TECHNICAL FIELD

[0001] The application belongs to the application field based on the hydrogeophysical cross-disciplinary theory and deep learning technology, and particularly relates to a non-homogeneous aquifer structure intelligent identification method and system fusing resistivity tomography and hydrological observation data. BACKGROUND

[0002] In the field of earth science, accurately identifying the complex lithofacies structure of the underground non-homogeneous aquifer is a core challenge. A reliable aquifer structure model is not only a key prerequisite for pollutant migration simulation and risk warning, but also an important basis for groundwater resource management and protection.

[0003] The development of the non-homogeneous aquifer structure identification method has experienced two main stages of random simulation and data assimilation. With the introduction of deep learning technology, the advantages of random simulation and data assimilation can be combined to realize accurate and rapid inversion identification of the non-homogeneous aquifer structure. The rise of the hydrogeophysical cross-disciplinary subject provides new impetus for the development of the current non-homogeneous aquifer structure identification method, and the key technical node lies in how to scientifically and efficiently extract the required hydrogeological information from the geophysical data. Compared with the discrete but highly valuable concentration, the water head observation data, the geophysical data are relatively continuous and easy to obtain, so fusing the geophysical data and the hydrological observation data can effectively improve the accuracy of the non-homogeneous aquifer structure identification and reduce the uncertainty.

[0004] The process of fusing the geophysical data needs to repeatedly call the geophysical forward model, which is relatively time-consuming and laborious. With the development of deep learning technology, such as attention mechanism, group convolution residual module, etc., the geophysical data can be effectively predicted to realize rapid and accurate identification of the non-homogeneous aquifer structure. SUMMARY

[0005] To solve the above technical problems, the application provides a non-homogeneous aquifer structure intelligent identification method and system fusing resistivity tomography and hydrological observation data, which can provide more valuable information for the inversion process by fusing the apparent resistivity data and the concentration and water head observation data, improve the structure identification accuracy, and reduce the uncertainty of the generated structure.

[0006] To achieve the above purpose, the application adopts the following technical solutions:

[0007] A non-homogeneous aquifer structure intelligent identification method fusing resistivity tomography and hydrological observation data, the method comprising:

[0008] Training a non-homogeneous aquifer structure generation model using a random simulation method based on lithofacies observation condition data;

[0009] Based on the heterogeneous aquifer structure generation model and solute transport simulation, the dataset of "facies structure-concentration, water head data" is constructed;

[0010] Based on the petrophysical relationship equation and resistivity tomography forward simulation, the dataset of "facies structure-apparent resistivity data" is constructed;

[0011] Based on the dataset of "facies structure-concentration, water head data" and the dataset of "facies structure-apparent resistivity data", the surrogate model for predicting hydrological observation data and the surrogate model for predicting apparent resistivity data are trained respectively;

[0012] Based on different types of actual observation data, the data fusion algorithm is used to optimize the random variables, and finally a set of heterogeneous aquifer structure with the minimum error between dynamic response observation data and actual observation data is obtained.

[0013] Preferably, based on the facies observation condition data, the random simulation method is used to train the heterogeneous aquifer structure generation model, which includes:

[0014] Carrying out field investigation in the research area, obtaining drilling and outcrop profile facies distribution data, converting them into random simulation condition data, and setting the volume proportion and extension length parameter prior interval of different facies according to the condition data;

[0015] Based on the Latin hypercube sampling method, N e parameters are extracted within the determined prior interval, and the parameters are combined with the condition data to carry out random simulation, obtaining N e simulated facies;

[0016] The obtained simulated facies is scaled to the interval [0, 1], and after being arranged into a dataset, the heterogeneous aquifer structure generation model is trained.

[0017] Preferably, based on the heterogeneous aquifer structure generation model and solute transport simulation, the dataset of "facies structure-concentration, water head data" is constructed, which includes:

[0018] Based on the Latin hypercube sampling method, N c random variables are extracted within the standard normal distribution interval, and then input into the trained heterogeneous aquifer structure generation model, obtaining N c facies structure expressed in the range of [0, 1];

[0019] The obtained facies structure is subjected to the inverse operation of the trained heterogeneous aquifer structure generation model, and is enlarged from the interval [0, 1] to the same facies data as the N e simulated facies expression mode, obtaining N c random facies structure;

[0020] The obtained lithofacies structure is subjected to solute transport simulation, concentration fields at different times matched with the heterogeneous structure are obtained, and the data set R1 from the structure to the hydrological response data is obtained after normalization processing of different types of response data. c

[0021] Preferably, the data set of the lithofacies structure and the apparent resistivity data is constructed by the forward simulation based on the rock physical relationship equation and the resistivity tomography method, and includes:

[0022] The obtained concentration state field at different times is converted into the true resistivity field at different times based on the rock physical relationship equation;

[0023] The forward simulation of the resistivity tomography method is carried out on the true resistivity field to obtain the corresponding apparent resistivity data at different times, and the data set R2 from the structure to the apparent resistivity original monitoring data is obtained after normalization processing of the apparent resistivity data. c

[0024] Preferably, the data fusion algorithm is used to optimize the random variables based on different types of actual observation data, and finally a set of heterogeneous aquifer structures with minimum error between the dynamic response observation data and the actual observation data is obtained, and includes:

[0025] N i initial random variables are extracted in the standard normal distribution interval based on the Latin hypercube sampling method, N i heterogeneous structures are obtained, and the concentration, the water head state field and the apparent resistivity data original monitoring data at the corresponding observation positions are obtained after inputting the surrogate model for predicting the hydrological observation data and the surrogate model for predicting the apparent resistivity data;

[0026] The obtained simulation observation data and the actual observation data are input into the data assimilation algorithm to complete the optimization of the N i initial random variables;

[0027] When the optimization reaches the set iteration number, the N i final random variables after optimization are obtained;

[0028] The obtained N i random variables are input into the generating model to obtain N i heterogeneous aquifer structures after optimization.

[0029] Meanwhile, the application also provides a heterogeneous aquifer structure intelligent identification system fusing the resistivity tomography and the hydrological observation data, which includes a first model training module, a first data set construction module, a second data set construction module, a second model training module and an optimization module.

[0030] ​​The first model training module is configured to train a heterogeneous aquifer structure generation model based on lithofacies observation condition data using a random simulation method.

[0031] The first data set construction module is configured to construct a data set of "lithofacies structure-concentration, water head data" based on the heterogeneous aquifer structure generation model and solute transport simulation.

[0032] The second data set construction module is configured to construct a data set of "lithofacies structure-apparent resistivity data" based on a lithophysical relationship equation and forward simulation of electrical resistivity tomography.

[0033] The second model training module is configured to train a substitute model for predicting hydrological observation data and a substitute model for predicting apparent resistivity data based on the data set of "lithofacies structure-concentration, water head data" and the data set of "lithofacies structure-apparent resistivity data", respectively.

[0034] The optimization module is configured to optimize random variables using a data fusion algorithm based on different types of actual observation data, and finally obtain a set of heterogeneous aquifer structures with minimum error between dynamic response observation data and actual observation data.

[0035] Preferably, the first model training module comprises an extension unit, a simulation unit and a training unit.

[0036] The extension unit is configured to carry out field investigation of a study area, obtain drilling and outcrop profile lithofacies distribution data, convert the data into condition data for random simulation, and set the volume proportion and extension length parameter prior interval of different lithofacies according to the condition data.

[0037] The simulation unit is configured to extract N e parameters in the determined prior interval based on a Latin hypercube sampling method, combine the parameters with the condition data to carry out random simulation, and obtain N e simulated lithofacies.

[0038] The training unit is configured to scale the obtained simulated lithofacies to the interval [0, 1], arrange the lithofacies into a data set, and train a heterogeneous aquifer structure generation model.

[0039] Preferably, the optimization module comprises an acquisition unit, an optimization unit, an iteration unit and a generation unit.

[0040] The acquisition unit is configured to extract N i initial random variables in a standard normal distribution interval based on a Latin hypercube sampling method, obtain N i heterogeneous structures, and input the structures into the substitute model for predicting hydrological observation data and the substitute model for predicting apparent resistivity data to obtain concentration, water head state field and apparent resistivity data original monitoring data at corresponding observation positions, respectively.

[0041] The optimization unit is used for inputting the obtained simulation observation data and actual observation data into a data assimilation algorithm, and completing optimization of N i initial random variables;

[0042] The iteration unit is used for obtaining N i final random variables when optimization reaches a set number of iterations;

[0043] The generation unit is used for inputting the obtained N i random variables into a generation model, and obtaining N i optimized heterogeneous aquifer structures.

[0044] Compared with the prior art, the method has the beneficial effects that:

[0045] The application relates to a heterogeneous aquifer structure intelligent identification method and system fusing resistivity tomography and hydrological observation data, mainly comprising the following steps: training an aquifer structure generation model by using a random simulation method based on lithofacies observation condition data; constructing a lithofacies structure-concentration, water head data dataset based on the aquifer structure generation model and solute transport simulation; constructing a lithofacies structure-apparent resistivity data dataset based on a rock physical relationship equation and forward simulation of the resistivity tomography method; training a substitute model for predicting hydrological observation data and a substitute model for predicting apparent resistivity data respectively; optimizing random variables based on different types of actual observation data by using a data fusion algorithm, and finally obtaining a group of heterogeneous aquifer structures with minimum error between dynamic response observation data and actual observation data. The application fully considers effective information provided by field lithofacies observation data, concentration, water head observation data and resistivity tomography original monitoring data, and realizes scientific fusion of different types of data by using deep learning technology, accelerates the identification process, improves the identification accuracy of the heterogeneous aquifer structure, and provides important technical support for pollutant migration simulation and risk early warning. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed to be used in the embodiments. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0047] Figure 1 It is a flowchart of the heterogeneous aquifer structure intelligent identification method in the embodiments of the present application.

[0048] Figure 2A flowchart of a generation module for a heterogeneous aquifer structure in an embodiment of the present application;

[0049] Figure 3 A schematic diagram of a substitute model for predicting apparent resistivity data in an embodiment of the present application;

[0050] Figure 4 A schematic diagram of a dynamic response data state field corresponding to a target heterogeneous aquifer structure in an embodiment of the present application; (a) is a concentration state field at 6 time points, (b) is a water head field, and (c) is an apparent resistivity field;

[0051] Figure 5 A schematic diagram of a target heterogeneous aquifer structure and identification results in an embodiment of the present application; (a) is a target heterogeneous aquifer structure, (b) is a separate concentration and water head inversion result, (c) is a separate apparent resistivity data inversion result, and (d) is a joint inversion result. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0053] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be understood as the common meanings understood by those skilled in the art to which the embodiments of the present application belong. The terms “first”, “second”, and similar terms used in the embodiments of the present application do not represent any order, number, or importance, but are only used to distinguish different components. The terms “include” or “contain” and similar terms mean that the elements or objects before the terms cover the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms “connect” or “connected” and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms “up”, “down”, “left”, “right”, and the like only represent relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships can also change accordingly.

[0054] Embodiment One

[0055] As known from the background, the following core challenges are currently faced:

[0056] The process of fusing geophysical data needs to repeatedly call geophysical forward model, which is relatively time-consuming and laborious. With the development of deep learning technology, such as attention mechanism, group convolution residual module, etc., it can effectively help to predict geophysical data and realize the rapid and accurate identification of non-homogeneous aquifer structure.

[0057] The embodiment of the application provides a non-homogeneous aquifer structure intelligent identification method fusing resistivity tomography and hydrological observation data, as shown in Figure 1 The method comprises the following steps:

[0058] Based on the lithofacies observation condition data, a non-homogeneous aquifer structure generation model is trained by using a random simulation method;

[0059] Based on the non-homogeneous aquifer structure generation model and the solute transport simulation, a data set of 'lithofacies structure-concentration, water head data' is constructed;

[0060] Based on the lithophysical relationship equation and the forward simulation of the resistivity tomography method, a data set of 'lithofacies structure-apparent resistivity data' is constructed;

[0061] Based on the data set of 'lithofacies structure-concentration, water head data' and the data set of 'lithofacies structure-apparent resistivity data', a substitute model for predicting hydrological observation data and a substitute model for predicting apparent resistivity data are respectively trained;

[0062] Based on different types of actual observation data, a data fusion algorithm is used to optimize random variables, and finally a set of non-homogeneous aquifer structures with minimum error between dynamic response observation data and actual observation data is obtained.

[0063] As an optional implementation, as shown in Figure 2 Based on the lithofacies observation condition data, the non-homogeneous aquifer structure generation model is trained by using a random simulation method, which comprises the following steps:

[0064] Step 1: carry out field investigation of the research area, obtain lithofacies distribution data such as drill hole and outcrop profile, convert the lithofacies distribution data into condition data for random simulation, and set the volume proportion and extension length parameter prior interval of different lithofacies according to the condition data;

[0065] Step 2: based on the Latin hypercube sampling method, N e parameters are extracted in the prior interval determined in step 1, and the parameters are combined with the condition data in step 1 to carry out random simulation, and N e simulated lithofacies are obtained;

[0066] Step 3: the simulated lithofacies obtained in step 2 are scaled to the interval [0, 1], and the non-homogeneous aquifer structure generation model is trained after the data set is arranged.

[0067] The random simulation method in step 1 is completed by GEOST software, and automatic loop calling is realized by writing Python code.

[0068] The generating model in step 3 is Octave Convolution Adversarial Autoencoder (OCAAE) model, and the formula of the last layer activation function is Sigmoid:

[0069]

[0070] The output range of the function is [0, 1], so the lithofacies data obtained in step 2 needs to be scaled to the interval [0, 1]. Taking three lithofacies as an example, the lithofacies indicator data is (0, 1, 2), and the corresponding value after scaling is (0, 0.5, 1).

[0071] After the training of the generating model in step 3 is completed, the high-dimensional aquifer structure can be represented by a low-dimensional random vector, and the size of the random vector can be customized by the user. The larger the size, the more refined the structure adjustment, but it will increase the computational burden, so the user needs to balance the calculation accuracy and the calculation cost, and it is generally recommended to set the size of each direction of the structure to one quarter.

[0072] As an optional implementation, the data set of "lithofacies structure-concentration, water head data" is constructed based on the heterogeneous aquifer structure generating model and solute transport simulation, which includes:

[0073] Step 4: N c random variables are extracted in the standard normal distribution interval based on the Latin hypercube sampling method, and then input into the generating model trained in step 3 to obtain N c lithofacies structures expressed in the range of [0, 1];

[0074] Step 5: The lithofacies structure obtained in step 4 is subjected to the inverse operation of step 3, and is enlarged from the interval [0, 1] to the same lithofacies data as in step 2 to obtain N c random lithofacies structures;

[0075] Step 6: Solute transport simulation is carried out on the lithofacies structure obtained in step 5 to obtain concentration and water head state fields at different times matched with the heterogeneous structure, and different types of response data are normalized to obtain a data set R1 containing N c groups of structure to hydrological response data.

[0076] In step 4, the purpose of extracting N c random variables is to train F1 and F2 in step 9, and generally N cThe larger the training effect is, but it will occupy too many computer GPU resources, users need to combine computer hardware performance self-set.

[0077] As an optional implementation, the data set of "facies structure-apparent resistivity data" is constructed based on the forward simulation of petrophysical relationship equation and resistivity tomography method, including:

[0078] Step 7: Based on the petrophysical relationship equation, the concentration state field at different times obtained in step 6 is converted into the true resistivity field at different times;

[0079] Step 8: The true resistivity field in step 7 is subjected to forward simulation of resistivity tomography method, and the corresponding apparent resistivity data at different times is obtained. After normalization processing of the apparent resistivity data, the data set R2 containing N c facies structure-apparent resistivity original monitoring data is obtained.

[0080] Wherein, the solute transport simulation and the forward simulation of resistivity tomography method in steps 6 and 8 are completed by TOUGHREACT and SimPEG, and in the modeling process, the structure model needs to be extended to the boundary to avoid boundary effect;

[0081] The data normalization formula in steps 6 and 8 is:

[0082]

[0083] In formula (2): And respectively represent the normalized and original values of the i-th type of dynamic observation data on the j-th grid, And respectively represent the maximum and minimum values of the data in the data set;

[0084] The petrophysical relationship equation in step 7 is established based on Archie formula:

[0085]

[0086] In formula (3): σ is the true conductivity of facies, φ is the porosity of facies itself, S w is the water saturation, and m and n respectively represent the cementation coefficient and saturation index. σ w is the conductivity of water phase, and the calculation formula is:

[0087]

[0088] The conversion of concentration state field c to true resistivity field can be realized by formula (3) and formula (4), and the conversion process is automatically realized by writing Python code.

[0089] As an optional implementation, step 9: using R1 and R2 respectively to complete the alternative model F1 for predicting hydrological observation data and the alternative model F2 for predicting apparent resistivity data; as shown in Figure 3 .

[0090] The input of the two alternative models in step 9 regards the lithofacies structure as image data, and the division of the grid is equivalent to the pixels of the image, wherein the output of the hydrological observation data alternative model F1 is image data with the same grid division scheme as the lithofacies structure, and the output of the alternative model F2 for predicting apparent resistivity data is a vector data with a self-definable size. The output result of F1 needs to be extracted on the corresponding observation point through Python code in the process of obtaining the observation data, and the output result of F2 does not need additional processing, and the output size is based on the electrode configuration setting in the resistivity tomography survey;

[0091] The hydrological observation data alternative model F1 in step 9 uses Deep Octave Convolution Residual Network (DOC RN), and the alternative model F2 for predicting apparent resistivity data uses Hybrid Residual Octave-Convolution Network with Squeeze-and-Excitation Attention (HROCN-SE), wherein HROCN-SE is constructed based on discrete convolution dense residual block, group convolution residual block and channel attention mechanism, and is proposed for the first time in order to handle the highly nonlinear relationship between the lithofacies structure and the apparent resistivity data.

[0092] As an optional implementation, the data fusion algorithm is used to optimize the random variables based on different types of actual observation data, and finally a set of non-homogeneous aquifer structures with the minimum error between the dynamic response observation data and the actual observation data is obtained, including:

[0093] Step 10: N i initial random variables are extracted in the standard normal distribution interval based on the Latin hypercube sampling method, and after repeating steps 4-5, a non-homogeneous structure is obtained, which is input into F1 and F2 in step 9 to obtain the concentration, water head state field and apparent resistivity data original monitoring data at the corresponding observation positions;

[0094] Step 11: input the simulated observation data obtained in step 10 and the actual observation data into the data assimilation algorithm to complete the optimization of the N i initial random variables in step 10;

[0095] Step 12: repeat steps 10-11, and after reaching the set number of iterations, the optimized Ni a final random variable;

[0096] Step 13: input the N i random variables obtained in step 12 into the generation model, so as to obtain N i optimized heterogeneous aquifer structures.

[0097] Among them, the Latin hypercube sampling method in steps 2, 4 and 10 is realized by writing Matlab code. Latin hypercube sampling is a structured sampling method that can ensure that the samples extracted in each dimension are representative enough;

[0098] Among them, the data assimilation algorithm in step 11 is Iterative local updating ensemble smoother (ILUES), which can reduce the error between the simulated observation data and the actual observation data by optimizing the N i initial random variables in step 10. It is generally recommended that N i is set to 2000, and the user can adjust it according to the specific task situation;

[0099] This method realizes the scientific integration of resistivity tomography and hydrological observation data, which are two completely different data types in the process of identifying heterogeneous aquifer structures, fully considers the effective information provided by different data, solves the problem of insufficient field observation data or conditional data, and improves the accuracy and computational efficiency of aquifer structure inversion combined with deep learning methods.

[0100] As an optional implementation, in order to test the performance of the proposed method, the specification provides a test case for two-dimensional heterogeneous aquifer structure inversion and identification. This case is a synthetic case, representing an aquifer depth of 20 m, length of 80 m, and containing gravel, coarse sand, and fine sand. This case generates a target heterogeneous aquifer structure based on five virtual borehole data using GEOST, and the true geostatistical parameters of the target heterogeneous aquifer structure are shown in Table 1. Based on this heterogeneous aquifer structure, a tracer test scenario is simulated using TOUGHREACT to generate corresponding water head and concentration distribution fields, and the corresponding apparent resistivity data are calculated based on the SimPEG code. Figure 4 Fig. 1 is a schematic diagram of the dynamic response data state field corresponding to the target heterogeneous aquifer structure; wherein (a) is the concentration state field at 6 time points, (b) is the water head field, and (c) is the apparent resistivity field; 3 hydrological observation wells are selected, and an electrode is arranged every 2 m on the ground surface to obtain observation data, and the ILUES data fusion algorithm is used to invert the heterogeneous aquifer structure based on these observation data;

[0101] In this case:

[0102] In step 2, 20,000 sets of aquifer structure parameters are extracted, and GEOST is automatically called using python code to complete the indicator kriging simulation.

[0103] In step 3, the indicator data of gravel, fine sand and coarse sand are scaled to 0, 0.5 and 1 respectively, and the structure data set is constructed to start the training of the aquifer structure generation model.

[0104] In step 4, 10,000 sets of random variables are randomly extracted to generate 10,000 sets of heterogeneous aquifer structure samples.

[0105] In steps 5-6, the rock facies indicator data is first enlarged and then input into TOUGHREACT to obtain the normalized dynamic observation data set of concentration and water head.

[0106] In step 7, the rock physical relationship equation is converted into the corresponding resistivity state field, with the water saturation set to 1 and the cementation coefficient set to 0.2.

[0107] In step 8, the resistivity tomography method is used to simulate the resistivity field at different times, and the normalized apparent resistivity data and the heterogeneous aquifer structure samples in step 5 are sorted into a dynamic observation data set.

[0108] In step 9, the data set of the two alternative models is 10,000 sets, and 80% of them are randomly extracted as the training set, and the remaining 20% are used as the validation set to complete the training of the two alternative models.

[0109] In step 10, 2,000 sets of random variables are randomly extracted, and steps 4-5 are repeated. The initial random structure obtained is input into the two alternative models constructed in step 9 to obtain different types of observation data at the corresponding observation positions.

[0110] Steps 11 and 12 are the optimization process of the initial random variables.

[0111] In step 13, the iteration number is set to 20 times.

[0112] In the final obtained posterior heterogeneous aquifer set, one structure is randomly extracted and compared with the target heterogeneous aquifer structure as shown in Figure 5 , where (a) is the target heterogeneous aquifer structure, (b) is the separate concentration and water head inversion result, (c) is the separate apparent resistivity data inversion result, and (d) is the joint inversion result. The accuracy identification result of the aquifer structure after separate concentration and water head inversion, separate geophysical data inversion and joint inversion can reach more than 90%.

[0113] Therefore, the non-homogeneous aquifer structure inversion identification method based on data fusion provided by the patent can effectively integrate hydrological observation data and geophysical data to effectively invert and identify the non-homogeneous aquifer structure.

[0114] Table 1. Prior distribution and true value of target non-homogeneous aquifer structure parameters

[0115]

[0116] It should be noted that the method of the embodiments of the present disclosure can be executed by a single device, such as a computer or a server. The method of the embodiments can also be applied to a distributed scenario, and completed by multiple devices cooperating with each other. In this distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiments of the present disclosure, and the multiple devices can interact with each other to complete the method.

[0117] It should be noted that some embodiments of the present disclosure have been described above. Other embodiments are within the scope of the appended claims. In some cases, it should be understood that the size of the serial number of each step in the above-mentioned embodiments does not mean the order of execution. The execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure. The actions or steps recited in the claims can be executed in an order different from that in the above-mentioned embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0118] Embodiment two

[0119] Based on the same inventive concept, corresponding to the method of any of the above embodiments, the present disclosure also provides an intelligent identification system for non-homogeneous aquifer structure based on fusion of resistivity tomography and hydrological observation data, comprising: a first model training module, a first data set construction module, a second data set construction module, a second model training module and an optimization module.

[0120] The first model training module is used to train a non-homogeneous aquifer structure generation model based on lithofacies observation condition data using a random simulation method;

[0121] The first data set construction module is used to construct a data set of "lithofacies structure-concentration, water head data" based on the non-homogeneous aquifer structure generation model and solute transport simulation;

[0122] The second data set construction module is configured to construct a data set of "facies structure-apparent resistivity data" based on a petrophysical relationship equation and forward simulation of resistivity tomography;

[0123] The second model training module is configured to train a substitute model for predicting hydrological observation data and a substitute model for predicting apparent resistivity data based on the data set of "facies structure-concentration, water head data" and the data set of "facies structure-apparent resistivity data", respectively.

[0124] The optimization module is configured to optimize the random variables based on different types of actual observation data using a data fusion algorithm, and finally obtain a set of heterogeneous aquifer structures with minimum error between dynamic response observation data and actual observation data.

[0125] As an optional implementation, the first model training module comprises an extension unit, a simulation unit and a training unit.

[0126] The extension unit is configured to carry out field investigation of a research area, obtain drilling and outcrop profile facies distribution data, convert the facies distribution data into conditional data for random simulation, and set the volume proportion and extension length parameter prior interval of different facies according to the conditional data.

[0127] The simulation unit is configured to extract N e parameters based on a Latin hypercube sampling method within the determined prior interval, and carry out random simulation by combining the parameters with the conditional data to obtain N e simulated facies.

[0128] The training unit is configured to scale the obtained simulated facies to the interval [0, 1], and train the heterogeneous aquifer structure generation model after arranging the data set.

[0129] As an optional implementation, the construction of the data set of "facies structure-concentration, water head data" based on the heterogeneous aquifer structure generation model and solute transport simulation comprises:

[0130] N c random variables are extracted within a standard normal distribution interval based on a Latin hypercube sampling method, and then input into the trained heterogeneous aquifer structure generation model to obtain N c facies structures expressed in the range [0, 1];

[0131] The obtained facies structures are subjected to inverse operation of the trained heterogeneous aquifer structure generation model, and are enlarged from the interval [0, 1] to facies data in the same expression mode as the N e simulated facies to obtain N c random facies structures.

[0132] The obtained lithofacies structure is subjected to solute transport simulation, and concentration, water head state field at different times are obtained, different types of response data are subjected to normalization processing, and N c group data set R1 from structure to hydrological response data.

[0133] As an optional implementation, the data set of 'lithofacies structure-apparent resistivity data' is constructed based on the rock physical relationship equation and the forward simulation of resistivity tomography, which includes:

[0134] The obtained concentration state field at different times is converted into true resistivity field at different times based on the rock physical relationship equation;

[0135] The forward simulation of resistivity tomography is carried out on the true resistivity field, and the corresponding apparent resistivity data at different times are obtained, and after normalization processing of the apparent resistivity data, N c group data set R2 from structure to apparent resistivity original monitoring data is obtained.

[0136] As an optional implementation, the optimization module includes an acquisition unit, an optimization unit, an iteration unit and a generation unit;

[0137] The acquisition unit is used for extracting N i initial random variables in the standard normal distribution interval based on the Latin hypercube sampling method, obtaining N i heterogeneous structures, and inputting the surrogate model for predicting hydrological observation data and the surrogate model for predicting apparent resistivity data to obtain concentration, water head state field and apparent resistivity data original monitoring data at corresponding observation positions respectively;

[0138] The optimization unit is used for inputting the obtained simulation observation data and the actual observation data into the data assimilation algorithm, and completing optimization of N i initial random variables;

[0139] The iteration unit is used for obtaining N i final random variables after optimization when the optimization reaches the set iteration number;

[0140] The generation unit is used for inputting the obtained N i random variables into the generation model to obtain N i optimized heterogeneous aquifer structures.

[0141] The system of the above embodiment is used to realize the intelligent identification method of the heterogeneous aquifer structure fused with resistivity tomography and hydrological observation data in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be described here.

[0142] It should be noted that the above-mentioned identification system is embodied in the form of functional units. The term "module" herein can be realized in the form of software and / or hardware, and is not specifically limited.

[0143] For example, the "module" can be a software program, a hardware circuit, or a combination of both, which realizes the above-mentioned functions. The hardware circuit can include an application specific integrated circuit (ASIC), an electronic circuit, a processor (such as a shared processor, a dedicated processor, or a group processor, etc.) and a memory for executing one or more software or firmware programs, a combination logic circuit, and / or other suitable components supporting the described functions.

[0144] The embodiments of the present disclosure are intended to cover all such alternatives, modifications, and variations of the present disclosure falling within the broadest scope of the appended claims. Accordingly, any and all such modifications, variations, omissions, and equivalents should be considered within the scope of the present disclosure.

Claims

1. A method for intelligent identification of heterogeneous aquifer structure by integrating electrical resistivity tomography and hydrological observation data, characterized in that: The method comprises: Based on the lithofacies observation condition data, a stochastic simulation method is used to train a model for heterogeneous aquifer structure generation; Construct a data set of "lithofacies structure-concentration, hydraulic head data" based on heterogeneous aquifer structure generation model and solute transport simulation; A dataset of "lithofacies structure-apparent resistivity data" was constructed based on rock physics equations and forward simulation of electrical resistivity tomography; Based on the "lithofacial structure-concentration, hydraulic head data" dataset and the "lithofacial structure-apparent resistivity data" dataset, we trained a surrogate model for predicting hydrological observation data and a surrogate model for predicting apparent resistivity data, respectively. Based on different types of actual observation data, a data fusion algorithm is used to optimize random variables, and finally a set of heterogeneous aquifer structures with the smallest error between the dynamic response observation data and the actual observation data is obtained; Based on different types of actual observation data, a data fusion algorithm is used to optimize random variables, and finally a set of heterogeneous aquifer structures with the smallest error between dynamic response observation data and actual observation data are obtained, including: Based on the Latin hypercube sampling method, the data are drawn from the standard normal distribution interval. Initial random variables, get After the input of the alternative model for predicting hydrological observation data and the alternative model for predicting apparent resistivity data, the concentration, head state field and apparent resistivity data original monitoring data of the corresponding observation position are obtained respectively; The obtained simulated observation data and actual observation data are input into the data assimilation algorithm to complete the Optimization of initial random variables; When the optimization reaches the set number of iterations, the optimized final random variables; Will obtain random variables are input into the generative model to obtain An optimized heterogeneous aquifer structure.

2. The method according to claim 1, characterized in that Based on the lithofacies observation data, the stochastic simulation method is used to train the heterogeneous aquifer structure generation model, including: Conduct field surveys in the study area, obtain borehole and outcrop profile lithofacies distribution data, convert them into conditional data for random simulation, and set the volume ratio of different lithofacies and extend the prior interval of length parameters based on the conditional data; Extraction based on Latin hypercube sampling method within a certain prior interval Set parameters and combine the parameters with conditional data to carry out random simulation to obtain simulated lithofacies; The obtained simulated lithofacies were scaled to the interval [0, 1] and organized into a data set to train the heterogeneous aquifer structure generation model.

3. The method according to claim 1, characterized in that The dataset of "lithofacies structure-concentration, hydraulic head data" constructed based on the heterogeneous aquifer structure generation model and solute transport simulation includes: Based on the Latin hypercube sampling method, the data are drawn from the standard normal distribution interval. random variables, which are then input into the trained heterogeneous aquifer structure generation model to obtain A lithofacies structure expressed as data in the range [0, 1]; The obtained lithofacies structure is subjected to the inverse operation of training the heterogeneous aquifer structure generation model, and is enlarged from the [0, 1] interval to the interval The lithofacies data with the same simulated lithofacies expression pattern are obtained. A random facies structure; The solute transport simulation is carried out on the obtained lithofacies structure to obtain the concentration and head state field at different times that match the heterogeneous structure. Different types of response data are normalized separately to obtain the Datasets from structural to hydrological response data .

4. The method according to claim 3, characterized in that The dataset of "lithofacities structure-apparent resistivity data" constructed by forward simulation based on rock physics equations and electrical resistivity tomography includes: The concentration state fields obtained at different times are converted into true resistivity fields at different times based on rock physics equations; The forward simulation of resistivity tomography is carried out on the true resistivity field to obtain the corresponding apparent resistivity data at different times. After normalization of the apparent resistivity data, the A dataset of raw monitoring data from structure to apparent resistivity .

5. A system for intelligently identifying heterogeneous aquifer structures by integrating electrical resistivity tomography and hydrological observation data, the system being used to implement the method according to any one of claims 1 to 4, characterized in that: include: A first model training module, a first data set construction module, a second data set construction module, a second model training module and an optimization module; The first model training module is used to train a heterogeneous aquifer structure generation model using a random simulation method based on lithofacies observation condition data; The first data set construction module is used to construct a data set of "lithofacies structure-concentration, hydraulic head data" based on the heterogeneous aquifer structure generation model and solute transport simulation; The second data set construction module is used to construct a data set of "lithofacies structure-apparent resistivity data" based on rock physics relationship equations and forward simulation of electrical resistivity tomography; The second model training module is used to train a surrogate model for predicting hydrological observation data and a surrogate model for predicting apparent resistivity data based on the "lithofacie structure-concentration, hydraulic head data" data set and the "lithofacie structure-apparent resistivity data" data set, respectively; The optimization module is used to optimize random variables using a data fusion algorithm based on different types of actual observation data, and ultimately obtain a set of heterogeneous aquifer structures with the smallest error between dynamic response observation data and actual observation data.

6. The system according to claim 5, characterized in that The first model training module includes: an extension unit, a simulation unit and a training unit; The extension unit is used to conduct on-site surveys in the study area, obtain drilling and outcrop profile lithofacies distribution data, convert them into conditional data for random simulation, and set the volume ratio of different lithofacies and the prior interval of the extension length parameter based on the conditional data; The simulation unit is used to extract the data based on the Latin hypercube sampling method within a certain priori interval. Set parameters and combine the parameters with conditional data to carry out random simulation to obtain simulated lithofacies; The training unit is used to scale the obtained simulated lithofacies to the interval [0, 1], organize them into a data set, and then train the heterogeneous aquifer structure generation model.

7. The system according to claim 5, characterized in that The optimization module includes: an acquisition unit, an optimization unit, an iteration unit and a generation unit; The acquisition unit is used to extract data within the standard normal distribution interval based on the Latin hypercube sampling method. Initial random variables, get After the input of the alternative model for predicting hydrological observation data and the alternative model for predicting apparent resistivity data, the concentration, head state field and apparent resistivity data original monitoring data of the corresponding observation position are obtained respectively; The optimization unit is used to input the obtained simulated observation data and actual observation data into the data assimilation algorithm to complete the Optimization of initial random variables; The iterative unit is used to obtain the optimized final random variables; The generating unit is used to obtain random variables are input into the generative model to obtain An optimized heterogeneous aquifer structure.

Images