Water drive front intelligent prediction method based on Fourier neural operator and related device

Abstract

The invention belongs to the field of petroleum engineering and underground fluid dynamics, and discloses a water drive front intelligent prediction method based on a Fourier neural operator and a related device.The water drive front intelligent prediction method comprises the steps that water saturation distribution and water drive front position information at the first moment during water drive in a geological reservoir are obtained; and processing the water saturation distribution and the water drive leading edge position information at the first moment through a pre-trained FNO-based water drive leading edge intelligent prediction model to obtain water saturation distribution and water drive leading edge position information at a second moment. According to the method, the FNO is applied to the water drive front intelligent prediction problem in the petroleum engineering field for the first time, the FNO can be used for learning the mapping characteristic between infinite-dimensional function spaces, and front water drive and corresponding saturation field information at different moments can be dynamically predicted; and the limitation of a traditional method and an existing machine learning model in solving the Buck-Leveett equation is overcome.

Description

Technical Field

[0001] The invention belongs to the field of petroleum engineering and underground fluid dynamics, and in particular relates to a water drive front intelligent prediction method based on Fourier neural operator and a related device. Background Art

[0002] In the field of underground fluid dynamics and oil and gas field exploitation, predicting the water flooding front can describe the dynamic behavior of multiphase flow in porous media in this field, which plays a vital role in further optimizing water injection schemes, preventing water in oil wells, and formulating production-increasing measures to increase production and economic benefits. By solving the Buckley-Leverett equation, the position information of the water flooding front and the corresponding saturation field information at each moment can be obtained. Traditional solutions to this equation mainly rely on analytical methods and numerical simulation methods. Due to the high nonlinearity and complexity of partial differential equations (PDEs), analytical methods are usually limited to solving special conditions or simplified models, and the calculation process is cumbersome and difficult to meet the needs of actual engineering applications; numerical simulation methods solve this equation by computer simulation of the flow process in porous media. This method can handle complex problems and has high computational accuracy, but it has a large amount of calculations, high demand for computing resources, and the accuracy and stability of the model have a great impact on the results.

[0003] With the rapid development of machine learning and deep learning technologies, relevant technicians have begun to explore the use of neural networks to predict water drive fronts. Machine learning methods can learn the inherent laws of equations through training data, thereby predicting new flow conditions. However, such methods often require a large amount of training data, and have high requirements on the quality and quantity of data, and have limited generalization capabilities, which restricts their application in practical engineering. Deep learning methods approximate the solution of the Buckley-Leverett equation by constructing complex neural network models, which can handle high-dimensional data and complex nonlinear relationships, and show high accuracy and generalization capabilities. However, existing deep learning methods also face challenges such as large amounts of training data, complex training processes, high computational costs, and strong grid dependence.

[0004] Traditional intelligent prediction methods for waterflooding fronts are overly dependent on grid size, which requires repeated complex calculations when dealing with different parameters or grid size changes, thereby increasing computational costs and time consumption. In addition, although the current mainstream water-to-oil proxy models based on machine learning have improved solution efficiency to a certain extent, they are usually only applicable to mappings between finite-dimensional function spaces and have weak generalization capabilities. When equation parameters or grid size change, these models often need to be retrained or solved, which limits their widespread application in the field of reservoir engineering. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a water drive front intelligent prediction method and related devices based on Fourier Neural Operator (FNO). The present invention can solve the problems of strong grid correlation, high calculation complexity and long time consumption existing in traditional prediction methods.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A water drive front intelligent prediction method based on Fourier neural operator, comprising:

[0008] Obtaining the water saturation distribution and water flooding front position information at the first moment of water flooding in the geological reservoir;

[0009] The water saturation distribution and the water drive front position information at the first moment are processed by a pre-trained FNO-based water drive front intelligent prediction model to obtain the water saturation distribution and the water drive front position information at the second moment, where the second moment is the next moment of the first moment.

[0010] Preferably, the training process of the FNO-based water drive front intelligent prediction model includes: training the Fourier neural operator using data in a pre-established data body to obtain the FNO-based water drive front intelligent prediction model;

[0011] The process of establishing the data body includes:

[0012] The total Darcy velocity in the Buckley-Leverett equation is taken as a random variable, and different values ​​are assigned to the total Darcy velocity. Then, each total Darcy velocity corresponds to a water flooding front model.

[0013] The water injection rate in each water drive front model is used as a standard to distinguish different moments, and a water drive front model at a third moment and a water drive front model at a fourth moment are determined, wherein the fourth moment is the next moment of the third moment;

[0014] Obtaining the water drive front distance and water saturation distribution in the water drive front model at the third moment, and the water drive front distance and water saturation distribution in the water drive front model at the fourth moment;

[0015] Acquire water drive front position information at the third moment according to the water drive front distance in the water drive front model at the third moment, and acquire water drive front position information at the fourth moment according to the water drive front distance in the water drive front model at the fourth moment;

[0016] The water saturation distribution and water flooding front position information corresponding to the third moment are taken as initial conditions, and the water saturation distribution and water flooding front position information corresponding to the fourth moment are taken as true solutions. The initial conditions and true solutions are used to form the data body required by the water flooding front intelligent prediction model based on FNO.

[0017] Preferably, the total Darcy velocity in the Buckley-Leverett equation is taken as a random variable, and different values ​​are assigned to the total Darcy velocity:

[0018] The total Darcy velocity is 10 -5 Up to 10 -3 The total Darcy velocity is uniformly sampled within the range to obtain multiple different values, thereby assigning different values ​​to the total Darcy velocity.

[0019] Preferably, the water injection rate in each water drive front model is used as a standard to distinguish different moments, and when the water drive front model at the third moment and the water drive front model at the fourth moment are determined:

[0020] The water drive front model corresponding to the water injection rate N1 times is determined as the water drive front model at the third moment, and the water drive front model corresponding to the water injection rate N2 times is determined as the water drive front model at the fourth moment, wherein N1≠N2.

[0021] Preferably, the value of N1 is 1, and the value range of N2 is 1.5-25.

[0022] Preferably, the Welge graphical method is used to obtain the water drive front distance and water saturation distribution in the water drive front model at the third moment, and the water drive front distance and water saturation distribution in the water drive front model at the fourth moment.

[0023] Preferably, the water drive front distance in the water drive front model at the third moment is discretized into the water drive front position information at the third moment by using a piecewise linear interpolation method, and the water drive front distance in the water drive front model at the fourth moment is discretized into the water drive front position information at the fourth moment by using a piecewise linear interpolation method.

[0024] The present invention also provides a water drive front intelligent prediction system based on Fourier neural operator, comprising:

[0025] Data acquisition unit: used to obtain the water saturation distribution and water drive front position information at the first moment of water drive in the geological reservoir;

[0026] Prediction unit: used to process the water saturation distribution and water drive front position information at the first moment through the pre-trained FNO-based water drive front intelligent prediction model to obtain the water saturation distribution and water drive front position information at the second moment, where the second moment is the next moment of the first moment.

[0027] The present invention also provides an electronic device, comprising:

[0028] one or more processors;

[0029] a storage device having one or more programs stored thereon;

[0030] When the one or more programs are executed by the one or more processors, the one or more processors implement the water drive front intelligent prediction method based on Fourier neural operator as described above in the present invention.

[0031] The present invention also provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the water drive front intelligent prediction method based on Fourier neural operator as described above in the present invention is implemented.

[0032] The present invention has the following beneficial effects:

[0033] The present invention applies FNO to the intelligent prediction problem of water flooding front in the field of petroleum engineering for the first time. The present invention uses the characteristics of FNO that can learn the mapping between infinite-dimensional function spaces to realize dynamic prediction of front water flooding and corresponding saturation field information at different times, overcoming the limitations of traditional methods and existing machine learning models in solving the Buckley-Leverett equation. Through the method of the present invention, efficient and accurate intelligent prediction of water flooding front can be achieved under the premise of ensuring controllable computing costs, providing strong technical support for reservoir engineers, helping them to reasonably adjust development plans to achieve economical and efficient oil and gas resource exploitation. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a flow chart of the water drive front intelligent prediction method based on Fourier neural operator.

[0035] FIG. 2( a ) is a diagram of T in an embodiment of the present invention. i The water drive front position information and the corresponding water saturation distribution diagram at the moment; FIG2(b) is the actual T in the embodiment of the present invention. i+1 FIG2(c) is a diagram showing the water flooding front position information and water saturation distribution diagram at the time T predicted by the FNO-based water flooding front intelligent prediction method in an embodiment of the present invention. i+1 The water drive front position information and water saturation distribution diagram at the moment; FIG2(d) is a diagram of T in the embodiment of the present invention. i+1 A visual comparison chart of the actual water saturation distribution and the predicted water saturation distribution at each moment. DETAILED DESCRIPTION

[0036] In order to more clearly introduce the cases and methods of the present invention and to show the similarities and differences between the method and the prior art solutions, the established model and the drawings required for use in the specific embodiments are briefly described below. It is hereby noted that the drawings involved below are all implementation methods of the present invention, and ordinary technicians in this field can obtain other drawings based on the following drawings without any creative work.

[0037] In order to efficiently predict the water drive front position information and the corresponding water saturation field, and solve the problems of strong grid dependence, high computational cost and weak generalization faced by traditional solution methods and existing proxy models in solving the Buckley-Leverett equation, the present invention combines the high efficiency of Fourier transform and the powerful learning ability of neural networks, and proposes a water drive front intelligent prediction method and related devices based on Fourier Neural Operator (FNO). The present invention can improve the prediction accuracy, efficiency and generalization ability, and provide a more reliable and efficient solution for oil and gas field exploitation and underground fluid dynamics under the premise of controllable budget cost.

[0038] The present invention adapts the water-to-oil model to the network structure of the Fourier neural operator, introduces the Welge graphical method, uses the interpolation grid water drive front distance, and converts the initial time T i The water flooding front position information and the corresponding water saturation are used as the initial condition a(x) of FNO, and the next moment T i+1 The water drive front position information and the corresponding water saturation are taken as the true solution u(x) of FNO, which is used as the data body of a new method for intelligent prediction of water drive front based on the Fourier neural operator architecture.

[0039] Specifically, refer to Figure 1 The water drive front intelligent prediction method based on Fourier neural operator of the present invention comprises the following steps:

[0040] Step 1: Since the actual geological reservoir is relatively complex and contains many uncertain factors, the total Darcy velocity ν in the Buckley-Leverett equation is d The parameter is a random variable, so that the parameter v d In 10 -5 Up to 10 -3 Sampling uniformly in the range, generating N ν d value, each v d Substitute them into the Buckley-Leverett equation respectively, then each ν d Each of them will correspond to a water drive front model, so N water drive front models will be generated. Further, the Buckley-Leverett equation is expressed as:

[0041]

[0042] S(t,x)| t＝0 =S wi #(2)

[0043] S(t,x)| x＝0 =1-S or #(3)

[0044] in, is the total Darcy velocity, the water injection rate q(x) remains constant, the porosity φ(x) is uniformly distributed in space, S w is water saturation, t is time, f w ′(S) is the derivative of the split flow function.

[0045] Step 2: Use the water injection rate q(x) in the water flooding front model in step 1 as the standard for distinguishing different moments: Specifically, the N water flooding front models corresponding to the original water injection rate q(x) (i.e., 1 times the water injection rate q(x)) are limited to T i The water flooding front model at the time (i.e., the third time) defines the N water flooding front models corresponding to the n-fold water injection rate q(x) as T i+1 The water drive front model at the fourth moment is: N1=1, N2=n×N2, and the value range of n is 1.5-25.

[0046] Step 3: Use Welge's graphical method to obtain T i The water drive front distance {L1, L2, ..., L N} and water saturation distribution {S w1 , S w2 ,…,S wN} and T i+1 The water drive front distance {L1, L2, ..., L N} and water saturation distribution {S w1 , S w2 ,…,S wN}.

[0047] Step 4: Use piecewise linear interpolation to interpolate the T obtained in step 3 i The water front distance {L1, L2, ..., L N} and T i+1 The water front distance {L1, L2, ..., L N} are discretized into T i The water drive front position information at the time {X sw1 , X sw2 ,…,X swN} and T i+1The water drive front position information at the time {X sw1 , X sw2 ,…,X swN}.

[0048] Step 5: Get T from step 3 to step 4 respectively i Moment and T i+1 The N water saturation distributions of the water flooding front model at time {S w1 , S w2 ,…,S wN} and water drive front position information {X sw1 , X sw2 ,…,X swN}, T i The water saturation distribution and water flooding front position information at the time are used as the initial condition a(x) in the data body required by the FNO-based water flooding front intelligent prediction model, as shown in Figure 2(a). i+1 The water saturation distribution and water flooding front position information at the moment are used as the real solution u(x) in the data volume required by the FNO-based water flooding front intelligent prediction model, as shown in Figure 2(b).

[0049] Step 6: Divide the data obtained in steps 4 and 5 (initial condition a(x), true solution u(x)) into training set, test set and evaluation set in a ratio of 7:2:1. Furthermore, the number of samples in the training set is 7 / 10N, the number of samples in the test set is 2 / 10N, and the number of samples in the evaluation set is 1 / 10N. The FNO is trained using the training set, and the water drive front intelligent prediction model based on FNO is obtained. Furthermore, Represents the nuclear integral operator in FNO:

[0050]

[0051] The Fourier transform expressions involved are as follows:

[0052]

[0053] Furthermore, the input data set is filtered out of high-order modes in the Fourier space of FNO, and low-order modes are retained to remove redundancy and improve the efficiency of solving the equation. Then, the data is restored to the original dimensional space through the inverse Fourier transform. The inverse Fourier transform expression involved is as follows:

[0054]

[0055] Furthermore, the kernel integral operator expression and the Fourier transform expression can be deduced:

[0056]

[0057] From this, the expression of the Fourier integral operator used can be obtained as follows:

[0058]

[0059] The FNO used solves the integral term in operator learning, and the expression of the established model can be obtained by combining the iterative relationship:

[0060]

[0061] in, is the spatial domain of PDE, x∈D is a point in the spatial domain, represents the coefficient function of the input, represents the target solution function, D j indicates (a j ,u j ), φ represents the parameter of the kernel network k, t=0,…,T represents the time step, σ represents the activation function, and is the Fourier transform and its inverse transform, W represents a linear transformation, v j For j = 0, 1, ... T-1, it is a series of functions, and each function takes a value in v t to v t+1 is defined as the nonlocal kernel integral operator and a local nonlinear activation function σ.

[0062] After the four transformation operations of the above Fourier layer, the processed training data is combined with the activation function and sent out of the Fourier space. The activation function used here is Gelu, which is defined as:

[0063]

[0064] Where X is a Gaussian random variable with zero mean and unit variance that is used to model the input data, and P(X≤x) is the probability that the input can be less than or equal to a given value.

[0065] Furthermore, the Gaussian error function can be used for calculation, which is expressed as:

[0066]

[0067] Where tanh(x) is the hyperbolic tangent function.

[0068] The solution model trained after m iterations was tested. After testing with 2 / 10N samples at different times, an intelligent prediction model for the water drive front based on FNO was successfully established. Subsequently, 1 / 10N evaluation samples can be used for evaluation and prediction at different times to obtain a more accurate water drive front position and corresponding water saturation distribution at the next moment, as shown in Figure 2(c). The error between the prediction result and the actual distribution is small. After the model is established, only the initial conditions (water drive front position information and corresponding water saturation distribution at time Ti) as shown in Figure 2(a) need to be input, and the model can be used to quickly predict the water drive front position information and corresponding water saturation field at the next moment Ti+1 as shown in Figure 2(c), thereby realizing dynamic prediction of the water drive front over the entire life cycle. Furthermore, the evaluation equation used is:

[0069]

[0070]

[0071] Where n represents the number of samples, represents the true solution, Represents the predicted value. At the same time, the root mean square error RMSE and determination coefficient R2 are used to evaluate the accuracy of the proposed water flooding front intelligent prediction model, with RMSE as the loss function and R2 as the goodness of fit;

[0072] When performing intelligent prediction of the water flooding front, the water saturation distribution and water flooding front position information at time t0 (i.e., the first moment) during water flooding in the geological reservoir are input into the pre-trained FNO-based water flooding front intelligent prediction model, and the FNO-based water flooding front intelligent prediction model is used for data processing, and finally the water saturation distribution and water flooding front position information at time t1 (i.e., the second moment) are output, that is, the intelligent prediction of the water flooding front at time t1 is realized. Among them, time t1 is the next moment of time t0.

[0073] Example

[0074] The water flooding front intelligent prediction method based on Fourier neural operator in this embodiment includes the following process: converting the total Darcy velocity ν in the oil-water two-phase Buckley-Leverett equation into d The parameter is a random variable such that the parameter ν d In 10 -5 Up to 10 -3 The range is uniformly sampled to obtain 2048 Buckley-Leverett equations with 1 times the injection rate q(x), i.e., the water flooding front model at time t0;

[0075] With the help of MATLAB platform, the Welge graphical method was used to obtain 2048 groups of water saturation distribution (S w The range is 0.1 to 0.9) and the corresponding waterflood front distance (L is 100 m);

[0076] The water drive front position information (X) corresponding to the 2048 water drive front distances at time t0 is obtained by linear interpolation method. sw =8192), taking 2048 sets of water saturation and corresponding water flooding front position information at time t0 as the initial condition a(x), that is, the size of a(x) is [2048, 8192];

[0077] The water flooding front model at 1.5 times the water injection rate q(x) is used as the water flooding front model at time t1, and the 2048 groups of water saturations and corresponding water flooding front distances (L = 100m) corresponding to the water flooding front model at time t1 are obtained by the Welge graphical method;

[0078] The water drive front position information (X) corresponding to the 2048 water drive front distances at time t1 is obtained by linear interpolation method. sw =8192), and the 2048 sets of water saturation distribution and water flooding front position information at time t1 are taken as the true solution u(x), that is, the size of u(x) is also [2048, 8192];

[0079] The obtained initial conditions a(x) and true solution u(x) are used as the data body for establishing the water flooding front intelligent prediction model based on FNO. 1500 sets of data are taken from the data body for model training, 300 sets of data are taken for model testing, and 200 sets of data are taken for model evaluation.

[0080] After the model is trained and tested, you can enter the model evaluation and prediction phase, as follows: Figure 2(a)-Figure 2(d) As shown, by inputting the water saturation distribution and the corresponding water flooding front position information a(x) at the current time t0, the water saturation distribution and the water flooding front position information at the time t1 can be quickly output.

[0081] Visualization model output results, Figure 2 (a) is the water saturation distribution at time t0 and its corresponding water drive front position information, Figure 2 (b) is the actual water saturation distribution at time t1 and its corresponding water drive front position information, Figure 2 (c) is the water saturation distribution at time t1 based on the FNO water drive front intelligent prediction and its corresponding water drive front position information, Figure 2 (d) is an intuitive comparison of the actual water saturation distribution and the predicted water saturation distribution at time t1. Figure 2(a)-Figure 2(d)It can be seen that the accuracy of using the established model to predict water saturation at different times is as high as 0.998, which is enough to show that the proposed new method has the potential to efficiently, accurately and dynamically predict the distribution of water saturation in oil reservoirs.

[0082] The present invention also provides a system based on the water flooding front intelligent prediction method as described above, the system comprising:

[0083] Data acquisition unit: used to obtain the water saturation distribution and water drive front position information at the first moment of water drive in the geological reservoir;

[0084] Prediction unit: used to process the water saturation distribution and water drive front position information at the first moment through the pre-trained FNO-based water drive front intelligent prediction model to obtain the water saturation distribution and water drive front position information at the second moment, where the second moment is the next moment of the first moment.

[0085] The embodiments of the present invention also provide corresponding electronic devices and computer-readable storage media for implementing the solutions provided by the embodiments of the present invention.

[0086] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes, so that the device executes the water drive front intelligent prediction method described in any embodiment of the present application.

[0087] The storage medium stores a computer program, wherein the computer program, when executed by the processor, implements the water flooding front intelligent prediction method described in any embodiment of the present application.

[0088] It can be seen from the above scheme that the present invention applies FNO to the intelligent prediction problem of water drive front in the field of petroleum engineering for the first time. First, a data generation method adapted to the neural operator network structure is invented, and the water saturation distribution data is obtained in combination with the Welge graphical method; secondly, the water saturation position information is discretized, and the water drive front and the corresponding water saturation distribution at the initial moment and the water drive front and the corresponding water saturation distribution at the next moment are used as training data. The model is trained using the training data and tested and evaluated, and an intelligent prediction model for water drive front based on FNO is established. The model inputs the water drive front position information and the corresponding water saturation distribution at the current moment, and can realize accurate dynamic prediction of the water drive front position information and the corresponding water saturation field at the next moment. Compared with the traditional PDE equation solving proxy model and machine learning method, this method still describes the dynamic distribution of the reservoir at the next moment with a higher prediction accuracy after modifying the parameters or grid number of the equation. It has stronger generalization than the previous machine learning method, and has a higher solution efficiency. It can realize accurate prediction under the condition of controllable computing cost, and provides a new idea and approach for the development and design of oil and gas fields.

[0089] The benefit of the present invention is that the constructed FNO-based water drive front intelligent prediction model uses neural networks to approximate infinite-dimensional operators to describe the dynamic behavior of the reservoir system, and can directly learn the mapping from arbitrary function parameters to solutions, and use fast Fourier transform to capture the frequency domain characteristics of the function, convert the time domain signal to the frequency domain, and perform nonlinear transformation based on this. It has more powerful expression ability than the traditional water drive front prediction method based on machine learning. The present invention has absolute advantages such as high solution accuracy, grid independence, high prediction efficiency, strong expression ability and generalization ability, and has important theoretical guidance significance and engineering application value for intelligent oil and gas field development decision-making in the field of underground fluid dynamics and petroleum engineering.

[0090] Therefore, the present invention has overcome the following technical problems faced by the prior art: the traditional method for predicting the water drive front solves the time and space domain through gridding, discretizes the partial differential equation, and uses an iterative algorithm to extract the numerical solution of the grid nodes. The grid has strong correlation, and the dynamic prediction of the water drive front is highly complex and time-consuming; the pure data-driven method ignores the physical laws contained in the Buckley-Leverett equation when predicting the water drive front, requires a large number of training samples, is prone to overfitting, and has weak generalization; the physical information neural network algorithm embeds the physical mechanism into the neural network, and uses automatic differentiation to optimize the network parameters, which can efficiently solve the PDE, but can only solve a specific instance in a class of PDE. When the equation parameters change, the model needs to be retrained. In actual reservoir engineering applications, due to the high complexity of the training process and the high computational cost, this technology faces certain limitations.

[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A water flooding front intelligent prediction method based on Fourier neural operator, characterized in that: include: Obtaining the water saturation distribution and water flooding front position information at the first moment of water flooding in the geological reservoir; The water saturation distribution and the water drive front position information at the first moment are processed by a pre-trained FNO-based water drive front intelligent prediction model to obtain the water saturation distribution and the water drive front position information at the second moment, where the second moment is the next moment of the first moment.

2. The method for intelligent prediction of water flooding front based on Fourier neural operator according to claim 1 is characterized in that: The training process of the FNO-based water drive front intelligent prediction model includes: using data in a pre-established data body to train the Fourier neural operator to obtain the FNO-based water drive front intelligent prediction model; The process of establishing the data body includes: The total Darcy velocity in the Buckley-Leverett equation is taken as a random variable, and different values ​​are assigned to the total Darcy velocity. Then, each total Darcy velocity corresponds to a water flooding front model. The water injection rate in each water drive front model is used as a standard to distinguish different moments, and a water drive front model at a third moment and a water drive front model at a fourth moment are determined, wherein the fourth moment is the next moment of the third moment; Obtaining the water drive front distance and water saturation distribution in the water drive front model at the third moment, and the water drive front distance and water saturation distribution in the water drive front model at the fourth moment; Acquire water drive front position information at the third moment according to the water drive front distance in the water drive front model at the third moment, and acquire water drive front position information at the fourth moment according to the water drive front distance in the water drive front model at the fourth moment; The water saturation distribution and water flooding front position information corresponding to the third moment are taken as initial conditions, and the water saturation distribution and water flooding front position information corresponding to the fourth moment are taken as true solutions. The initial conditions and true solutions are used to form the data body required by the water flooding front intelligent prediction model based on FNO.

3. The method for intelligent prediction of water flooding front based on Fourier neural operator according to claim 2 is characterized in that: When the total Darcy velocity in the Buckley-Leverett equation is taken as a random variable and different values ​​are assigned to the total Darcy velocity: The total Darcy velocity is 10 -5 Up to 10 -3 The total Darcy velocity is uniformly sampled within the range to obtain multiple different values, thereby assigning different values ​​to the total Darcy velocity.

4. The method for intelligent prediction of water flooding front based on Fourier neural operator according to claim 2 is characterized in that: When the water injection rate in each water drive front model is used as the standard to distinguish different moments, the water drive front model at the third moment and the water drive front model at the fourth moment are determined: The water drive front model corresponding to the water injection rate N1 times is determined as the water drive front model at the third moment, and the water drive front model corresponding to the water injection rate N2 times is determined as the water drive front model at the fourth moment, wherein N1≠N2.

5. The method for intelligent prediction of water flooding front based on Fourier neural operator according to claim 4 is characterized in that: The value of N1 is 1, and the value of N2 ranges from 1.5 to 25.

6. The method for intelligent prediction of water flooding front based on Fourier neural operator according to claim 2 is characterized in that: The Welge graphical method is used to obtain the water drive front distance and water saturation distribution in the water drive front model at the third moment, and the water drive front distance and water saturation distribution in the water drive front model at the fourth moment.

7. The method for intelligent prediction of water flooding front based on Fourier neural operator according to claim 2 is characterized in that: The water drive front distance in the water drive front model at the third moment is discretized into the water drive front position information at the third moment by piecewise linear interpolation method, and the water drive front distance in the water drive front model at the fourth moment is discretized into the water drive front position information at the fourth moment by piecewise linear interpolation method.

8. A water drive front intelligent prediction system based on Fourier neural operator, characterized in that: include: Data acquisition unit: used to obtain the water saturation distribution and water drive front position information at the first moment of water drive in the geological reservoir; Prediction unit: used to process the water saturation distribution and water drive front position information at the first moment through the pre-trained FNO-based water drive front intelligent prediction model to obtain the water saturation distribution and water drive front position information at the second moment, where the second moment is the next moment of the first moment.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the water drive front intelligent prediction method based on Fourier neural operator as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the water drive front intelligent prediction method based on Fourier neural operator as described in any one of claims 1 to 7 is implemented.

Images