Hydrofracture earthquake possibility prediction method

Abstract

The invention provides a hydrofracture earthquake possibility prediction method, and relates to the technical field of hydrofracture, and the method comprises the steps: constructing a physical constraint depth operator network model combined with an importance sampling method; collecting and processing data in different scenes as training data; performing iterative training on the physical constraint depth operator network model to obtain a physical constraint depth operator network model weight of an optimal step; and initializing the physical constraint depth operator network model, and obtaining a hydrofracture earthquake possibility prediction result in the target scene according to the target hydrofracture earthquake prediction task data. According to the method, importance sampling is fused into the depth operator network, and sampling distribution is adaptively changed by calculating the importance of each function in a function set, so that model convergence is accelerated, model stability is improved, the problems of high training difficulty, low convergence speed and low solving precision in a non-label data scene are solved, and the method is suitable for large-scale popularization and application. And the possibility of the hydrofracture earthquake can be predicted more accurately.

Description

Technical Field

[0001] The invention belongs to the technical field of hydraulic fracturing, and in particular relates to a method for predicting the possibility of hydraulic fracturing earthquakes. Background Art

[0002] Hydraulic fracturing technology has become a standard method for the energy industry to exploit deep gas, mineral and heat storage. Hydraulic fracturing is accompanied by the process of injecting fracturing fluid into the deep ground to form fracture zones, establishing fluid channels to exploit oil and gas energy, and reinjecting wastewater. Due to the existence of weak areas such as natural fracture areas or faults underground, the above industrial activities may cause obstruction of underground mining and even induce earthquakes. Most induced earthquakes related to fluid injection occur in the basement rock, 1 to 4 kilometers deeper than the target formation. The instability problems caused by these fluid injections will seriously hinder the efficient implementation of energy extraction activities, and also pose a threat to safe production and the lives of surrounding residents. Therefore, it is crucial to sort out the stability problems caused by the fluid injection process.

[0003] The possibility of fluid injection-induced earthquakes during hydraulic fracturing activities can be represented by the following mathematical modeling:

[0004] in, Relative to the reference stress rate The seismicity rate of the steady-state seismicity rate, Indicates time, represents the Coulomb stress rate, The characteristic time scale is related to the physical properties of the lithosphere (such as creep behavior or rheological properties) and the external stress loading rate that triggers earthquakes. Solving this ordinary differential equation can predict the seismic activity rate. However, the Coulomb stress rate in the above formula is It needs to be calculated and estimated through spline curve fitting and solver. There is no explicit expression, so the equation has no analytical solution and can only be approximated by numerical solution method.

[0005] Traditional numerical solution methods include the finite difference method (FDM), the finite element method (FEM), and the finite volume method (FVM). These methods discretize the equations so that the state of the fluid can be approximately solved in space and time. For example, the finite difference method is suitable for some simple geometric structures by approximating differential operations at discrete nodes, but it is not flexible enough in complex fractures of hydraulic fracturing. Although the finite element method is suitable for dealing with complex geological structures and boundary conditions, it is computationally expensive and not suitable for real-time simulation. The finite volume rule is popular for its good description of conservation equations, but it lacks accuracy when dealing with highly nonlinear conditions. Although traditional numerical solution methods are quite mature, they still face the problems of long calculation time and high resource consumption.

[0006] In recent years, the progress of deep learning technology has provided new ideas for solving PDEs. Among them, the Deep Operator Network (DeepONet) is an innovative deep learning model for fast solving of partial differential equations. The core of the deep operator network is that it can learn the mapping relationship from input to output of partial differential equations, that is, it can directly obtain the distribution of solutions through training without the need for discrete solutions point by point. By inputting initial conditions, boundary conditions, etc., the deep operator network can directly predict the pressure distribution and fluid flow pattern in the entire area, greatly shortening the calculation time. For the hydraulic fracturing process, the application advantages of the deep operator network are particularly obvious: after training, it can realize real-time prediction of fluid pressure and flow conditions in the fracture network. This enables engineers to adjust injection parameters in real time according to the model results, optimize the fracturing process, and reduce the consumption of computing resources.

[0007] The subsequently proposed physics-informed DeepONet adds a partial differential equation loss term to the loss function of the deep operator network, which greatly reduces or even completely avoids the reliance on labeled training data, further expanding the application scenarios of the deep operator network. For the sake of simplicity, the "deep operator network" in the following text refers to the deep operator network that incorporates physics constraints.

[0008] Although deep operator networks have to some extent avoided the problems of traditional numerical solvers, such as long calculation time, discretized grids, and difficulty in solving inverse problems, deep operator networks have the problems of high training difficulty, slow convergence speed, and low solution accuracy for some complex partial differential equation problems, problems with little labeled data, and problems with no labeled data. In the hydraulic fracturing scenario, the problem of seismic activity rate prediction is also one of them. Summary of the invention

[0009] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting the possibility of hydraulic fracturing earthquakes, which solves the problems of high training difficulty, slow convergence speed and low solution accuracy of deep operator networks in unlabeled data scenarios, and thus makes a more accurate prediction of the possibility of hydraulic fracturing earthquakes.

[0010] In order to achieve the above objectives, the technical solution adopted by the present invention is: a method for predicting the possibility of hydraulic fracturing earthquakes, comprising the following steps: S1. Construct a physical constraint deep operator network model PDIS combined with importance sampling method; S2, collect and process the data of the seismicity rate mathematical model under different scenarios as training data; S3. Using the training data and partial differential equation constraints, the physical constraint deep operator network model PDIS is iteratively trained to obtain the optimal step physical constraint deep operator network model PDIS weight; S4. Initialize the physical constraint deep operator network model PDIS using the weights obtained in step S3, and obtain the hydraulic fracturing earthquake possibility prediction result under the target scenario based on the target hydraulic fracturing earthquake prediction task data.

[0011] The beneficial effects of the present invention are as follows: the present invention integrates importance sampling into the deep operator network by constructing a physically constrained deep operator network model PDIS, calculates the importance of each function in a function set, and adaptively changes the sampling distribution, thereby accelerating model convergence and improving model stability, solving the problems of deep operator networks being difficult to train, slow convergence, and low solution accuracy in unlabeled data scenarios, thereby making a more accurate prediction of the possibility of hydraulic fracturing earthquakes.

[0012] Furthermore, the expression of the bias reweighted loss function of the physical constraint deep operator network model PDIS is as follows:

[0013]

[0014]

[0015] in, represents the bias reweighted loss function of the physical constraint deep operator network model PDIS, Represents the function set of the current batch, which is obtained by the importance sampling method of the function at the beginning of a training iteration. i Indicates i functions, Indicates i The sampling weights corresponding to the functions are: represents the loss function of the physical constraint deep operator network, Represents a function dataset N Middle i functions, represents the coordinate set of the collocation points input by the backbone network, represents the weight parameters of the physical constraint deep operator network model, , and They represent the partial differential equation loss, initial condition loss, and boundary condition loss in the physical constraint method, respectively. and All represent weights, Indicates i The original sampling probability of the function, represents the sampling probability of the function, Represents a function dataset.

[0016] The above further scheme has the beneficial effect that the bias reweighted loss function is used to implement the physical constraints of the partial differential equations, while preventing convergence deviations caused by changes in the sampling distribution. While changing the sampling distribution and accelerating the convergence of training, the bias reweighted loss function ensures that the convergence direction remains unchanged, which is an essential component for the correct implementation of the importance sampling method.

[0017] Furthermore, the training data includes Coulomb stress rates under different scenarios. The data, irrelevant constant parameters and initial boundary conditions under different scenarios, among which the Coulomb stress rate The data consist of a discrete set of points representing the Coulomb stress rate and a spline curve fit.

[0018] The beneficial effect of the above further scheme is that in the earthquake activity rate prediction scenario, sufficient Coulomb stress rate is obtained The data includes discrete point sets and spline curve fitting representation of Coulomb stress rate, which ensures the convergence stability of the training results of the physical constraint deep operator network model PDIS and the generalization ability of the model.

[0019] Furthermore, step S3 includes the following steps: S301, initializing the bias weight parameters of the physical constraint deep operator network model PDIS to generate a set of sampling points evenly distributed in the time dimension; S302. At the beginning of an iteration, enter the Coulomb stress rate The discrete point set and sampling point set are based on the output of the physical constraint deep operator network model PDIS and the Coulomb stress rate By calculating each group of Coulomb stress rates one by one The sum of partial differential equation errors of the data on the sampling point set gives each group of Coulomb stress rates The sampling probability of the function ; S303, batch sampling is performed according to the function sampling probability, and the sampled Coulomb stress rate The data and the randomly generated time coordinate sequence are used as the input of the branch network and the trunk network respectively, and the output of the physical constraint deep operator network model PDIS is obtained. The output of the physical constraint deep operator network model PDIS is the prediction result of the possibility of hydraulic fracturing earthquake under the target scenario; S304, based on the output of the physical constraint deep operator network model PDIS, calculate the deviation reweighted loss, and perform gradient optimization to adjust the physical constraint deep operator network model PDIS parameters; S305. Repeat steps S301 to S304 until the iteration ends and the weight of the physical constraint deep operator network model PDIS of the optimal step is retained, wherein the physical constraint deep operator network model PDIS includes a trunk network and a branch network.

[0020] The beneficial effect of the above further solution is that in step S03, the physical constraint deep operator network model PDIS calculates the Coulomb stress rate under different scenarios The importance of discrete point sets and adaptively changing sampling distribution are used to accelerate model convergence and improve the stability of hydraulic fracturing seismic activity rate prediction results. At the same time, the accuracy of the network model in predicting the possibility of hydraulic fracturing earthquakes is also improved.

[0021] Furthermore, the expression of the function sampling probability in step S302 is as follows:

[0022]

[0023] in, represents the sampling probability of the function, S represents the set of sampling points, The input function is hour PDE loss of position, The input function is hour PDE loss of position, and Represents the function data set respectively N Middle n Function and i functions, Represents the uniform time coordinate sequence within the target time interval. k time coordinates, represents the weight parameters of the physical constraint deep operator network model, represents the output of the hydraulic fracturing earthquake possibility prediction result under the target scenario, represents the characteristic time scale, represents the Coulomb stress rate.

[0024] The beneficial effect of the above further scheme is: by calculating the sum of the partial differential equation losses of each input function (i.e., the discretized representation of the Coulomb stress rate in different scenarios) on a uniform grid, the PDIS weight of the physically constrained deep operator network model approximates the importance distribution of each function without adding too many additional calculation steps, which facilitates subsequent function sampling.

[0025] Furthermore, step S4 includes the following steps: S401, using the weight of the physical constraint deep operator network model PDIS of the optimal step, initializing the physical constraint deep operator network model PDIS; S402. Obtain the Coulomb stress rate in the target scenario The discrete point set is represented as the input of the branch network; S403, randomly generate a time coordinate sequence of the target prediction time interval as the input of the backbone network; S404, performing forward propagation according to the input results of step S402 and step S403; S405. According to the forward propagation result, the outputs of the main network and the branch network are subjected to inner product operation to obtain the hydraulic fracturing earthquake possibility prediction result under the target scenario.

[0026] The beneficial effect of the above further scheme is that by loading the trained physical constraint deep operator network model PDIS, the present invention can output the seismic activity rate prediction result for a given coordinate according to any input function (discretized form of Coulomb stress rate under different physical scenarios), thereby realizing the prediction of hydraulic fracturing earthquakes. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Schematic diagram of the PDIS structure of the physical constrained deep operator network model.

[0028] Figure 2 Schematic diagram of the importance sampling process.

[0029] Figure 3 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0030] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0031] Example The present invention is based on a hydraulic fracturing earthquake possibility prediction method based on a physics-informed deep operator network with importance sampling (PDIS), which mainly predicts the possibility of hydraulic fracturing activities inducing earthquakes through PDIS. Figure 3As shown, the present invention provides a method for predicting the possibility of hydraulic fracturing earthquakes, and the implementation method thereof is as follows: S1. Construct a physical constraint deep operator network model PDIS combined with importance sampling method; In this embodiment, Figure 1 As shown ( Figure 1 middle, Indicates i Function , x Represents the input coordinates; b 1 to b p For branch network p outputs, t 1 to t p For backbone network p output, the role of the two is to obtain the final network through the inner product of the input function , enter coordinates x The predicted value under loss function M Represents the function set of the current batch, i 1 to M The function subscript of represents the sampling weight, X express x A collection of represents the weight parameters of the physical constraint deep operator network model, and Both represent weights), the physical constraint deep operator network model PDIS includes: The backbone network is used to obtain the time position of the seismicity rate based on the time coordinate sequence of the randomly generated target prediction time interval; A branch network, used for receiving a discrete point set representation of the Coulomb stress rate under the acquired target scenario; The output layer is used to perform inner product operations on the outputs of the trunk network and the branch network to obtain the prediction results of the possibility of hydraulic fracturing earthquakes in the target scenario.

[0032] In this embodiment, the physical constraint deep operator network model PDIS is a physical constraint deep operator model combined with the importance sampling method, and its purpose is to fit the solution operator of a class of partial differential equations during the training process, so as to achieve the purpose of quickly solving this class of partial differential equations. In the scenario of the present invention, the purpose of the physical constraint deep operator network model PDIS is to solve the partial differential equation of the fluid injection induced seismic activity rate in the hydraulic fracturing scenario, so as to predict the change of the seismic activity rate over time under different Coulomb stress rate curves. Therefore, it is urgent to propose a new hydraulic fracturing earthquake possibility prediction method, which avoids the problems of large computing resource consumption and limited grid of traditional numerical solution methods, and further improves the convergence speed of deep learning in the training process and the accuracy of solving partial differential equations, so as to more efficiently and accurately predict the hydraulic fracturing process, avoid instability problems to the greatest extent, and ensure the safety and efficiency of energy extraction activities.

[0033] In this embodiment, the physical constraint deep operator network model PDIS includes: a backbone network, a branch network and a deviation reweighted loss function. Among them, the backbone network is usually composed of a fully connected network, and its input is the coordinate position of a matching point (i.e., the time coordinate sequence of the target time interval), which is used to define the position of the objective function (i.e., the seismic activity rate); the branch network can be composed of a fully connected layer, a convolutional layer or other network structure, and its input is a discrete sampling point of a function (i.e., a discrete point set of Coulomb stress rate), usually a vector, and the output of the backbone network and the branch network are inner-producted to finally obtain the predicted value of the objective function under the current input function and input coordinates. The deviation reweighted loss function is used to implement the physical constraints of the partial differential equation, while preventing convergence deviations caused by changes in the sampling distribution.

[0034] In this embodiment, in a training iteration process, the importance sampling probability of each function in the function set is first calculated, and then the function importance sampling is performed to obtain the function batch of the current iteration, and then the coordinate batch is randomly sampled. Finally, the sampling data of both are used to train the network, calculate the deviation reweighted loss, and optimize the network model parameters.

[0035] Function sampling probability: In the physical constraint deep operator network model PDIS, the function sampling probability The ratio of the sum of the loss function values ​​of the sampling point set in its definition domain to the sum of all functions:

[0036] in, S represents the set of sampling points, generated by uniform sampling over the entire domain, represents a function dataset, The input function is expressed as hour The partial differential equation loss of the position is the mean square error between the actual calculation result of the partial differential equation expression and the target value. Through the above process, the present invention can evaluate the error size of each function in the function set, so that important functions are more likely to be selected for training. In the hydraulic fracturing earthquake prediction scenario of the present invention, S is a uniform time coordinate sequence within the target time interval, The data set corresponding to the discretization of the Coulomb stress rate under different environmental conditions, that is, in the hydraulic fracturing earthquake prediction scenario of the present invention, the input function set is actually the data set corresponding to the discretization of the Coulomb stress rate under different environmental conditions; the set of sampling point sets is actually the uniform time coordinate sequence within the target time interval, The input function is hour The partial differential equation loss of the position is expressed as the mathematical model of seismic activity rate:

[0037] in, represents the sampling probability of the function, S represents the set of sampling points, The input function is hour PDE loss of position, The input function is hour PDE loss of position, and Represents the function data set respectively N Middle n Function and i functions, Represents the uniform time coordinate sequence within the target time interval. k time coordinates, represents the weight parameters of the physical constraint deep operator network model, represents the output of the hydraulic fracturing earthquake possibility prediction result under the target scenario, represents the Coulomb stress rate, n From 1 to , Represents a collection of functions N The n function, that is, the Coulomb stress rate under different environmental conditions The discretized representation n A representation, k From 1 to S , That is S The first in the collection kelement, that is, the uniform time coordinate sequence within the target time interval. k time coordinates, It represents the characteristic time scale and is regarded as a constant in this method. Its specific value is usually derived from experimental data or observation fitting. It is related to the physical properties of the lithosphere (such as creep behavior or rheological properties) and the external stress loading rate that triggers earthquakes.

[0038] In this embodiment, as shown in the above formula, the present invention calculates the partial differential equation error of each function on the sampling point set before the training starts, and then calculates the probability of each function being sampled in equal proportion. The coordinate set is obtained by uniform sampling in the domain of definition. Its purpose is to minimize the growth of performance overhead while ensuring accurate estimation of the function error. The greater the sum of the errors of the function on the sampling point set, the greater the probability of being sampled. This importance sampling method allows the network to adaptively tilt the sampling distribution toward high-error data points during training, so it can ultimately increase the convergence speed of the network while improving the convergence stability and prediction accuracy of the network. The schematic diagram of the sampling process can be seen Figure 2 .

[0039] In this embodiment, the total loss function of the physical constraint deep operator network model PDIS is:

[0040] in, Represents the function set of the current batch, which is obtained by the importance sampling method of the function at the beginning of a training iteration. Represents the coordinate set of the collocation points input by the main network, including the coordinates within the definition domain and the coordinates of the boundary of the definition domain. For the i The sampling weight corresponding to the function is multiplied by the loss corresponding to the function to achieve the bias reweighting operation. is the loss function of the physical constraint deep operator network, expressed as follows:

[0041] The three terms in the formula correspond to the partial differential equation loss, initial condition loss and boundary condition loss in the physical constraint method. The coefficients of the latter two terms are used to adjust the proportion of the three losses.

[0042] Sampling weights for functions , which is equal to the ratio of the random sampling probability to the importance sampling probability:

[0043] in, represents the bias reweighted loss function of the physical constraint deep operator network model PDIS, Represents the function set of the current batch, which is obtained by the importance sampling method of the function at the beginning of a training iteration. i Indicates i functions, Indicates i The sampling weights corresponding to the functions are: represents the loss function of the physical constraint deep operator network, Represents a function dataset N Middle i functions, represents the coordinate set of the collocation points input by the backbone network, represents the weight parameters of the physical constraint deep operator network model, , and They represent the partial differential equation loss, initial condition loss, and boundary condition loss in the physical constraint method, respectively. and All represent weights, Indicates i The original sampling probability of the function, represents the sampling probability of the function, Represents a function dataset.

[0044] S2, collect and process the data of the seismicity rate mathematical model under different scenarios as training data; In this embodiment, the training data includes the Coulomb stress rate under different scenarios. The data, irrelevant constant parameters and initial boundary conditions under different scenarios, among which the Coulomb stress rate The data consists of a discrete set of points and a spline curve fit representation of the Coulomb stress rate. After the curve is discretized, it is used as the input function of the branch network.

[0045] In this embodiment, the Coulomb stress rate It is the most important training data, which represents the rate of change of the stress state on the fault plane in the prediction of seismic activity rate. It combines the changes of shear stress and effective normal stress into one quantity, reflecting the slip tendency of the fault in an earthquake-induced environment. It is the most important training data, which represents the rate of change of the stress state on the fault plane in the prediction of seismic activity rate. It combines the changes of shear stress and effective normal stress into one quantity, reflecting the slip tendency of the fault in an earthquake-induced environment.

[0046] In this embodiment, a set of Coulomb stress rates The data consists of two parts: a discrete point set of Coulomb stress rate and a spline curve fitting representation. The former is used as the input of the branch network, and the latter is used as an additional variable of the partial differential equation loss during training. Collect enough Coulomb stress rate The data can provide enough training samples for the subsequent physical constraint deep operator network model PDIS to fit the solution operator, thereby improving the generalization ability of the model and ensuring that the model can correctly learn the mapping relationship from Coulomb stress rate data to seismic activity rate.

[0047] S3. Using the training data and partial differential equation constraints, the physical constraint deep operator network model PDIS is iteratively trained to obtain the optimal step physical constraint deep operator network model PDIS weight. The implementation method is as follows: S301, initializing the bias weight parameters of the physical constraint deep operator network model PDIS to generate a set of sampling points evenly distributed in the time dimension; S302. At the beginning of an iteration, enter the Coulomb stress rate The discrete point set and sampling point set are based on the output of the physical constraint deep operator network model PDIS and the Coulomb stress rate By calculating each group of Coulomb stress rates one by one The sum of partial differential equation errors of the data on the sampling point set gives each group of Coulomb stress rates The sampling probability of the function ; S303, batch sampling is performed according to the function sampling probability, and the sampled Coulomb stress rate The data and the randomly generated time coordinate sequence are used as the input of the branch network and the trunk network respectively, and the output of the physical constraint deep operator network model PDIS is obtained. The output of the physical constraint deep operator network model PDIS is the prediction result of the possibility of hydraulic fracturing earthquake under the target scenario; S304, based on the output of the physical constraint deep operator network model PDIS, calculate the deviation reweighted loss, and perform gradient optimization to adjust the physical constraint deep operator network model PDIS parameters; S305. Repeat steps S301 to S304 until the iteration ends and the weight of the physical constraint deep operator network model PDIS of the optimal step is retained, wherein the physical constraint deep operator network model PDIS includes a trunk network and a branch network.

[0048] In this embodiment, during the training process, the bias weight parameters of the physical constraint deep operator network model PDIS are first initialized, and a set of sampling points evenly distributed in the time dimension is generated for subsequent evaluation of function importance. Subsequently, at the beginning of an iteration, the Coulomb stress rate is input Discrete point set and sampling point set, calculate each group of Coulomb stress rate one by one according to the output of the physical constraint deep operator network model PDIS and the spline curve interpolation of Coulomb stress rate The sum of partial differential equation errors of the data on the sampling point set is converted geometrically into the calculation of each group of Coulomb stress rate Then, batch sampling is performed according to the sampling probability, and the sampled Coulomb stress rate The data and randomly generated time coordinate sequence are used as the input of the branch network and the main network respectively to obtain the network output. The bias reweighted loss is calculated based on the model output, and gradient optimization is performed to adjust the model parameters. The above steps are repeated until the iteration ends and the PDIS weight of the physical constraint deep operator network model of the optimal step is retained.

[0049] S4. Initialize the physical constraint deep operator network model PDIS using the weights obtained in step S3, and obtain the hydraulic fracturing earthquake possibility prediction result under the target scenario according to the target hydraulic fracturing earthquake prediction task data. The implementation method is as follows: S401, using the weight of the physical constraint deep operator network model PDIS of the optimal step, initializing the physical constraint deep operator network model PDIS; S402. Obtain the Coulomb stress rate in the target scenario The discrete point set is represented as the input of the branch network; S403, randomly generate a time coordinate sequence of the target prediction time interval as the input of the backbone network; S404, performing forward propagation according to the input results of step S402 and step S403; S405. According to the forward propagation result, the outputs of the main network and the branch network are subjected to inner product operation to obtain the hydraulic fracturing earthquake possibility prediction result under the target scenario.

[0050] In this embodiment, when the physical constraint deep operator network model PDIS is predicted, the physical constraint deep operator network model PDIS is first initialized using the optimal physical constraint deep operator network model PDIS parameters. Then, the Coulomb stress rate under the target scene is obtained by detection or simulation. The discrete point set is represented as the input of the branch network; then, the time coordinate sequence of the target prediction time interval is generated as the input of the backbone network. Finally, the data is input into the initialized physical constraint deep operator network model PDIS, and the output result is obtained after forward propagation, that is, the prediction result of the possibility of hydraulic fracturing fluid-induced earthquakes in the target scenario.

[0051] In summary, compared with the traditional numerical solver, the present invention has fewer hyperparameters and does not require fixed-step time grid division. Curve, the present invention can directly use the discretization result as the network model input without additional preprocessing operations. Compared with the existing deep operator network, the present invention can The importance of the curve adaptively selects training batches, accelerates model training, and improves convergence stability and accuracy.

[0052] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0053] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for predicting the possibility of hydraulic fracturing earthquakes, characterized in that: The following steps are involved: S1. Construct a physical constraint deep operator network model PDIS combined with importance sampling method; S2, collect and process the data of the seismicity rate mathematical model under different scenarios as training data; S3. Using the training data and partial differential equation constraints, the physical constraint deep operator network model PDIS is iteratively trained to obtain the optimal step physical constraint deep operator network model PDIS weight; S4. Initialize the physical constraint deep operator network model PDIS using the weights obtained in step S3, and obtain the hydraulic fracturing earthquake possibility prediction result under the target scenario based on the target hydraulic fracturing earthquake prediction task data.

2. The method for predicting the possibility of hydraulic fracturing earthquakes according to claim 1, characterized in that: The expression of the bias reweighted loss function of the physical constraint deep operator network model PDIS is as follows: in, represents the bias reweighted loss function of the physical constraint deep operator network model PDIS, Represents the function set of the current batch, which is obtained by the importance sampling method of the function at the beginning of a training iteration. i Indicates i functions, Indicates i The sampling weights corresponding to the functions are: represents the loss function of the physical constraint deep operator network, Represents a function dataset N Middle i functions, represents the coordinate set of the collocation points input by the backbone network, represents the weight parameters of the physical constraint deep operator network model, , and They represent the partial differential equation loss, initial condition loss, and boundary condition loss in the physical constraint method, respectively. and All represent weights, Indicates i The original sampling probability of the function, represents the sampling probability of the function, Represents a function dataset.

3. The method for predicting the possibility of hydraulic fracturing earthquakes according to claim 1, characterized in that: The training data includes Coulomb stress rates under different scenarios The data, irrelevant constant parameters and initial boundary conditions under different scenarios, among which the Coulomb stress rate The data consist of a discrete set of points representing the Coulomb stress rate and a spline curve fit.

4. The method for predicting the possibility of hydraulic fracturing earthquakes according to claim 3, characterized in that: The step S3 comprises the following steps: S301, initializing the bias weight parameters of the physical constraint deep operator network model PDIS to generate a set of sampling points evenly distributed in the time dimension; S302. At the beginning of an iteration, enter the Coulomb stress rate The discrete point set and sampling point set are based on the output of the physical constraint deep operator network model PDIS and the Coulomb stress rate By calculating each group of Coulomb stress rates one by one The sum of partial differential equation errors of the data on the sampling point set gives each group of Coulomb stress rates The sampling probability of the function ; S303, batch sampling is performed according to the function sampling probability, and the sampled Coulomb stress rate The data and the randomly generated time coordinate sequence are used as the input of the branch network and the trunk network respectively, and the output of the physical constraint deep operator network model PDIS is obtained. The output of the physical constraint deep operator network model PDIS is the prediction result of the possibility of hydraulic fracturing earthquake under the target scenario; S304, based on the output of the physical constraint deep operator network model PDIS, calculate the deviation reweighted loss, and perform gradient optimization to adjust the physical constraint deep operator network model PDIS parameters; S305. Repeat steps S301 to S304 until the iteration ends and the weight of the physical constraint deep operator network model PDIS of the optimal step is retained, wherein the physical constraint deep operator network model PDIS includes a trunk network and a branch network.

5. The method for predicting the possibility of hydraulic fracturing earthquakes according to claim 4, characterized in that: The expression of the function sampling probability in step S302 is as follows: in, represents the sampling probability of the function, S represents the set of sampling points, The input function is hour PDE loss of position, The input function is hour PDE loss of position, and Represents the function data set respectively N Middle n Function and i functions, Represents the uniform time coordinate sequence within the target time interval. k time coordinates, represents the weight parameters of the physical constraint deep operator network model, represents the output of the hydraulic fracturing earthquake possibility prediction result under the target scenario, represents the characteristic time scale, represents the Coulomb stress rate.

6. The method for predicting the possibility of hydraulic fracturing earthquakes according to claim 4, characterized in that: The step S4 comprises the following steps: S401, using the weight of the physical constraint deep operator network model PDIS of the optimal step, initializing the physical constraint deep operator network model PDIS; S402. Obtain the Coulomb stress rate in the target scenario The discrete point set is represented as the input of the branch network; S403, randomly generate a time coordinate sequence of the target prediction time interval as the input of the backbone network; S404, performing forward propagation according to the input results of step S402 and step S403; S405. According to the forward propagation result, the outputs of the main network and the branch network are subjected to inner product operation to obtain the hydraulic fracturing earthquake possibility prediction result under the target scenario.

Images