Nonlinear crustal stress field inversion method and device based on physical data neural network

Abstract

The invention provides a nonlinear crustal stress field inversion method and device based on a physical data neural network, and is used for constructing a corresponding back propagation neural network to carry out optimal interpretation by taking a nonlinear crustal stress field boundary condition as an optimization target. According to the method, a non-linear characteristic rule of crustal stress field distribution caused by stratum structure complexity and mechanical parameter heterogeneity is analyzed, so that an optimal non-linear crustal stress field boundary condition is obtained, and the problems that in engineering practice, a non-linear crustal stress field inversion result is large in error and the inversion result is not reliable enough are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geostress field inversion, in particular to a nonlinear geostress field inversion method and device based on a physical data neural network. BACKGROUND

[0002] Geostress field analysis can provide important data support for related geological engineering analysis work, and typical further mechanical analysis can be promoted based on the inverted geostress field.

[0003] However, in actual situations, the nonlinear mechanical characteristics of rock-soil media and the existence of regional geological structure make it difficult for traditional linear geostress inversion methods to well reveal the distribution characteristics of the geostress field in the study area.

[0004] More specifically, in the process of simulating surface subsidence caused by shallow ore body mining, numerical analysis of the mechanical response of deep large underground tunnel networks or large plant surrounding rock, and other specific application scenarios, the mechanical analysis numerical simulation based on the regional geostress field inversion results is severely restricted due to the contradiction between the traditional geostress field inversion analysis method based on the elastic constitutive and the non-elastic constitutive model used in the mechanical analysis process.

[0005] That is, the existing geostress field inversion analysis method has the defects of large result error and unreliable inversion results. SUMMARY

[0006] The present application provides a nonlinear geostress field inversion method and device based on a physical data neural network, which is used to construct a corresponding back propagation neural network for optimization interpretation with nonlinear geostress field boundary conditions as the optimization target, analyze the nonlinear characteristics of the geostress field distribution caused by the complexity of the stratum structure and the non-homogeneity of the mechanical parameters, and obtain the optimal nonlinear geostress field boundary conditions to solve the problem of large error in the nonlinear geostress field inversion results and unreliable inversion results in engineering practice.

[0007] In a first aspect, the present application provides a nonlinear geostress field inversion method based on a physical data neural network, the method comprising: On the basis of the engineering geological condition data and the geostress measurement results in the study area, the multiple linear regression geostress field calculation results of the initial boundary conditions are determined by constructing a geostress field inversion model and performing multiple linear regression processing of the traditional elastic constitutive model. On the basis of the orthogonal test table constructed by the initial boundary conditions, the numerical simulation of the nonlinear constitutive model is performed on each boundary condition, and the neural network data set is configured after normalizing the numerical simulation results of the corresponding nonlinear constitutive model. The back propagation neural network is trained using the neural network dataset as a training sample, to obtain a trained neural network meeting the requirements of a loss function; According to the boundary condition coefficients obtained by calculation of the trained neural network, the model is substituted to perform nonlinear constitutive calculation, to inverse each measured point stress value and calculate the corresponding measured point root mean square error, if the measured point root mean square error is not lower than a threshold value, the corresponding data is added to the neural network dataset for cyclic processing, until the errors of all measured points meet the requirement that the error is not lower than the threshold value, the current boundary condition coefficients are output, and the regional stress field is obtained by forward calculation.

[0008] In a second aspect, the present application provides a nonlinear regional stress field inversion device based on physical data neural network, the device comprising: A first processing unit is configured to determine the initial boundary condition multivariate linear regression stress field calculation result by constructing a stress field inversion model and performing multivariate linear regression processing of a traditional elastic constitutive model based on the engineering geological condition data and the stress measurement results of the research area; A second processing unit is configured to perform nonlinear constitutive model numerical simulation processing on each boundary condition based on the orthogonal test table constructed by the initial boundary condition, and configure a neural network dataset after normalizing the corresponding nonlinear constitutive model numerical simulation results; A third processing unit is configured to train a back propagation neural network using the neural network dataset as a training sample, to obtain a trained neural network meeting the requirements of a loss function; A fourth processing unit is configured to substitute the boundary condition coefficients obtained by calculation of the trained neural network into the model to perform nonlinear constitutive calculation, to inverse each measured point stress value and calculate the corresponding measured point root mean square error, if the measured point root mean square error is not lower than a threshold value, the corresponding data is added to the neural network dataset for cyclic processing, until the errors of all measured points meet the requirement that the error is not lower than the threshold value, the current boundary condition coefficients are output, and the regional stress field is obtained by forward calculation.

[0009] In a third aspect, the present application provides a processing device comprising a processor and a memory, the memory storing a computer program, and the processor executes the method provided in the first aspect of the present application when invoking the computer program in the memory.

[0010] In a fourth aspect, the present application provides a computer readable storage medium, the computer readable storage medium storing a plurality of instructions, and the instructions are adapted to be loaded by a processor to execute the method provided in the first aspect of the present application.

[0011] From the above, the present application has the following beneficial effects: For the demand of high-precision nonlinear stress inversion, the nonlinear stress field boundary condition is taken as the optimization target, the corresponding back propagation neural network is constructed to carry out optimization interpretation, the nonlinear characteristic law of the stress field distribution caused by the complexity of the stratum structure and the heterogeneity of the mechanical parameters is analyzed, and the optimal nonlinear stress field boundary condition is obtained, so as to solve the problems of large error of the nonlinear stress field inversion result and unreliable inversion result in engineering practice. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0013] Figure 1 A flowchart of the nonlinear stress field inversion method based on the physical data neural network of the present application; Figure 2 A logic diagram of the nonlinear stress field inversion method based on the physical data neural network of the present application; Figure 3 A structural diagram of the nonlinear stress field inversion device based on the physical data neural network of the present application; Figure 4 A structural diagram of the processing device of the present application. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0015] The terms "first", "second", and the like in the description and in the claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or modules does not necessarily have to be limited to only those steps or modules clearly listed, but can include other steps or modules not clearly listed or inherent to the process, method, product, or device. The naming or numbering of the steps appearing in the present application does not mean that the steps in the method flow must be performed in the time / logical order indicated by the naming or numbering, and the named or numbered flow steps can be changed in execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0016] The division of modules appearing in the present application is a logical division, and in actual application, there can be another division manner, for example, a plurality of modules can be combined or integrated in another system, or some features can be ignored or not executed, in addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, the indirect coupling or communication connection between the modules can be electrical or other similar forms, which are not limited in the present application. Moreover, the modules or sub-modules described as separate components can or can not be physically separated, can or can not be physical modules, or can be distributed to a plurality of circuit modules, and part or all of the modules can be selected according to actual needs to achieve the purpose of the present application scheme.

[0017] Before introducing the nonlinear stress field inversion method based on physical data neural network provided by the present application, the background content involved in the present application is first introduced.

[0018] The nonlinear stress field inversion method, device and computer readable storage medium based on physical data neural network provided by the present application can be applied to a processing device, for taking the nonlinear stress field boundary condition as an optimization target, constructing a corresponding back propagation neural network to perform optimization interpretation, analyzing the nonlinear characteristic law of the in-situ stress field distribution caused by the complexity of the formation structure and the inhomogeneity of the mechanical parameters, and thus obtaining the optimal nonlinear stress field boundary condition, to solve the problem of large error and unreliable inversion result of the nonlinear stress field inversion result in engineering practice.

[0019] The physical data neural network based nonlinear stress field inversion method mentioned in the present application can be executed by a physical data neural network based nonlinear stress field inversion device, or a server, a physical host, a user equipment (UE) or other types of processing devices integrated with the physical data neural network based nonlinear stress field inversion device. The physical data neural network based nonlinear stress field inversion device can be implemented in hardware or software, and the UE can be a terminal device such as a smartphone, a tablet computer, a notebook computer, a desktop computer or a personal digital assistant (PDA), and the processing device can be set in a device cluster.

[0020] It can be understood that the present application is based on existing data or data that has been collected, and therefore the processing device executing the physical data neural network based nonlinear stress field inversion method or the corresponding application service of the physical data neural network based nonlinear stress field inversion method usually only needs to meet the required data processing capability, and the specific device type and device deployment form are flexible.

[0021] If the direct collection of existing data is also involved, the processing device needs to be further adapted in terms of software and hardware to have the corresponding data collection capability.

[0022] In addition, if the content display requirement of the processing process or the processing result is involved, the processing device can also be configured with a display screen (including a touch screen), and the processing device can also use an external display device or other devices with a display screen to complete the content display target.

[0023] Next, the physical data neural network based nonlinear stress field inversion method provided by the present application will be introduced.

[0024] Firstly, referring to Figure 1 , Figure 1 Fig. 1 shows a flowchart of the physical data neural network based nonlinear stress field inversion method of the present application. The physical data neural network based nonlinear stress field inversion method provided by the present application can specifically include the following steps S101 to S104: Step S101, based on the engineering geological condition data and the in-situ stress measurement results of the study area, the initial boundary condition multiple linear regression stress field calculation result is determined by constructing a stress field inversion model and performing a traditional elastic constitutive model multiple linear regression processing. It can be understood that the implementation of the scheme of the present application is focused on a pre-selected, determined certain research area, engineering area, in which case, the engineering geological condition data and the in-situ stress measurement results (i.e. the in-situ stress test results of the relevant measurement points) of the research area currently required to be subjected to the nonlinear stress field inversion can be obtained to preliminarily construct the in-situ stress field inversion model, and the traditional multivariate linear regression process of the elastic constitutive model is continued to be carried out, so as to complete the preliminary determination of the initial boundary conditions of the in-situ stress field inversion model of the research area by the calculated multivariate linear regression in-situ stress field calculation results.

[0025] Among them, the data of the engineering geological condition data and the in-situ stress measurement results are common data collected in geological engineering, so they can usually be directly extracted from online systems and the like, of course, real-time collection and processing are also possible, and the specific situation in the actual application of the scheme of the present application can be flexibly adjusted.

[0026] As an exemplary embodiment here, the corresponding test method or data source of the in-situ stress measurement results can specifically include water pressure cracking method, stress relief method, stress recovery method and acoustic emission method and the like in-situ stress test methods.

[0027] On the other hand, the types of the initial boundary conditions involved here can specifically include stress boundary conditions and displacement boundary conditions.

[0028] Further, as an exemplary embodiment, the construction of the in-situ stress field inversion model and the traditional elastic constitutive model multivariate linear regression processing here can specifically involve the following processing contents: 1) azimuth transformation is performed on the in-situ stress measurement results to convert the corresponding magnitude, azimuth angle and inclination angle to the coordinate system adopted in numerical simulation to obtain the measured stress component values of each known point; Specifically, the principal stress azimuth angle , the inclination angle and the principal stress magnitude value can be obtained from the in-situ stress measurement results, and the directional cosines between the principal stress axis and the axes of the coordinate system O-xyz are obtained according to the following formula in combination with the related theories of elastic mechanics: , , , Among them, is the azimuth angle of the measured principal stress value, the north direction is regarded as 0°, the clockwise direction is positive (+), and the counterclockwise direction is negative (-); is the measured principal stress inclination angle, the elevation angle is positive (+), and the depression angle is negative (-); , and are the direction cosines of the principal stress axes with respect to the x, y, z axes respectively.

[0029] Then, the six stress components in the x, y, z coordinate system are obtained by using the following coordinate transformation formula: It can be understood that in the exemplary embodiments herein, the data other than the ground stress is extracted from the engineering geological condition information obtained in the foregoing.

[0030] 2) Select the main influencing factors of the initial ground stress field, calculate the stress of each known point, and obtain the calculated stress component values of each known point; Generally, the initial ground stress field is a function of the following form: , wherein, is the initial ground stress value, which exists in the form of six stress components, , and represent the coordinate position, , and are rock mass parameters, which are the elastic modulus, Poisson's ratio and unit weight respectively, is the self-weight factor, , and are geological structure factors, is the temperature factor.

[0031] It is generally assumed that the rock mass in the study range is in the elastic stage before construction, so that the elastic analysis can be carried out for the study, and on this basis, it is assumed that each factor is linearly superimposed, so that the initial ground stress value can be obtained: , wherein, is the regression coefficient (also the target to be solved in the next step), is the main influencing factor of the selected regional ground stress field, and the commonly selected influencing factors are, for example, displacement boundary constraint condition, stress / displacement boundary condition, cross-sectional coordinate system, X-direction uniform action condition, Y-direction uniform action condition, Z-direction uniform action condition, X-direction triangular action condition, Y-direction triangular action condition, self-weight action condition, XY plane shear condition, XZ plane shear condition, YZ plane shear condition, self-weight action force, etc., is the observation error, which is a random variable, and it is considered that the expected value is zero when the number of observation data is large enough.

[0032] 3) Based on the measured stress component values and calculated stress component values of each known point, the corresponding regression coefficients are calculated using the least square analysis method; Based on the principle of multiple linear regression, the regression calculation value of the ground stress (i.e. measured stress component value) and the stress calculation value calculated under each operating condition obtained by numerical simulation (i.e. calculated stress component value) have the following relationship: , Wherein, is the observation point number, is the stress component mark, ranging from 1 to 6, is the th observation point, th stress component regression calculation value, is the regression coefficient relative to the independent variable , is the factor of the sub-load , the numerical calculation value of the th observation point th stress component, is the total number of factors including self-weight stress and construction stress.

[0033] If there are measuring points, the residual sum of squares is: , Wherein, is the observation value of the observation point stress component , according to the least square principle, the equation that makes is the minimum value is: .

[0034] 4) According to the regression coefficient, re-calculate the numerical value on the model to obtain the initial ground stress field distribution, and complete the determination of the multiple linear regression ground stress field calculation result of the initial boundary condition.

[0035] By solving the above matrix equation, we can get undetermined coefficients, and then add the calculated values calculated by numerical calculation under each operating condition, that is, the regression ground stress value of any point position in the model: , According to the measurement point multiple linear regression ground stress field calculation result, determine the initial data of the subsequent analysis process, and the corresponding data structure can be referred to as follows: , wherein, Part is the stress component inversion data of the measuring point, Part is the boundary regression coefficient data, Part is the root mean square error of the measured data and the inversion data of the measuring point.

[0036] Step S102, on the basis of the orthogonal test table constructed under the initial boundary condition, the nonlinear constitutive model numerical simulation process is carried out respectively for each boundary condition, and the normalized nonlinear constitutive model numerical simulation result is configured into a neural network data set; It can be understood that after the initial boundary condition is preliminarily determined (quantified by the corresponding boundary condition coefficient), an orthogonal test table can be constructed to better promote the subsequent data processing.

[0037] Specifically, as an exemplary embodiment, the initial data of the orthogonal test table can specifically include 1) model measuring point stress inversion value, 2) boundary condition regression coefficient, 3) root mean square error of the inversion result and the measured result of the measuring point.

[0038] In addition, it should be noted that the present application starts from step S102 here, and involves a loop process in the subsequent step S104, therefore, the content of the orthogonal test table will also be updated with the loop process, and therefore, the exemplary embodiment mentioned here refers to the initial data, and the three content items involved are not changed, and the content recorded in the content items will be dynamically changed.

[0039] It can be understood that for the numerical model, the boundary conditions at different positions can be classified into the above-mentioned stress boundary condition and displacement boundary condition.

[0040] At this time, the nonlinear constitutive model numerical simulation process of each boundary condition can be carried out for the numerical model, and after the nonlinear constitutive model numerical simulation result is obtained, normalization can be continued, and then the neural network data set, i.e. the training sample, can be configured according to the configuration scheme of the training sample, such as the proportion (such as 7:3) and the extraction method (such as random extraction) of the training set and the test set, for the subsequent neural network training.

[0041] In addition, for the numerical model of the present application, i.e. the nonlinear constitutive model (involved in step S102 and subsequent step S104), as an exemplary embodiment, the Mohr-Coulomb model, the Drucker-Prager model, the strain hardening / softening model, the double yield surface model or the Hoek-Brown model can be used.

[0042] Step S103, using the neural network data set as a training sample to train the back propagation neural network to obtain a trained neural network that meets the loss function requirements; After obtaining the previous training samples, we can proceed with the specific network training work and train the backpropagation (BP) neural network in combination with the configured loss function.

[0043] Specifically, this application has specially designed a configuration scheme for the loss function to help promote efficient and high-precision training effects.

[0044] Correspondingly, as an exemplary embodiment, the overall loss function used in training the back-propagation neural network can be specifically expressed as follows: , , , in, is the total loss, is the weight coefficient, is the physical loss term, is the data loss term, is the number of test set samples in the training sample, is the number of output data, is the jth true output value of the i-th test set, is the input of the i-th test set and combination The j-th predicted output value under the combination The optimal weight With bias combination of is the jth physical constraint component of the boundary condition data, The measured ground stress results are When inputting data, The boundary condition prediction value of the neural network output under .

[0045] Specifically, the back propagation neural network consists of an input layer, multiple hidden layers and an output layer. The input layer is responsible for receiving external data, the hidden layer performs nonlinear transformation on the data, and the output layer outputs the final result. Through the information exchange between neurons, the nonlinear mapping between input and output is realized. When processing data, for the first The first hidden layer in the neurons, through forward propagation according to the following formula based on the input data (the output of the input layer or the previous hidden layer) using weights and bias After calculating the weighted sum, the activation function Activation is the output of the neurons in this layer : , After forward propagation obtains the output of each neuron in the output layer, the error between the output result and the true value is calculated according to the following formula: The first hidden layer in the The error signal of a neuron : , in, The activation function outputs The derivative value of is the weight of the neurons in this layer and the next layer, For the Tier The error signal of a neuron.

[0046] During the back propagation process, according to the error signal of the neurons in this layer and the weight of the input data on the neurons in this layer and bias To update: , , in, is the learning rate, The input result of the neurons in this layer (that is, the output result of the neurons in the previous layer).

[0047] In order to obtain the optimal weight matrix and bias combination so that the network's prediction results are best fitted with the true values, it is necessary to construct a data-driven loss function to quantify the error between the two and determine the optimal weight matrix and bias combination by minimizing the loss function, as shown in the following formula: , in, is the optimal weight and bias combination, It is the combination of the current weight matrix and bias The data-driven loss function under is the number of test set samples, is the number of output data, For the The test set The actual output value, For the The input of the test set and The next predicted output values.

[0048] At the same time, the physical driving data corresponding to the minimum RMSE value is obtained, and the physical driving loss function value is calculated.

[0049] The physical drive part uses an iterative process to gradually approximate the real results and uses the linear elastic multivariate regression results as the initial values ​​of the physical constraints, as shown in the following formula: , in, Represents the physical constraint results of boundary condition data, The stress data of the measuring point representing the first elastic calculation result, Represents the nonlinear numerical simulation calculation process, Indicates the error of the stress value at the measuring point of the corresponding process.

[0050] , in, In order to make The minimum number of iterations, Represents the corresponding nonlinear numerical simulation calculation process.

[0051] Thus, there is the total loss mentioned above .

[0052] In this case, based on the calculated dual-constraint loss results of the physical data, the neural network weights and biases are updated through backpropagation, and the iterations are repeated until the loss function is lower than the threshold, thereby obtaining a standardized output that conforms to the test dataset and physical information.

[0053] It is worth noting that in the design of the present application, the so-called physical data neural network, or the physical data dual-driven neural network, does not refer to the situation where the loss function here involves physical loss terms and data loss terms. The loss function is only the convergence condition of the model. Its essence is to solve the nonlinear problem of the geostress field. It is reflected in a series of data processing starting from step S101 to obtain the engineering geological conditions data and geostress measured results of the study area, or the underlying design idea behind a series of data processing starting from step S101 to obtain the engineering geological conditions data and geostress measured results of the study area, which corresponds to the processing architecture of the overall solution, so as to achieve high-precision nonlinear geostress field inversion effect.

[0054] Step S104, according to the boundary condition coefficient calculated by the trained neural network, substitute into the model for nonlinear constitutive calculation, to inverse each measured point stress value and calculate the corresponding measured point root mean square error, if the measured point root mean square error is not less than the threshold value, the corresponding data is added to the neural network data set for cyclic processing, until all the errors of the measured points meet the requirement that the error is not less than the threshold value, output the current boundary condition coefficient, and calculate to obtain the regional stress field.

[0055] After the training of the back propagation neural network is completed, at this time, the nonlinear stress field inversion based on the physical data neural network specially designed in the present application can be promoted, the determined boundary condition coefficient, i.e. the calculated boundary condition coefficient, obtained by the trained neural network can be substituted into the nonlinear constitutive model mentioned in the previous step S102 for nonlinear constitutive calculation to inverse the stress value of each measured point.

[0056] It can be seen that the purpose of training the back propagation neural network is to calculate the boundary condition coefficient that best meets the previous situation (embodied by the training sample) through iterative search.

[0057] At this time, for each measured point, if the measured point stress value obtained by nonlinear constitutive calculation has a measured point root mean square error lower than the threshold value, it can be considered to meet the requirement, otherwise, if it is not lower than the threshold value, it can be considered not to meet the requirement, and it needs to return to the previous step S102 for cyclic / iterative calculation, so that the measured point stress value obtained by nonlinear constitutive calculation of all measured points has a measured point root mean square error lower than the threshold value and meets the requirement.

[0058] The measured point root mean square error, i.e. the root mean square error of the measured point, is calculated according to the conventional calculation logic of the root mean square error, and is calculated from the stress inversion value obtained by nonlinear constitutive calculation and the measured stress value.

[0059] Specifically, the measured point root mean square error RMSE can be calculated according to the following formula: , Wherein, represents the root mean square error of data input , is the number of measured points, is the number of stress components to be inverted, is the measured value of the first stress component of the first measured point, is the inversion value of the first stress component of the first measured point.

[0060] ​​​​And in the above-mentioned cycle / iteration processing, in the case that the measured point root mean square error of the measured point stress value obtained by the nonlinear constitutive calculation of all measured points is less than the threshold value, the corresponding boundary condition coefficient of the case can be recorded as the current boundary condition coefficient, which is output as the result, and the final regional stress field of the current study area is obtained by forward calculation, realizing the nonlinear stress field inversion target based on physical data neural network.

[0061] Among them, for the nonlinear constitutive model numerical simulation processing of the above step S102 and the nonlinear constitutive calculation involved in this step S104, as an exemplary embodiment, the finite difference method can be used to promote the specific calculation process.

[0062] At the same time, as the specific content of the calculated boundary condition and the current boundary condition coefficient, the specific content of the calculated boundary condition can specifically include X-direction extrusion, Y-direction extrusion, Z-direction extrusion, XY plane shear, YZ plane shear, XZ plane shear and self weight, in addition, the corresponding boundary condition application type can specifically include uniform type and triangular type.

[0063] It can be understood that the above-mentioned Figure 2 The logic schematic diagram of the nonlinear stress field inversion method based on physical data neural network of the present application is shown, which is used to better understand the above-mentioned scheme.

[0064] Finally, in general, for the high-precision nonlinear stress inversion requirement, the nonlinear stress field boundary condition is taken as the optimization target, the corresponding back propagation neural network is constructed for optimization interpretation, the nonlinear characteristic law of the stress field distribution caused by the complexity of the stratum structure and the non-homogeneity of the mechanical parameters is analyzed, and the optimal nonlinear stress field boundary condition is obtained, so as to solve the problem of large error and unreliable inversion result of the nonlinear stress field inversion result in engineering practice.

[0065] The above is the introduction of the nonlinear stress field inversion method based on physical data neural network provided by the present application. In order to better implement the nonlinear stress field inversion method based on physical data neural network provided by the present application, the present application also provides a nonlinear stress field inversion device based on physical data neural network from the functional module angle.

[0066] Referring to Figure 3 , Figure 3 A structural schematic diagram of the nonlinear stress field inversion device based on physical data neural network of the present application, in the present application, the nonlinear stress field inversion device based on physical data neural network 300 can specifically include the following structures: The first processing unit 301 is configured to determine the multiple linear regression in-situ stress field calculation result of the initial boundary condition by constructing an in-situ stress field inversion model and performing multiple linear regression processing of a traditional elastic constitutive model based on the engineering geological condition data of the research area and the measured results of the in-situ stress. The second processing unit 302 is configured to perform numerical simulation processing of the non-linear constitutive model on each boundary condition respectively based on the orthogonal test table constructed by the initial boundary condition, and configure a neural network data set after normalizing the numerical simulation results of the corresponding non-linear constitutive model. The third processing unit 303 is configured to train the back propagation neural network using the neural network data set as a training sample to obtain a trained neural network meeting the requirement of a loss function. The fourth processing unit 304 is configured to perform non-linear constitutive calculation by substituting the boundary condition coefficient obtained by the trained neural network into the model, to invert the stress values of each measured point and calculate the root mean square error of the corresponding measured point, and if the root mean square error of the measured point is not lower than a threshold value, the corresponding data is added to the neural network data set for cyclic processing until the errors of all measured points meet the requirement that the error is not lower than the threshold value, and the current boundary condition coefficient is outputted, and the regional in-situ stress field is obtained by forward calculation.

[0067] In an exemplary embodiment, the corresponding test methods of the in-situ stress measurement results include the hydraulic fracturing method, the stress relief method, the stress recovery method and the acoustic emission method. The types of the initial boundary condition include the stress boundary condition and the displacement boundary condition.

[0068] In another exemplary embodiment, the construction of the in-situ stress field inversion model and the multiple linear regression processing of the traditional elastic constitutive model involve the following processing contents: The measured stress component values of each known point are obtained by performing azimuth transformation on the in-situ stress measurement results to convert the corresponding values, azimuth angles and inclination angles to the coordinate system used in numerical simulation. The main influencing factors of the initial in-situ stress field are selected to calculate the stress of each known point to obtain the calculated stress component values of each known point. The regression coefficients are calculated using the least square analysis method based on the measured stress component values and the calculated stress component values of each known point. The initial in-situ stress field distribution is obtained by re-performing numerical calculation on the model according to the regression coefficients, and the determination of the multiple linear regression in-situ stress field calculation result of the initial boundary condition is completed.

[0069] In another exemplary embodiment, the initial data of the orthogonal test table includes the three contents of the in-situ stress inversion values of the model measured points, the boundary condition regression coefficients, and the root mean square error of the inversion results and the measured results of the measured points.

[0070] In yet another exemplary embodiment, the overall loss function employed in training the back-propagation neural network is represented as follows: , , , wherein, is the total loss, is the weight coefficient, is the physical loss term, is the data loss term, is the number of test set samples in the training sample, is the number of output data, is the jthreal output value of the ithtest set, is the input of the ithtest set and the jthpredicted output value under the combination of the combination is the optimal weight and the combination of bias , is the jthphysical constraint component of the boundary condition data, is the measured stress result at the measuring point is the boundary condition prediction value of the neural network output under the combination when the input data is.

[0071] In yet another exemplary embodiment, the nonlinear constitutive model specifically employs the Mohr-Coulomb model, the Drucker-Prager model, the strain hardening / softening model, the dual yield surface model, or the Hoek-Brown model.

[0072] In yet another exemplary embodiment, the nonlinear constitutive model numerical simulation processing and the nonlinear constitutive calculation both specifically employ the finite difference method. In the specific content of the calculated boundary condition coefficient and the current boundary condition coefficient, the X-direction extrusion, the Y-direction extrusion, the Z-direction extrusion, the XY plane shear, the YZ plane shear, the XZ plane shear, and the self-weight correspond to the types of boundary conditions applied, including the uniform type and the triangular type.

[0073] The present application also provides a processing device from the perspective of hardware structure. Referring to Figure 4 , Figure 4 a structural schematic diagram of the processing device of the present application is shown. Specifically, the processing device of the present application can include a processor 401, a memory 402, and an input / output device 403. The processor 401 is used to execute the computer program stored in the memory 402 to realize the functions as Figure 1corresponding embodiments of the nonlinear stress field inversion method based on physical data neural network; or the processor 401 is configured to implement the above functions of the units in the corresponding embodiments of the nonlinear stress field inversion method based on physical data neural network when executing the computer program stored in the memory 402. Figure 3 corresponding embodiments of the nonlinear stress field inversion method based on physical data neural network; or the processor 401 is configured to implement the above functions of the units in the corresponding embodiments of the nonlinear stress field inversion method based on physical data neural network when executing the computer program stored in the memory 402. Figure 1 corresponding embodiments of the nonlinear stress field inversion method based on physical data neural network.

[0074] For example, the computer program can be divided into one or more modules / units, one or more modules / units are stored in the memory 402 and executed by the processor 401 to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the computer device.

[0075] The processing device can include, but is not limited to, the processor 401, the memory 402, the input and output device 403. Those skilled in the art can understand that the schematic diagram is only an example of the processing device, and does not constitute a limitation on the processing device, which can include more or less components than the schematic diagram, or combine certain components, or different components, for example, the processing device can also include a network access device, a bus, etc., and the processor 401, the memory 402, the input and output device 403 are connected through the bus.

[0076] The processor 401 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the processing device, which connects all parts of the device through various interfaces and lines.

[0077] The memory 402 can be used to store computer programs and / or modules, and the processor 401 realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 402, and calling data stored in the memory 402. The memory 402 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, and the like; and the data storage area can store data created according to the use of the processing device, and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0078] When the processor 401 is used to execute the computer programs stored in the memory 402, the following functions can be realized: On the basis of the engineering geological condition data of the research area and the measured results of the ground stress, the ground stress field inversion model is constructed, and the multiple linear regression processing of the traditional elastic constitutive model is performed to determine the multiple linear regression ground stress field calculation results of the initial boundary conditions; On the basis of the orthogonal test table constructed by the initial boundary conditions, the numerical simulation of the nonlinear constitutive model is performed on each boundary condition, and the corresponding numerical simulation results of the nonlinear constitutive model are normalized and configured into a neural network data set; The neural network data set is used as a training sample to train the back propagation neural network, and a trained neural network meeting the requirements of the loss function is obtained; According to the boundary condition coefficients obtained by the calculation of the trained neural network, the nonlinear constitutive calculation is performed by substituting the model, the measured point stress values are inverted, and the corresponding root mean square errors of the measuring points are calculated. If the root mean square error of the measuring point is not less than the threshold value, the corresponding data is added to the neural network data set for cyclic processing until the errors of all measuring points meet the requirement that the error is not less than the threshold value, the current boundary condition coefficient is output, and the regional ground stress field is obtained by direct calculation.

[0079] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the nonlinear ground stress field inversion device based on the physical data neural network, the processing device and the corresponding units described above can be referred to as Figure 1 For the description of the nonlinear ground stress field inversion method based on the physical data neural network in the corresponding embodiment, details are not repeated here.

[0080] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware by the instructions, which can be stored in a computer readable storage medium and loaded and executed by a processor.

[0081] To this end, the present application provides a computer readable storage medium, which stores a plurality of instructions capable of being loaded by a processor to execute the present application as Figure 1 The steps of the nonlinear geostress field inversion method based on the physical data neural network in the corresponding embodiments can refer to the specific operations of the method for inverting a nonlinear geostress field based on a physical data neural network as Figure 1 The description of the nonlinear geostress field inversion method based on the physical data neural network in the corresponding embodiments is not repeated here.

[0082] The computer readable storage medium can include a read only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0083] Due to the instructions stored in the computer readable storage medium, the present application as Figure 1 The steps of the nonlinear geostress field inversion method based on the physical data neural network in the corresponding embodiments, therefore, can implement the present application as Figure 1 The beneficial effects of the nonlinear geostress field inversion method based on the physical data neural network in the corresponding embodiments are described in detail above, and are not repeated here.

[0084] The nonlinear geostress field inversion method, device, processing equipment and computer readable storage medium based on the physical data neural network provided by the present application are described in detail above, and the principles and implementation modes of the present application are described by applying specific examples. The above embodiment descriptions are only used to help understand the core idea of the present application; meanwhile, for those skilled in the art, the specific implementation modes and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present description should not be understood as a limitation of the present application.

Claims

1. A nonlinear stress field inversion method based on physical data neural network, characterized in that, The method comprises: On the basis of engineering geological condition data and in-situ stress measurement results of a study area, a stress field inversion model is constructed, and a multivariate linear regression processing of a traditional elastic constitutive model is performed to determine a multivariate linear regression stress field calculation result of an initial boundary condition; On the basis of an orthogonal test table constructed by the initial boundary condition, a numerical simulation processing of a non-linear constitutive model is performed on each boundary condition, and after a corresponding numerical simulation result of the non-linear constitutive model is normalized, a neural network data set is configured; The neural network data set is used as a training sample to train a back propagation neural network, and a trained neural network meeting a loss function requirement is obtained; According to boundary condition coefficients calculated by the trained neural network, a non-linear constitutive calculation is performed on the model to invert each in-situ stress value of a measurement point and calculate a corresponding root mean square error of the measurement point, if the root mean square error of the measurement point is not lower than a threshold value, corresponding data is added to the neural network data set for a cyclic processing, until errors of all measurement points meet a requirement that the errors are not lower than the threshold value, a current boundary condition coefficient is output, and a regional stress field is obtained by forward calculation.

2. The method of claim 1, wherein, Corresponding test methods of the in-situ stress measurement results include a hydraulic fracturing method, a stress relief method, a stress recovery method and an acoustic emission method; Types of the initial boundary condition include stress boundary conditions and displacement boundary conditions.

3. The method of claim 1, wherein, The construction of the stress field inversion model and the multivariate linear regression processing of the traditional elastic constitutive model involve the following processing contents: Azimuthal transformation is performed on the in-situ stress measurement results to convert corresponding values, azimuth angles and inclination angles to a coordinate system adopted in numerical simulation to obtain measured stress component values of each known point; Main influencing factors of an initial stress field are selected to calculate stresses of the known points to obtain calculated stress component values of the known points; On the basis of the measured stress component values and the calculated stress component values of the known points, regression coefficients are calculated by using a least square analysis method; According to the regression coefficients, numerical calculation is performed on the model again to obtain an initial stress field distribution, and determination of the multivariate linear regression stress field calculation result of the initial boundary condition is completed.

4. The method of claim 1, wherein, Initial data of the orthogonal test table include three contents of model measurement point stress inversion values, boundary condition regression coefficients, root mean square errors of measurement point inversion results and measured results.

5. The method of claim 1, wherein, An overall loss function used for training the back propagation neural network is expressed as follows: ， ， ， wherein, is the total loss, is a weight coefficient, is a physical loss term, is a data loss term, is the number of test set samples in the training sample, is the number of output data, is the jth real output value of the ith test set, is the input of the ith test set and the combination of the jth predicted output value under the combination is the optimal weight and the combination of bias, is the jth physical constraint component of the boundary condition data, is the measured stress result at the measurement point is the boundary condition prediction value of the neural network output under the combination when the input data is the input data.​ 6. The method of claim 1, wherein, The non-linear constitutive model specifically adopts a Mohr-Coulomb model, a Drucker-Prager model, a strain hardening / softening model, a double yield surface model or a Hoek-Brown model.

7. The method of claim 1, wherein, The numerical simulation processing of the non-linear constitutive model and the non-linear constitutive calculation specifically adopt a finite difference method; In specific contents of the calculated boundary condition coefficients and the current boundary condition coefficients, X-direction extrusion, Y-direction extrusion, Z-direction extrusion, XY plane shearing, YZ plane shearing, XZ plane shearing and self weight are included, and corresponding boundary condition application types include uniform distribution and triangle type.

8. A device for nonlinear stress field inversion based on physical data neural networks, characterized in that, The device comprises: The first processing unit is configured to determine the multiple linear regression geostress field calculation result of the initial boundary condition by constructing a geostress field inversion model and performing multiple linear regression processing of a traditional elastic constitutive model based on the engineering geological condition data and the measured results of the ground stress of the research area. The second processing unit is configured to perform nonlinear constitutive model numerical simulation processing on each boundary condition based on the orthogonal test table constructed by the initial boundary condition, normalize the corresponding nonlinear constitutive model numerical simulation result, and configure a neural network data set. The third processing unit is configured to train a back propagation neural network using the neural network data set as a training sample to obtain a trained neural network meeting the requirement of a loss function. The fourth processing unit is configured to perform nonlinear constitutive calculation by substituting the boundary condition coefficient obtained by calculation of the trained neural network into a model, to invert each measured point stress value and calculate the corresponding root mean square error of the measuring point, and if the root mean square error of the measuring point is not lower than a threshold value, the corresponding data is added to the neural network data set for cyclic processing until the errors of all measuring points meet the requirement that the errors are not lower than the threshold value, and the current boundary condition coefficient is output and the regional geostress field is calculated.

9. A processing device, characterized by The computer readable storage medium stores a plurality of instructions adapted to be loaded by the processor to execute the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a plurality of instructions adapted to be loaded by the processor to execute the method of any one of claims 1 to 7.

Images