Coal rock full space-time fracture field inversion method and system based on interpretable graph neural network

Abstract

The invention discloses a coal rock full-time-space fracture field inversion method and system based on an interpretable graph neural network. The inversion method comprises the following steps: obtaining staged time CT detection and full-time-space acoustic emission monitoring information through a loaded fracture test of a testing machine, and constructing a data set; training the graph neural network to extract data features to obtain a full space-time fracture parameter real-time quantitative inversion model; inputting acoustic emission scalar parameter data of a test set into the model to obtain a full space-time fracture parameter inversion result; and obtaining an influence factor of the acoustic emission scalar parameter on a fracture parameter inversion result through a Bayesian interpreter, and revealing an inversion mechanism model. The inversion model can realize the inversion of the whole space-time fracture field through the whole space-time acoustic emission monitoring data in the loading process, and fracture parameters can be accurately inverted in real time. Meanwhile, a mechanism model of fracture parameter inversion is disclosed.

Description

Technical Field

[0001] The present invention relates to the field of coal rock monitoring technology, and in particular to a coal rock full-time and space fracture field inversion method and system using an interpretable graph neural network. Background Art

[0002] With the continuous increase in the depth of coal mining, the severity of underground disasters has become increasingly prominent, which greatly threatens the safe production and efficient mining of coal mines. Studies have shown that during coal mining, mining activities will destroy the stress equilibrium state of the original coal rock mass, leading to the redistribution of the stress field and the structural mutation of the coal rock mass. This process usually involves the rapid and unstable expansion of cracks and induces large-scale rupture (fracture) disasters, which is the root cause of dynamic disasters such as rock burst, coal and gas outburst. Therefore, in order to deeply understand the mechanism of coal rock dynamic disasters and realize early warning of disasters, it is necessary to carry out in-depth research on the distribution and evolution characteristics of the fracture field.

[0003] Although existing rock mechanics theories and methods attempt to explain the evolution of coal rock fracture fields caused by mining activities, this complex process is still regarded as a "black box" that is difficult to observe and understand. There are still challenges in achieving real-time monitoring of fracture fields, accurate quantitative inversion, exploring their breeding mechanisms and revealing their development and evolution processes. At present, industrial X-rays, CT scanning and numerical simulation methods have made certain progress. These technologies can be used to reconstruct coal rock structures, especially in constructing three-dimensional digital models of discontinuous structures and fracture fields. For example, patent CN108072467A discloses a method for measuring the internal stress field of a discontinuous structure, patent CN114169182A discloses a method and equipment for reconstructing an ellipsoid model of rock face fractures, and patent CN114187423A discloses a method, electronic equipment and storage medium for reconstructing surrounding rock fractures in three-dimensional simulation tests. However, these methods have some limitations. First, they often rely on high laboratory conditions and complex model reconstruction methods, which limits their wide application in actual engineering. Secondly, the prospects for the promotion and application of fracture field reconstruction inversion are still unclear and need further research and improvement. In addition, these methods cannot provide real-time dynamic spatiotemporal evolution characteristics of fracture field distribution, which may cause certain limitations in actual production processes.

[0004] In recent years, some experimental methods combining detection-monitoring technology with numerical simulation have been applied in laboratories. However, the current use of these detection and monitoring information is still mainly focused on analyzing its temporal variation law. At the same time, relying solely on numerical simulation for fracture field inversion requires high computational costs, and even small geometric changes are likely to cause bottlenecks in the iteration process. Therefore, how to use detection and monitoring information to achieve full-time and space fracture field inversion and make up for the limitations of numerical simulation methods still needs further research. With the continuous advancement and development of science and technology, the breakthrough of deep learning methods has attracted people's attention to the use of intelligent algorithms in fracture field inversion. In particular, the use of graph neural networks and Bayesian networks can effectively extract the features of different types of data. This provides the possibility of using deep learning methods to achieve accurate inversion, real-time reproduction and mechanism interpretation of fracture parameters and evolution processes of fracture fields in loaded coal and rock masses, but there is still a lack of specific feasible ideas and implementation methods. Therefore, it is urgent to carry out actual measurements of detection and monitoring signals during the loading process of coal and rock masses, and establish the relationship between acoustic emission waveforms, characteristic parameters, event locations, full-space fracture positioning and fracture parameter data. At the same time, the proposed coal-rock full-time and space fracture field inversion method and system based on interpretable graph neural network has a significant role in revealing the "black box" evolution process of coal-rock dynamic disasters and realizing early warning of potential disasters. Summary of the invention

[0005] In response to the problems and needs raised above, this solution proposes a coal rock full-time and space fracture field inversion method and system based on an interpretable graph neural network. Due to the adoption of the following technical features, it can achieve the above technical objectives and bring about many other technical effects.

[0006] One object of the present invention is to propose a coal-rock full-time and space fracture field inversion method using an interpretable graph neural network, comprising the following steps:

[0007] S10: arranging CT detection sensors and acoustic emission monitoring sensors on the coal-rock mass, and placing the coal-rock mass in a testing machine for a load fracture test, and collecting CT detection information at staged moments and full-time and full-space acoustic emission monitoring information during the process of the coal-rock mass being loaded and fractured by the CT detection sensors and the acoustic emission monitoring sensors; wherein the CT detection information includes full-space fracture parameter information at staged moments, and the acoustic emission monitoring information includes full-time and full-space waveform information, full-time and full-space event information, and full-time and full-space scalar parameter information;

[0008] S20: matching the full-space fracture parameter information and the full-space acoustic emission monitoring information at the stage moments in the process of coal-rock mass load fracture according to the time, to obtain a fracture parameter data set; dividing the fracture parameter data set into a training set, a validation set and a test set according to corresponding proportions;

[0009] S30: The graph neural network extracts features from the full-space fracture parameter data at staged moments in the training set and the validation set of the fracture parameter data set and the acoustic emission monitoring data at the corresponding moments, and trains to obtain a real-time quantitative inversion model for the full-space fracture parameters; the full-space waveform data and scalar parameter data in the test set are input into the inversion model to obtain the full-space fracture parameter inversion results, i.e., the fracture parameters of the full-space fracture field during the loading process; and the full-space fracture field parameter evolution process of the coal rock mass (30) under loading is obtained;

[0010] S40: The scalar parameter data of the full-time and space acoustic emission monitoring data in the test set of the fracture parameter data set and the full-time and space fracture parameter inversion results are input into the graph neural network interpreter based on the Bayesian network to obtain the influencing factors between the acoustic emission scalar parameters and the fracture parameter inversion results, and then obtain the real-time quantitative inversion mechanism model of the fracture parameters based on the acoustic emission scalar parameters.

[0011] In addition, the coal-rock full-time and space fracture field inversion method based on the interpretable graph neural network of the present invention may also have the following technical features:

[0012] In one example of the present invention, in step S30, the graph neural network extracts features of the full-space crack parameters at staged moments and the acoustic emission monitoring information data at corresponding moments in the training set and the validation set of the crack parameter data set, including the following steps:

[0013] Encoding the acoustic emission data into acoustic emission graph structure data through a coding layer;

[0014] The fusion layer fuses the waveform and scalar parameters in the input acoustic emission image structure data;

[0015] The feature extraction layer extracts the features of the result after scalar parameter data fusion, extracts node features and obtains new acoustic emission waveforms and scalar parameter features;

[0016] The weight vector The parameterized self-attention mechanism extracts higher-dimensional features;

[0017] All selections at time j are normalized by the Softmax function, and the feature similarity of the acoustic emission scalar parameters at different times is calculated;

[0018] The corresponding linear combination of characteristics is obtained by coefficient calculation, and the characteristics of the final output of the scalar parameters of the acoustic emission position at different times are obtained;

[0019] Connect multiple self-attention mechanisms through the average calculation method to stabilize the calculation process and obtain the final features extracted by the multi-head graph attention mechanism;

[0020] The features extracted by the multi-head graph attention mechanism are normalized through the Layer Normalization function to improve training stability and speed;

[0021] The features extracted by the Layer Normalization function are linearly transformed through the fully connected layer to extract higher-dimensional features;

[0022] The features extracted by the fully connected layer are regressed through the Readout function to obtain the regression parameters of each graph structure.

[0023] In one example of the present invention, the expression for extracting features from the result of scalar parameter data fusion through the feature extraction layer is as follows:

[0024]

[0025] Where h represents the acoustic emission scalar parameter in the acoustic emission graph structure of the input graph attention neural network; F represents the number of acoustic emission characteristic parameters; h ′ The acoustic emission scalar parameter in the new graph structure representing the output of the feature extraction layer; F ′ The number of AE scalar parameters representing potentially different cardinalities.

[0026] In one example of the present invention, the expression of extracting higher-dimensional features by the self-attention mechanism is as follows:

[0027]

[0028] j∈N i

[0029]

[0030] Among them, e ij Represents the characteristic similarity of the acoustic emission waveform at time i to time j; N i represents the neighborhood at time j; a represents the self-attention mechanism, which consists of a single-layer feedforward neural network; represents the weight vector; W represents the weight matrix; and They represent the acoustic emission scalar parameters at different times in the acoustic emission image structure data.

[0031] In one example of the present invention, in step S30, the graph neural network extracts features from the full-space crack parameter data at the stage moment and the acoustic emission data at the corresponding moment in the training set and the validation set of the crack parameter data set, including:

[0032] The square difference between the true value and the inversion value of the full-space fracture parameter is calculated by the mean square error loss function, and the model is optimized according to the function calculation value. The calculation formula of the mean square error loss function is as follows:

[0033]

[0034] Among them, MSE represents the calculation result of the mean square error loss function; Represents the true value of the crack parameter in the entire space; Represents the inversion value of the full-space fracture parameters.

[0035] In one example of the present invention, in step S40, all acoustic emission graph structure data G in the test set are input into Φ I The output of the model is T I , the calculation formula is as follows:

[0036] Φ I :G→T I

[0037] Among them, Φ I represents the full-time and space crack quantitative inversion model; G represents all acoustic emission structure data; T I Represents the inversion space of the full-space and time-fracture quantitative inversion model.

[0038] In an example of the present invention, the step S40 specifically includes the following:

[0039] S401: Encode the acoustic emission data in the test set into acoustic emission graph structure data G and input it into the graph neural network interpreter. The data generation module activates the acoustic emission feature v at each moment in G. The data generation module controls the activation probability of the acoustic emission feature in the graph structure through the parameter p.

[0040] S402: Using the variable screening module to filter D I The variable data in the extractor is used to speed up the mechanism extraction efficiency of the interpreter. The variable screening module extracts important variables I, MB through the Markov layer P (I) is the minimum set of variables obtained by screening, denoted as U(I), I is independent of other variables in P. At the same time, Set the random variable s in v Perfect mapping of distribution B * , Provided separately with Same statistics;

[0041] S403: The interpreter passes the BIC function D I Learning to interpret Bayesian networks;

[0042] S404: The full-time and space fracture parameter inversion mechanism model is obtained according to the influencing factors output by the interpreter. The calculation formula is as follows:

[0043] M I :T I =m1×M1+m2×M2+…+m n ×M n

[0044] Among them, M n represents the acoustic emission scalar parameter in the full-time and space fracture parameter inversion mechanism; M I Represents the full-time and space fracture parameter inversion mechanism model.

[0045] In an example of the present invention, step S401 includes the following steps:

[0046] Set the random variable s v To indicate whether the acoustic emission feature v is activated, set s = {s v}, obtain the activated acoustic emission graph structure data G(s);

[0047] Input G(s) into Φ I The output of the model is Φ I (G(s)), calculate Φ through Ι(·) I (G(s)) and Φ I The difference between (G) and (s) is used to obtain the target variable I of the real-time quantitative inversion model of all-time and space fracture parameters based on G(s);

[0048] The acoustic emission scalar parameter variable at each moment is set to Δv = {s v}, denoted as D I ;

[0049] The neighborhood time of the acoustic emission scalar parameter at each moment Set to V, after n samplings, D I The size of is n×(|V|+1), and the calculation formula is as follows:

[0050] p∈(0,1)

[0051] I={Ι(Φ I (G(s)))};Φ I :g→T I

[0052] Δv={s v}; v∈V

[0053] G(s); G(s)∈g

[0054] Among them, the parameter p represents the activation probability of the feature in the graph structure; Ι(·) represents the calculation of Φ I(G(s)) and Φ I (G) Function of whether there is a significant difference; g represents all the acoustic emission map structure data in the dataset.

[0055] In one example of the present invention, in step S403, the interpreter uses the BIC function D I Learn to explain the Bayesian network in the following formula:

[0056]

[0057] in, represents the objective function for calculating the impact factor of each acoustic emission feature; BIC represents the function for calculating the impact factor; B represents the Bayesian network; θ B represents the Bayesian network parameters; l(θ B :D I [U(I)]) represents θ B and D I [U(I)]; Dim[B] represents the dimension of the Bayesian network; m n The acoustic emission scalar parameter influence factor representing the full-time and space fracture parameter inversion mechanism model; represents the parameter value that maximizes the log-likelihood (maximum likelihood estimator); represents the full-time and space-time crack parameter inversion mechanism obtained by the interpreter; ε represents the interpretable domain; V(B) represents the set of random variables in the Bayesian network; M represents the number of variables in the restriction B.

[0058] Another object of the present invention is to propose a coal-rock full-time and space fracture field inversion system with an interpretable graph neural network, comprising:

[0059] The information acquisition module is configured to arrange CT detection sensors and acoustic emission monitoring sensors on the coal rock mass, and place the coal rock mass in a testing machine for a load fracture test, and acquire CT detection information at staged moments and full-time and full-space acoustic emission monitoring information during the process of the coal rock mass being loaded and fractured by the CT detection sensors and the acoustic emission monitoring sensors; wherein the CT detection information includes full-space fracture parameter information at staged moments, and the acoustic emission monitoring information includes full-time and full-space waveform information, full-time and full-space event information, and full-time and full-space scalar parameter information;

[0060] The data set division module is configured to correspond the full-space fracture parameter information and the full-time and space acoustic emission monitoring information at the stage moments in the process of coal rock mass load fracture according to the time to obtain the fracture parameter data set; the fracture parameter data set is divided into a training set, a verification set and a test set according to the corresponding proportions;

[0061] The model building and inversion module is configured to extract features of the full-space fracture parameter data and the acoustic emission monitoring data at the corresponding time in the training set and the verification set of the fracture parameter data set through a graph neural network, and train to obtain a real-time quantitative inversion model of the full-time and space fracture parameters; input the full-time and space waveform data and scalar parameter data in the test set into the inversion model to obtain the full-time and space fracture parameter inversion results, that is, the fracture parameters of the full-time and space fracture field during the loading process; and obtain the full-time and space fracture field parameter evolution process of the coal rock mass (30) under loading;

[0062] The inversion mechanism model module is configured to input the scalar parameter data of the full-time and space acoustic emission monitoring data in the test set of the fracture parameter data set and the full-time and space fracture parameter inversion results into a graph neural network interpreter based on a Bayesian network, obtain the influencing factors between the acoustic emission scalar parameters and the fracture parameter inversion results, and then obtain a real-time quantitative inversion mechanism model of fracture parameters based on the acoustic emission scalar parameters.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. The present invention focuses on the deep learning method to invert the full-time and space fracture field of the loaded coal-rock mass through the detection-monitoring information in the coal-rock mass load fracture experiment, combining the graph neural network and the Bayesian network to reveal the mechanism model of quantitative inversion of full-time and space fracture parameters, and realize the high-precision inversion and full-time and space reproduction of the fracture field and its evolution of the loaded coal-rock mass. Compared with the traditional numerical simulation method, this coal-rock full-time and space fracture field inversion method and system with interpretable graph neural network has higher computational efficiency and better adaptability, and can effectively make up for the limitations of traditional methods.

[0065] 2. The real-time quantitative inversion model of all-time and space fracture parameters in the present invention is obtained by joint feature extraction of fracture parameters through graph neural network, and a fracture parameter feature extraction method for acoustic emission information during coal rock mass loading is proposed. Through this fracture parameter extraction method, the fracture parameters can be inverted through the acoustic emission information during coal rock mass loading obtained in the laboratory, which lays a certain foundation for the subsequent all-time and space fracture field inversion and the evolution process of all-time and space fracture field parameters of coal rock mass loading.

[0066] 3. The real-time quantitative inversion mechanism interpreter of all-space fracture parameters in the present invention samples, extracts data and extracts structural features of the inversion results of all-space fracture parameters, acoustic emission scalar parameters and models through Bayesian networks, and obtains the influencing factors in the real-time quantitative inversion model of all-space fracture parameters. The model inversion mechanism interpreter can directly obtain the inversion mechanism model of all-space fracture parameters in the process of coal and rock mass loading through acoustic emission data and the prediction results of each model, revealing the black box process of fracture parameter inversion of coal and rock mass in the process of loading.

[0067] The best embodiment for carrying out the present invention will be described in more detail below with reference to the accompanying drawings so that the features and advantages of the present invention can be easily understood. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings of the embodiments of the present invention, wherein the drawings are only used to illustrate some embodiments of the present invention, but not to limit all embodiments of the present invention thereto.

[0069] Figure 1 It is a flow chart of a method for coal-rock full-time and space fracture field inversion using an interpretable graph neural network according to an embodiment of the present invention;

[0070] Figure 2 It is a schematic diagram of the arrangement structure of the coal rock mass exploration-monitoring system according to an embodiment of the present invention;

[0071] Figure 3 A schematic diagram of the structure of a real-time quantitative inversion model for the full-time and space fracture field evolution process according to an embodiment of the present invention;

[0072] Figure 4 A flow chart of the positioning and inversion mechanism is obtained for the interpreter according to an embodiment of the present invention.

[0073] List of reference numerals:

[0074] CT detection sensor 10;

[0075] Acoustic emission monitoring sensor 20;

[0076] X-ray detection sensor 21;

[0077] X-ray source 22;

[0078] Coal and rock mass 30;

[0079] Testing machine 40;

[0080] Coupling device 50. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the technical solution of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described in conjunction with the drawings of specific embodiments of the present invention. The same figure marks in the drawings represent the same parts. It should be noted that the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0082] Unless otherwise defined, the technical terms or scientific terms used herein shall have the usual meanings understood by persons of ordinary skill in the field to which the present invention belongs. The words "first", "second" and similar words used in the patent application specification and claims of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not necessarily indicate a quantity limitation. Words such as "include" or "comprise" mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0083] According to the first aspect of the present invention, a coal-rock full-time and space fracture field inversion method using an interpretable graph neural network is provided. Figures 1 to 4 As shown, the following steps are included:

[0084] S10: arranging a CT detection sensor 10 and an acoustic emission monitoring sensor 20 on the coal-rock mass 30, and placing the coal-rock mass 30 in a testing machine 40 for a load fracture test, and collecting CT detection information at a certain stage and full-time and full-space acoustic emission monitoring information during the load fracture process of the coal-rock mass 30 through the CT detection sensor 10 and the acoustic emission monitoring sensor 20; wherein the CT detection information includes full-space fracture parameter information at a certain stage, and the acoustic emission monitoring information includes full-time and full-space waveform information, full-time and full-space event information, and full-time and full-space scalar parameter information; for example, Figure 2 As shown, the testing machine 40 is placed in a closed box, and the coal rock body 30 is arranged in the testing machine 40, wherein the acoustic emission monitoring sensor 20 is installed in the coal rock body 30, and the CT detection sensor 10 includes an X-ray detection sensor 21 and an X-ray source 22 and is arranged outside the coal rock body 30, the X-ray detection sensor 21 and the X-ray source 22 are respectively placed on both sides of the coal rock body 30, and the coupling device 50 is also installed in the coal rock body 30 and arranged opposite to the acoustic emission monitoring sensor 20.

[0085] S20: matching the full-space fracture parameter information and the full-space acoustic emission monitoring information at the stage moments in the process of the coal-rock mass 30 being subjected to load fracture according to the time, to obtain a fracture parameter data set; dividing the fracture parameter data set into a training set, a validation set, and a test set according to corresponding proportions;

[0086] S30: The graph neural network extracts features from the full-space fracture parameter data at staged moments in the training set and the validation set of the fracture parameter data set and the acoustic emission monitoring data at the corresponding moments, and trains to obtain a real-time quantitative inversion model for full-time and space fracture parameters; the full-time and space waveform data and scalar parameter data in the test set are input into the inversion model to obtain the full-time and space fracture parameter inversion results, i.e., the fracture parameters of the full-time and space fracture field during the loading process; and the full-time and space fracture field parameter evolution process of the coal rock mass 30 under loading is obtained;

[0087] S40: The scalar parameter data of the full-time and space acoustic emission monitoring data in the test set of the fracture parameter data set and the full-time and space fracture parameter inversion results are input into the graph neural network interpreter based on the Bayesian network to obtain the influencing factors between the acoustic emission scalar parameters and the fracture parameter inversion results, and then obtain the real-time quantitative inversion mechanism model of the fracture parameters based on the acoustic emission scalar parameters.

[0088] This inversion method focuses on the deep learning method to invert the full-time and space fracture field of the loaded coal-rock mass 30 through the detection and monitoring information in the coal-rock mass 30 load fracture experiment, and combines the graph neural network and the Bayesian network to reveal the mechanism model of quantitative inversion of full-time and space fracture parameters, and achieve high-precision inversion and full-time and space reproduction of the fracture field and its evolution of the loaded coal-rock mass 30. Compared with traditional numerical simulation methods, this coal-rock full-time and space fracture field inversion method and system with interpretable graph neural network has higher computational efficiency and better adaptability, and can effectively make up for the limitations of traditional methods.

[0089] The real-time quantitative inversion model of all-time and space fracture parameters in the inversion method is obtained by joint feature extraction of fracture parameters through graph neural network, and a fracture parameter feature extraction method for acoustic emission information during the loading process of coal rock mass 30 is proposed. Through this fracture parameter extraction method, the fracture parameters can be inverted through the acoustic emission information during the loading process of coal rock mass 30 obtained in the laboratory, which lays a certain foundation for the subsequent all-time and space fracture field inversion and obtaining the evolution process of all-time and space fracture field parameters of coal rock mass 30 under loading.

[0090] The full-time and space crack parameter real-time quantitative inversion mechanism interpreter of the inversion method samples, extracts data and extracts structural features of the full-space crack parameter, acoustic emission scalar parameter and the inversion results of the model through the Bayesian network, and obtains the influencing factors in the full-time and space crack parameter real-time quantitative inversion model. The model inversion mechanism interpreter can directly obtain the full-time and space crack parameter inversion mechanism model during the loading process of the coal rock mass 30 through the acoustic emission data and the prediction results of each model, and reveal the black box process of the crack parameter inversion of the coal rock mass 30 during the loading process.

[0091] In one example of the present invention, Figure 2As shown, in the step S10, before placing the coal rock body 30 in the testing machine 40 for load fracture test, the following steps are also included: connecting the acoustic emission probe with the coupling device 50, and applying coupling agent to the probe of the coupling device 50 and arranging it on the surface of the coal rock body 30; setting the basic parameters of the acoustic emission and the monitoring information to be collected; setting the basic parameters of the CT detector and the detection information to be collected; setting the loading mode of the testing machine 40 and the data to be collected.

[0092] In one example of the present invention, Figure 3 As shown, in step S30, the graph neural network extracts features of the full-space crack parameters at staged moments and the acoustic emission monitoring information data at corresponding moments in the training set and the validation set of the crack parameter data set, including the following steps:

[0093] Encoding the acoustic emission data into acoustic emission graph structure data through a coding layer;

[0094] The fusion layer fuses the waveform and scalar parameters in the input acoustic emission image structure data;

[0095] The feature extraction layer extracts the features of the result after scalar parameter data fusion, extracts node features and obtains new acoustic emission waveforms and scalar parameter features;

[0096] The weight vector The parameterized self-attention mechanism extracts higher-dimensional features;

[0097] All selections at time j are normalized by the Softmax function, and the feature similarity of the acoustic emission scalar parameters at different times is calculated;

[0098] The corresponding linear combination of characteristics is obtained by coefficient calculation, and the characteristics of the final output of the scalar parameters of the acoustic emission position at different times are obtained;

[0099] Connect multiple self-attention mechanisms through the average calculation method to stabilize the calculation process and obtain the final features extracted by the multi-head graph attention mechanism;

[0100] The features extracted by the multi-head graph attention mechanism are normalized through the Layer Normalization function to improve training stability and speed;

[0101] The features extracted by the Layer Normalization function are linearly transformed through the fully connected layer to extract higher-dimensional features;

[0102] The features extracted by the fully connected layer are regressed through the Readout function to obtain the regression parameters of each graph structure.

[0103] In one example of the present invention, the expression for extracting features from the result of scalar parameter data fusion through the feature extraction layer is as follows:

[0104]

[0105] Where h represents the acoustic emission scalar parameter in the acoustic emission graph structure of the input graph attention neural network; F represents the number of acoustic emission characteristic parameters; h ′ The acoustic emission scalar parameter in the new graph structure representing the output of the feature extraction layer; F ′ The number of AE scalar parameters representing potentially different cardinalities.

[0106] In one example of the present invention, the expression of extracting higher-dimensional features by the self-attention mechanism is as follows:

[0107]

[0108] j∈N i

[0109]

[0110] Among them, e ij Represents the characteristic similarity of the acoustic emission waveform at time i to time j; N i represents the neighborhood at time j; a represents the self-attention mechanism, which consists of a single-layer feedforward neural network; represents the weight vector; W represents the weight matrix; and They represent the acoustic emission scalar parameters at different times in the acoustic emission image structure data.

[0111] In one example of the present invention, in step S30, the graph neural network extracts features from the full-space crack parameter data at the stage moment and the acoustic emission data at the corresponding moment in the training set and the validation set of the crack parameter data set, including:

[0112] The square difference between the true value and the inversion value of the full-space fracture parameter is calculated by the mean square error loss function, and the model is optimized according to the function calculation value. The calculation formula of the mean square error loss function is as follows:

[0113]

[0114] Among them, MSE represents the calculation result of the mean square error loss function; Represents the true value of the crack parameter in the entire space; Represents the inversion value of the full-space fracture parameters.

[0115] In one example of the present invention, in step S40, all acoustic emission graph structure data G in the test set are input into Φ I The output of the model is T I , the calculation formula is as follows:

[0116] Φ I :G→T I

[0117] Among them, Φ I represents the full-time and space crack quantitative inversion model; G represents all acoustic emission structure data; T I Represents the inversion space of the full-space and time-fracture quantitative inversion model.

[0118] In one example of the present invention, Figure 4 As shown, the step S40 specifically includes the following:

[0119] S401: Encode the acoustic emission data in the test set into acoustic emission graph structure data G and input it into the graph neural network interpreter. The data generation module activates the acoustic emission feature v at each moment in G. The data generation module controls the activation probability of the acoustic emission feature in the graph structure through the parameter p.

[0120] S402: Using the variable screening module to filter D I The variable data in the extractor is used to speed up the mechanism extraction efficiency of the interpreter. The variable screening module extracts important variables I, MB through the Markov layer P (I) is the minimum set of variables obtained by screening, denoted as U(I), I is independent of other variables in P. At the same time, Set the random variable s in v Perfect mapping of distribution B * , Provided separately with Same statistics;

[0121] S403: The interpreter passes the BIC function D I Learning to interpret Bayesian networks;

[0122] S404: The full-time and space fracture parameter inversion mechanism model is obtained according to the influencing factors output by the interpreter. The calculation formula is as follows:

[0123] M I :T I =m1×M1+m2×M2+…+m n ×M n

[0124] Among them, M nrepresents the acoustic emission scalar parameter in the full-time and space fracture parameter inversion mechanism; M I Represents the full-time and space fracture parameter inversion mechanism model.

[0125] In an example of the present invention, step S401 includes the following steps:

[0126] Set the random variable s v To indicate whether the acoustic emission feature v is activated, set s = {s v}, get the activated acoustic emission graph structure data G(s), input G(s) into Φ I The output of the model is Φ I (G(s)), calculate Φ through Ι(·) I (G(s)) and Φ I The difference between (G) and (s) is used to obtain the target variable I of the real-time quantitative inversion model of all-time and space crack parameters based on G(s). The acoustic emission scalar parameter variable at each moment is set to Δv = {s v}, denoted as D I To ensure that the acoustic emission scalar parameter at each moment has an impact on the inversion result, the neighborhood moment of the acoustic emission scalar parameter at each moment is Set to V, after n samplings, D I The size of is n×(|V|+1), and the calculation formula is as follows:

[0127] p∈(0,1)

[0128] I={Ι(Φ I (G(s)))};Φ I :g→T I

[0129] Δv={s v}; v∈V

[0130] G(s); G(s)∈g

[0131] Among them, the parameter p represents the activation probability of the feature in the graph structure; Ι(·) represents the calculation of Φ I (G(s)) and Φ I (G) Function of whether there is a significant difference; g represents all the acoustic emission map structure data in the dataset.

[0132] In one example of the present invention, in step S403, the interpreter uses the BIC function D I Learn to explain the Bayesian network in the following formula:

[0133]

[0134] in, represents the objective function for calculating the impact factor of each acoustic emission feature; BIC represents the function for calculating the impact factor; B represents the Bayesian network; θ B represents the Bayesian network parameters; l(θ B :D I [U(I)]) represents θ B and D I [U(I)]; Dim[B] represents the dimension of the Bayesian network; m n The acoustic emission scalar parameter influence factor representing the full-time and space fracture parameter inversion mechanism model; represents the parameter value that maximizes the log-likelihood (maximum likelihood estimator); represents the full-time and space-time crack parameter inversion mechanism obtained by the interpreter; ε represents the interpretable domain; V(B) represents the set of random variables in the Bayesian network; M represents the number of variables in B.

[0135] According to the second aspect of the present invention, a coal-rock full-time and space fracture field inversion system using an interpretable graph neural network comprises:

[0136] The information acquisition module is configured to arrange a CT detection sensor 10 and an acoustic emission monitoring sensor 20 on the coal-rock mass 30, and place the coal-rock mass 30 in a testing machine 40 for a load fracture test, and acquire the CT detection information at the stage moment and the full-time and full-space acoustic emission monitoring information during the load fracture process of the coal-rock mass 30 through the CT detection sensor 10 and the acoustic emission monitoring sensor 20; wherein the CT detection information includes the full-space fracture parameter information at the stage moment, and the acoustic emission monitoring information includes the full-time and full-space waveform information, the full-time and full-space event information, and the full-time and full-space scalar parameter information;

[0137] The data set division module is configured to correspond the full-space fracture parameter information and the full-time and space acoustic emission monitoring information at the stage moments in the process of the coal rock mass (30) being subjected to load and fracturing according to the time to obtain a fracture parameter data set; and to divide the fracture parameter data set into a training set, a validation set and a test set according to corresponding proportions;

[0138] The model building and inversion module is configured to extract features of the full-space fracture parameter data and the acoustic emission monitoring data at the corresponding time in the training set and the verification set of the fracture parameter data set through a graph neural network, and train to obtain a real-time quantitative inversion model of the full-time and space fracture parameters; input the full-time and space waveform data and scalar parameter data in the test set into the inversion model to obtain the full-time and space fracture parameter inversion results, that is, the fracture parameters of the full-time and space fracture field during the loading process; and obtain the full-time and space fracture field parameter evolution process of the coal rock mass (30) under loading;

[0139] The inversion mechanism model module is configured to input the scalar parameter data of the full-time and space acoustic emission monitoring data in the test set of the fracture parameter data set and the full-time and space fracture parameter inversion results into a graph neural network interpreter based on a Bayesian network, obtain the influencing factors between the acoustic emission scalar parameters and the fracture parameter inversion results, and then obtain a real-time quantitative inversion mechanism model of fracture parameters based on the acoustic emission scalar parameters.

[0140] The inversion system focuses on the deep learning method to invert the full-time and space fracture field of the loaded coal-rock mass 30 through the detection and monitoring information in the coal-rock mass 30 load fracture experiment, and combines the graph neural network and the Bayesian network to reveal the mechanism model of quantitative inversion of full-time and space fracture parameters, and realize the high-precision inversion and full-time and space reproduction of the fracture field and its evolution of the loaded coal-rock mass 30. Compared with the traditional numerical simulation method, this coal-rock full-time and space fracture field inversion method and system with interpretable graph neural network has higher computational efficiency and better adaptability, and can effectively make up for the limitations of traditional methods.

[0141] The real-time quantitative inversion model of all-time and space fracture parameters in the inversion system is obtained by joint feature extraction of fracture parameters through graph neural network, and a fracture parameter feature extraction method for acoustic emission information during the loading process of coal rock mass 30 is proposed. Through this fracture parameter extraction method, the fracture parameters can be inverted through the acoustic emission information during the loading process of coal rock mass 30 obtained in the laboratory, which lays a certain foundation for the subsequent all-time and space fracture field inversion and obtaining the evolution process of all-time and space fracture field parameters of coal rock mass 30 under loading.

[0142] The real-time quantitative inversion mechanism interpreter of the full-time and space crack parameters of the inversion system samples, extracts data and extracts structural features of the inversion results of the full-space crack parameters, acoustic emission scalar parameters and models through the Bayesian network, and obtains the influencing factors in the real-time quantitative inversion model of the full-time and space crack parameters. The model inversion mechanism interpreter can directly obtain the full-time and space crack parameter inversion mechanism model during the loading process of the coal rock mass 30 through the acoustic emission data and the prediction results of each model, revealing the black box process of the crack parameter inversion of the coal rock mass 30 during the loading process.

[0143] The exemplary implementation scheme of the coal rock full-time and space fracture field inversion method and system of the interpretable graph neural network proposed in the present invention is described in detail above with reference to the preferred embodiments. However, those skilled in the art will understand that, without departing from the concept of the present invention, various modifications and variations can be made to the above-mentioned specific embodiments, and various technical features and structures proposed in the present invention can be combined in various ways without exceeding the protection scope of the present invention, which is determined by the attached claims.

Claims

1. A coal-rock full-time and space fracture field inversion method based on an interpretable graph neural network, characterized in that: The steps include: S10: arranging a CT detection sensor (10) and an acoustic emission monitoring sensor (20) on the coal-rock mass (30), and placing the coal-rock mass (30) in a testing machine (40) to perform a load-bearing fracture test, and collecting CT detection information at staged moments and full-time and full-space acoustic emission monitoring information during the load-bearing fracture process of the coal-rock mass (30) through the CT detection sensor (10) and the acoustic emission monitoring sensor (20); wherein the CT detection information includes full-space fracture parameter information at staged moments, and the acoustic emission monitoring information includes full-time and full-space waveform information, full-time and full-space event information, and full-time and full-space scalar parameter information; S20: matching the full-space fracture parameter information and the full-space acoustic emission monitoring information at the stage moments in the process of the coal rock mass (30) being subjected to load fracture according to the time, to obtain a fracture parameter data set; and dividing the fracture parameter data set into a training set, a validation set and a test set according to corresponding proportions; S30: The graph neural network extracts features from the full-space fracture parameter data at staged moments in the training set and the validation set of the fracture parameter data set and the acoustic emission monitoring data at the corresponding moments, and trains to obtain a real-time quantitative inversion model for the full-space fracture parameters; the full-space waveform data and scalar parameter data in the test set are input into the inversion model to obtain the full-space fracture parameter inversion results, i.e., the fracture parameters of the full-space fracture field during the loading process; and the full-space fracture field parameter evolution process of the coal rock mass (30) under loading is obtained; S40: The scalar parameter data of the full-time and space acoustic emission monitoring data in the test set of the fracture parameter data set and the full-time and space fracture parameter inversion results are input into the graph neural network interpreter based on the Bayesian network to obtain the influencing factors between the acoustic emission scalar parameters and the fracture parameter inversion results, and then obtain the real-time quantitative inversion mechanism model of the fracture parameters based on the acoustic emission scalar parameters.

2. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 1 is characterized in that: In step S30, the graph neural network extracts features of the full-space crack parameters at staged moments and the acoustic emission monitoring information data at corresponding moments in the training set and the validation set of the crack parameter data set, including the following steps: Encoding the acoustic emission data into acoustic emission graph structure data through a coding layer; The fusion layer fuses the waveform and scalar parameters in the input acoustic emission image structure data; The feature extraction layer extracts the features of the result after scalar parameter data fusion, extracts node features and obtains new acoustic emission waveforms and scalar parameter features; The weight vector The parameterized self-attention mechanism extracts higher-dimensional features; All selections at time j are normalized by the Softmax function, and the feature similarity of the acoustic emission scalar parameters at different times is calculated; The corresponding linear combination of characteristics is obtained by coefficient calculation, and the characteristics of the final output of the scalar parameters of the acoustic emission position at different times are obtained; Connect multiple self-attention mechanisms through the average calculation method to stabilize the calculation process and obtain the final features extracted by the multi-head graph attention mechanism; The features extracted by the multi-head graph attention mechanism are normalized through the Layer Normalization function to improve training stability and speed; The features extracted by the Layer Normalization function are linearly transformed through the fully connected layer to extract higher-dimensional features; The features extracted by the fully connected layer are regressed through the Readout function to obtain the regression parameters of each graph structure.

3. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 2 is characterized in that: The expression for feature extraction of the result after fusion of scalar parameter data through the feature extraction layer is as follows: Where h represents the acoustic emission scalar parameter in the acoustic emission graph structure of the input graph attention neural network; F represents the number of acoustic emission scalar parameters; h ′ The acoustic emission scalar parameter in the new graph structure representing the output of the feature extraction layer; F ′ The number of AE scalar parameters representing potentially different cardinalities.

4. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 2 is characterized in that: The expression of higher-dimensional features extracted by the self-attention mechanism is as follows: j∈N i Among them, e ij Represents the characteristic similarity of the acoustic emission waveform at time i to time j; N i represents the neighborhood at time j; a represents the self-attention mechanism, which consists of a single-layer feedforward neural network; represents the weight vector; W represents the weight matrix; and They represent the acoustic emission scalar parameters at different times in the acoustic emission image structure data.

5. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 1 is characterized in that: In step S30, the graph neural network extracts features from the full-space crack parameter data at each stage and the acoustic emission data at the corresponding time in the training set and the validation set of the crack parameter data set, including: The square difference between the true value and the inversion value of the full-space fracture parameter is calculated by the mean square error loss function, and the model is optimized according to the function calculation value. The calculation formula of the mean square error loss function is as follows: Among them, MSE represents the calculation result of the mean square error loss function; T Ii Represents the true value of the crack parameter in the entire space; Represents the inversion value of the full-space fracture parameters.

6. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 1 is characterized in that: In step S40, all acoustic emission graph structure data G in the test set are input into Φ I The output of the model is T I , the calculation formula is as follows: F I :G→T I Among them, Φ I represents the full-time and space crack quantitative inversion model; G represents all acoustic emission structure data; T I Represents the inversion space of the full-space and time-fracture quantitative inversion model.

7. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 1 is characterized in that: The step S40 specifically includes the following: S401: Encode the acoustic emission data in the test set into acoustic emission graph structure data G and input it into the graph neural network interpreter. The data generation module activates the acoustic emission feature v at each moment in G. The data generation module controls the activation probability of the acoustic emission feature in the graph structure through the parameter p. S402: Using the variable screening module to filter D I The variable data in the extractor is used to speed up the mechanism extraction efficiency of the interpreter. The variable screening module extracts important variables I, MB through the Markov layer P (I) is the minimum set of variables obtained by screening, denoted as U(I), I is independent of other variables in P. At the same time, Set the random variable s in v Perfect mapping of distribution B * , Provided separately with Same statistics; S403: The interpreter passes the BIC function D I Learning to interpret Bayesian networks; S404: The full-time and space fracture parameter inversion mechanism model is obtained according to the influencing factors output by the interpreter. The calculation formula is as follows: M I ∶T I =m1×M1+m2×M2+…+m n ×M n Among them, M n represents the acoustic emission scalar parameter in the full-time and space fracture parameter inversion mechanism; M I Represents the full-time and space fracture parameter inversion mechanism model.

8. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 7 is characterized in that: The step S401 includes the following steps: Set the random variable s v To indicate whether the acoustic emission feature v is activated, set s = {s v }, obtain the activated acoustic emission graph structure data G(s); Input G(s) into Φ I The output of the model is Φ I (G(s)), calculate Φ through Ι(·) I (G(s)) and Φ I The difference between (G) and (s) is used to obtain the target variable I of the real-time quantitative inversion model of all-time and space fracture parameters based on G(s); The acoustic emission scalar parameter variable at each moment is set to Δv = {s v }, denoted as D I ; The neighborhood time of the acoustic emission scalar parameter at each moment Set to V, after n samplings, D I The size of is n×(|V|+1), and the calculation formula is as follows: p∈(0，1) I={Ι(Φ I (G(s)))};Φ I :g→T I Δv={s v };v∈V G(s); G(s)∈g Among them, the parameter p represents the activation probability of the feature in the graph structure; Ι(·) represents the calculation of Φ I (G(s)) and Φ I (G) Function of whether there is a significant difference; g represents all the acoustic emission map structure data in the dataset.

9. The coal-rock full-time and space fracture field inversion method using an interpretable graph neural network according to claim 7 is characterized in that: In step S403, the interpreter uses the BIC function D I Learn to explain the Bayesian network in the following formula: in, represents the objective function for calculating the impact factor of each acoustic emission feature; BIC represents the function for calculating the impact factor; B represents the Bayesian network; θ B represents the Bayesian network parameters; l(θ B :D I [U(I)]) represents θ B and D I [U(I)]; Dim[B] represents the dimension of the Bayesian network; m n The acoustic emission scalar parameter influence factor representing the full-time and space fracture parameter inversion mechanism model; represents the parameter value that maximizes the log-likelihood (maximum likelihood estimator); represents the full-time and space-time crack parameter inversion mechanism obtained by the interpreter; ε represents the interpretable domain; V(B) represents the set of random variables in the Bayesian network; M represents the number of variables in the restriction B.

10. A coal-rock full-time and space fracture field inversion system based on an interpretable graph neural network, characterized in that: include: The information collection module is configured to arrange a CT detection sensor (10) and an acoustic emission monitoring sensor (20) on a coal-rock mass (30), and place the coal-rock mass (30) in a testing machine (40) to perform a load-bearing fracture test, and to collect CT detection information at a certain time and full-time and full-space acoustic emission monitoring information during the load-bearing fracture process of the coal-rock mass (30) through the CT detection sensor (10) and the acoustic emission monitoring sensor (20); wherein the CT detection information includes full-space fracture parameter information at a certain time, and the acoustic emission monitoring information includes full-time and full-space waveform information, full-time and full-space event information, and full-time and full-space scalar parameter information; The data set division module is configured to correspond the full-space fracture parameter information and the full-time and space acoustic emission monitoring information at the stage moments in the process of the coal rock mass (30) being subjected to load and fracturing according to the time to obtain a fracture parameter data set; and to divide the fracture parameter data set into a training set, a validation set and a test set according to corresponding proportions; The model building and inversion module is configured to extract features of the full-space fracture parameter data and the acoustic emission monitoring data at the corresponding time in the training set and the verification set of the fracture parameter data set through a graph neural network, and train to obtain a real-time quantitative inversion model of the full-time and space fracture parameters; input the full-time and space waveform data and scalar parameter data in the test set into the inversion model to obtain the full-time and space fracture parameter inversion results, that is, the fracture parameters of the full-time and space fracture field during the loading process; and obtain the full-time and space fracture field parameter evolution process of the coal rock mass (30) under loading; The inversion mechanism model module is configured to input the scalar parameter data of the full-time and space acoustic emission monitoring data in the test set of the fracture parameter data set and the full-time and space fracture parameter inversion results into a graph neural network interpreter based on a Bayesian network, obtain the influencing factors between the acoustic emission scalar parameters and the fracture parameter inversion results, and then obtain a real-time quantitative inversion mechanism model of fracture parameters based on the acoustic emission scalar parameters.

Images