Coal rock implicit evolution damage constitutive model construction method and system based on ensemble learning

Abstract

The present application relates to the technical field of rock mass mechanical property prediction, in particular to a coal rock implicit evolution damage constitutive model construction method and system based on ensemble learning, comprising the following steps: S1, obtaining coal rock parameters and constructing a multi-dimensional damage feature index system and a sample data set; S2, preprocessing the sample and training multiple neural networks, optimizing the combination base learner and the meta-learner, and building an ensemble learning framework; S3, inputting multi-dimensional indexes for iterative training to construct an implicit evolution damage model; S4, verifying the model accuracy and comparing and analyzing the test and simulation results; S5, inputting nondestructive testing and environmental data to output the full stress-strain curve to evaluate the mechanical properties of coal rock. Through ensemble learning and loading path modeling, the present application realizes nondestructive and accurate prediction of the stress-strain curve of coal rock under multiple working conditions, significantly improving the intelligence and applicability of mechanical property evaluation.

Description

Technical Field

[0001] This invention relates to the field of rock mass mechanical property prediction technology, and in particular to a method and system for constructing implicit evolutionary damage constitutive models of coal and rock based on ensemble learning. Background Technology

[0002] During coal mining, long-term immersion in mine water significantly alters the physical and mechanical properties of coal and rock, leading to reduced strength, deterioration of microstructure, and development of pores and fractures. This, in turn, induces various geological disasters such as coal and rock pillar instability, roadway collapse, and water inrush and sand collapse, seriously threatening the safety and stability of mining projects. Currently, the assessment of the mechanical properties of coal and rock mainly relies on indoor mechanical tests. Although these tests can obtain strength parameters and stress-strain curves with a certain degree of accuracy, they are difficult to achieve in-situ non-destructive acquisition and rapid prediction of the mechanical properties of coal and rock under complex environments due to their inherent defects such as strong destructiveness, long testing cycle, high cost, and limited working conditions.

[0003] On the other hand, coal and rock are subjected to multi-field coupling in complex underground environments, and their damage evolution process exhibits highly nonlinear and path-dependent characteristics. Traditional coupled constitutive models are difficult to accurately characterize them. Existing models generally suffer from problems such as complex structure, too many parameters, and poor adaptability, making it difficult to meet the dual requirements of efficient prediction and engineering practicality. At the same time, although artificial intelligence technology has been widely used in geotechnical prediction modeling in recent years, coal and rock damage prediction methods based on single neural networks still have limitations such as weak generalization ability and serious overfitting. A stable prediction mechanism that can be widely applied to multiple working conditions and multiple loading paths has not yet been formed. Summary of the Invention

[0004] This invention provides a method and system for constructing a implicit evolutionary damage constitutive model for coal and rock based on ensemble learning. By integrating multi-dimensional indicators such as the properties of coal and rock itself, groundwater environmental factors, and stress loading paths, an ensemble learning framework composed of multiple neural network algorithms is constructed. Path prediction samples are iteratively generated through a strain-controlled loading strategy, and a implicit evolutionary damage model with path memory characteristics is trained to achieve non-destructive intelligent prediction of the full stress-strain curve of coal and rock. This significantly improves prediction accuracy, adaptability, and engineering practicality, and is applicable to scenarios such as in-situ assessment of the mechanical properties of coal and rock and early warning of geological disaster risks.

[0005] A method for constructing a implicit evolutionary damage constitutive model for coal and rock based on ensemble learning includes the following steps: S1. Through indoor simulation experiments and related tests, basic parameters of coal and rock are obtained. Combined with the coal and rock damage evolution mechanism and failure characteristics, a multidimensional coal and rock damage characteristic index system and sample dataset are established. S2, the sample dataset is preprocessed and divided into training set, test set and prediction set. By training and comparing various neural network algorithms, the base learner and meta learner are determined, and the key parameters of the learner are globally optimized to build an integrated learning framework. S3, by inputting a multidimensional coal and rock damage feature index system, the integrated learning framework is iteratively trained and debugged to construct a coal and rock implicit evolutionary damage model based on an integrated neural network; S4. The accuracy of the implicit evolutionary damage model of coal and rock is verified by using the test set. The test results are compared and analyzed with the results of indoor experiments and numerical simulations to obtain the prediction accuracy of the implicit evolutionary damage model of coal and rock. S5 inputs the non-destructive testing parameters and environmental index data of coal and rock under different working conditions into the implicit evolutionary damage model of coal and rock to obtain the full stress-strain curve of coal and rock damage and evaluate the mechanical properties of coal and rock.

[0006] Optionally, S1 includes: S11. Collect representative coal and rock samples from coal seam 16 in the mining area and their corresponding groundwater samples. The groundwater samples include Ordovician limestone water, Lithium 14 limestone water, and Jurassic sandstone water. Supplement the collection of water samples from the water inrush area and the goaf area. Determine the mineral composition of the coal and rock by XRD diffraction analysis. Process the coal and rock samples into standard specimens of Ф50×100 mm and test their physical properties and water quality parameters. S12, coal and rock immersion tests were conducted using solutions of different concentrations, pH values, flow rates, temperatures, and water pressures, as well as mine water. Uniaxial compression, nuclear magnetic resonance testing, and acoustic detection were used to collect the mechanical and microstructural parameters of coal and rock under different aqueous solutions and immersion times. S13. Based on the collected coal and rock mechanical parameters and microstructure parameters, and combined with PFC discrete element numerical simulation, a multidimensional coal and rock damage characteristic index system is constructed.

[0007] Optionally, the multidimensional coal and rock damage characteristic index system includes coal and rock intrinsic property indexes, groundwater environmental factor indexes, and stress-strain process indexes. The intrinsic properties of the coal and rock are expressed as follows: ; in, r For the density of the rock mass, v p For the speed of sound waves, f Porosity D w Damage caused by water factors, D l For load damage, D T Total damage, D It is the fractal dimension; The groundwater environmental factor indicators are expressed as follows: ; in, T Soaking time, Na is Na + Concentration, Ca is Ca 2+ Concentration, Mg is Mg 2+ Concentration, pH refers to pH value. v The velocity of the water flow; The stress-strain process index is expressed as follows: ; in, e i-2 For the first two strains, s i-2 For the first two stresses, e i-1 As a contingency plan, s i-1 For the previous stress, e i In response to the current situation, s i This represents the current stress.

[0008] Optionally, the rock mass density r、 sound wave speed v p and porosity f Obtained through indoor simulation experiments; The water factor damage D w Represented as: ; in, S T0 This represents the initial NMR peak area. S T The area of ​​the nuclear magnetic resonance peaks after water-rock interaction. a and b These are the conversion coefficients between water content and NMR peak area, respectively. V 总 For the volume of the rock; The load damage D l Represented as: ; in, e and e p These are strain and peak strain, respectively. m For shape factor, exp It is an exponential function; The total damage DT Represented as: ; The fractal dimension D Represented as: ; in, T 2 is the NMR peak. T 2max The largest NMR peak, W v For the lateral relaxation time to be less than T 2% of the cumulative pore volume. lg It is a logarithmic function.

[0009] Optionally, S2 includes: S21, Set the input and output samples according to the sample dataset; The input sample is represented as follows: ; The output sample is represented as: ; in, For the first i One input sample, y i For the first i The current stress corresponding to each sample; S22, using the training set Five-fold cross-validation was used to train various neural network algorithms to obtain prediction results. And by using BP, RF, SVM and LSTM as base learners, new training sets are generated. As training samples for the meta-learner, among which... ; S23, use the trained base learner to compute the predicted values ​​on the test set. P te A new test set is constructed by combining the sample dataset. As test samples for meta-learners By using Taylor charts to compare the performance of base learners, the base learners and meta-learners are determined, and an ensemble learning framework is established.

[0010] Optionally, S3 includes: S31 employs a strain-controlled stress method, setting the strain increment for each loading step to 0.1%. In the first loading step, the stress and strain of the first two steps are... All are 0, current strain The current stress value is 0.1%, and it is expressed as follows: ; S32, when predicting the current stress value in the second loading step e i-1 and s i-1 The stress values ​​predicted in the first loading step and 0.1% were taken respectively. Y 1. Current Response The value is 0.2%, and the current stress value of the second loading step is output as follows: ; S33, when predicting the current stress value in the third loading step. e i-2 and s i-2 The stress values ​​predicted in the first loading step and 0.1% were taken respectively. Y 1, e i-1 and s i-1 The stress values ​​predicted in the second loading step were taken as 0.2% and 2% respectively. Y 2. Current Response The value is 0.3%, and the current stress value of the third loading step is output as follows: ; S34, through continuous iteration, the strain value is gradually reduced until the specimen is loaded to failure, thus completing the stress-strain path prediction and constructing a constitutive model of implicit evolutionary damage in coal and rock.

[0011] Optionally, the stress-strain path prediction is expressed as: .

[0012] Optionally, S4 includes: S41, set the meta-learner to a CNN model, in which convolution, ReLU activation function and max pooling function are used in sequence in the hidden layer, the loss function is mean squared error, and the learning rate is set to 0.005. When the error convergence curve of the coal and rock implicit evolutionary damage constitutive model is less than 0.1, the iterative training is terminated. S42. Stress-strain curves predicted by the implicit evolutionary damage constitutive model of coal and rock, obtained from laboratory tests and numerical simulations, are plotted respectively. The accuracy of the implicit evolutionary damage constitutive model of coal and rock is evaluated based on the overlap of the stress-strain curves.

[0013] The system for constructing a coal and rock implicit evolutionary damage constitutive model based on ensemble learning is used to implement the aforementioned method for constructing a coal and rock implicit evolutionary damage constitutive model based on ensemble learning. It includes the following modules: Data acquisition module: Through indoor simulation experiments and related tests, acquire basic coal and rock parameter data, and combine coal and rock damage evolution mechanism and failure characteristics to establish a multidimensional coal and rock damage characteristic index system and sample dataset; Data processing module: preprocesses the sample dataset, divides it into training set, test set and prediction set, trains and compares various neural network algorithms to determine the base learner and meta learner, and globally optimizes the key parameters of the learner to build an integrated learning framework. Model training module: The integrated learning framework is iteratively trained and debugged by inputting a multidimensional coal and rock damage feature index system to construct a coal and rock implicit evolution damage model based on an integrated neural network; Performance evaluation module: The accuracy of the implicit evolutionary damage model of coal and rock is verified by using a test set. The test results are compared and analyzed with the results of indoor experiments and numerical simulations to obtain the prediction accuracy of the implicit evolutionary damage model of coal and rock. Damage prediction module: Input the non-destructive testing parameters and environmental index data of coal and rock under different working conditions into the implicit evolutionary damage model of coal and rock to obtain the full stress-strain curve of coal and rock damage and evaluate the mechanical properties of coal and rock.

[0014] The beneficial effects of this invention are: This invention integrates multi-source experimental data with the coal and rock damage evolution mechanism to construct a multi-dimensional coal and rock damage characteristic index system. It also introduces an integrated learning framework composed of multiple neural network models to realize stress-strain response modeling of coal and rock in complex hydrogeological environments. This invention breaks through the limitation of traditional methods that rely on destructive experiments to predict the mechanical properties of coal and rock, and can effectively improve the prediction accuracy and applicability.

[0015] This invention constructs an implicit evolutionary damage model for coal and rock with a path memory mechanism by designing an explicit loading path structure input. It is then combined with a CNN meta-learner for fusion optimization, enabling rapid output of stress-strain curves for coal and rock under any working conditions. This provides an intelligent and highly generalizable modeling solution for in-situ non-destructive prediction of coal and rock damage state and mechanical properties, and has good engineering practical value and promotion potential. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the construction method flow according to an embodiment of the present invention; Figure 2This is a schematic diagram of the multi-scale coal and rock damage index evaluation system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of stress-strain process index extraction according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the calculation process of the implicit evolutionary damage model of coal and rock according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the construction process of the implicit evolutionary damage model for coal and rock based on ensemble learning, according to an embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the comparison between the model's predicted values ​​and the actual values ​​in an embodiment of the present invention. Figure 7 This is a schematic diagram comparing the full stress-strain curves of predicted values, experimental values, and simulated values ​​in an embodiment of the present invention. Figure 8 This is a schematic diagram of the system functional modules according to an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0019] like Figure 1-Figure 7 As shown, the method for constructing a implicit evolutionary damage constitutive model for coal and rock based on ensemble learning includes the following steps: S1. Conduct indoor simulation experiments and related tests to obtain basic coal and rock parameter data. Combined with the coal and rock damage evolution mechanism and failure characteristics, construct a multidimensional coal and rock damage characteristic parameter system and sample dataset. X , Y ).

[0020] S11, the system collects and analyzes geological and hydrogeological data within the mining area, collecting coal samples from the main No. 16 coal seam, water samples from major aquifers (Ordovician limestone water, No. 14 limestone water, Jurassic sandstone water), inrush water, and water samples from the goaf. XRD diffraction analysis is used to determine the coal and petrological mineral composition, and the samples are processed into standard specimens of Ф50×100mm. A complete water quality analysis is performed on the collected mine water samples, and salt solutions are prepared according to the concentration ratio of major ions, with Ca being the main cation. 2+ Mg 2+ : The main anion is .

[0021] S12 was used to conduct coal and rock immersion tests with different concentrations (0.1, 0.3, 0.5, 0.7, 1.0 mol / L), pH values ​​(1, 3, 7, 10, 13), flow rates (0, 0.5, 1.0, 1.5, 2.0 m / s), temperatures (10, 20, 25, 30, 40℃), and water pressures (1.0, 2.0, 3.0, 4.0, 5.0 MPa) and mine water. Uniaxial compression, nuclear magnetic resonance testing, and acoustic wave detection were used to test the mechanical parameters and microstructure parameters of coal and rock under different aqueous solutions and immersion times.

[0022] S13, based on the test results from step S12 and combined with PFC discrete element numerical simulation, analyze the damage and deterioration mechanism and instability failure law of coal and rock under different aqueous solution immersion environments. Construct a stratified evaluation index framework from aspects such as the inherent properties of coal and rock, environmental conditions, and external loads, and establish a multi-scale coal and rock damage index evaluation system, such as... Figure 2 As shown.

[0023] S14, Selecting indicators to characterize the properties of coal and rock: ,in, r For the density of the rock mass, v p For the speed of sound waves, f Porosity D w Damage caused by water factors, D l For load damage, D T Total damage, D It is the fractal dimension.

[0024] S15, Select indicators to characterize groundwater environmental factors: ,in, T Soaking time, Na is Na + Concentration, Ca is Ca 2+ Concentration, Mg is Mg 2+ Concentration, pH refers to pH value. v The above indicators, representing water flow velocity, are primarily obtained from detection data.

[0025] S16, Select an index that can characterize the stress-strain process of coal and rock: ,in, e i-2 For the first two strains, s i-2 For the first two stresses, e i-1 As a contingency plan, s i-1 For the previous stress, e i In response to the current situation, s i The above indicators are obtained from the coal and rock stress-strain curves, representing the current stress.

[0026] S14 includes: S141, above, r , v p and f Water factor damage parameters obtained directly from experimental data D w It is expressed as follows: (1) In the formula, D w Damage caused by water factors, S T0 This represents the initial NMR peak area. S T The area of ​​the nuclear magnetic resonance peaks after water-rock interaction. a and b The conversion coefficient between water content and NMR peak area. V 总 The volume of the rock.

[0027] S142, above, load damage parameters D l It is expressed as follows: (2) In the formula, D l For load damage, e and e p These are strain and peak strain, respectively. m For shape factor, exp It is an exponential function.

[0028] S143, the above, total coal and rock damage parameters D T It is expressed as follows: (3) S144, above, fractal dimension D The fractal geometry formula is as follows: (4) In the formula, D For fractal dimension, T 2 is the NMR peak. T 2max The largest NMR peak, W v For the lateral relaxation time to be less than T 2% of the cumulative pore volume.lg It is a logarithmic function.

[0029] S2, the sample dataset ( X , Y The dataset underwent normalization and standardization preprocessing, resulting in a dataset of 405 samples. The sample dataset was then divided into a training set (7:2:1) and a training set (…). X tr , Y tr ), test set ( X te , Y te ) and prediction set ( X tp , Y tp The training set was randomly divided into 5 subsets using 5-fold cross-validation. By training and comparing various neural network algorithms, the base learner and meta-learner were determined, and the key parameters of the learners were globally optimized to construct an ensemble learning framework based on neural networks.

[0030] S21, determine the input and output samples, the... i The input parameters for each sample are represented as follows: The output sample is represented as: ,in, ; y i For the first i The current stress corresponding to each sample.

[0031] S22, uses five-fold cross-validation to train various neural network algorithms, that is, the training set ( X tr , Y tr The data is randomly divided into 5 parts. Each neural network algorithm uses 4 parts for training and 1 part for validation to obtain the prediction results. P tr-j This process is repeated five times to obtain the prediction results of each neural network algorithm on the training set. In this study, four base learners with good training performance—BP, RF, SVM, and LSTM—were selected and subjected to the same operations to generate a new training set. P tr , Y tr ), as training samples for the meta-learner, where , .

[0032] S23, use the trained base learner to make predictions on the original test set to obtain the predicted values. P teA new test set is constructed by combining experimental data. P te , Y te ), as test samples for meta-learners By using Taylor charts to compare the performance of base learners, the base learners and meta-learners are determined, thereby establishing an ensemble learning framework.

[0033] S3, by inputting 7 coal and rock intrinsic property indicators, 6 groundwater environmental factor indicators, and 5 stress-strain process indicators, the data acquisition method is as follows: Figure 3 As shown, the calculation process is as follows: Figure 4 The ensemble learning framework was iteratively trained and, after multiple adjustments, a latent evolutionary damage model for coal and rock based on neural network ensemble learning was constructed, such as... Figure 5 As shown.

[0034] S31, as mentioned above, the stress-strain process index uses strain-controlled stress, with the strain increasing by 0.1% in each loading step. When predicting the current stress in the first loading step, the strain and stress of the first two steps ( e i-2 , s i-2 , e i-1 , s i-1 All values ​​are set to 0, and the current strain is ( e i Take 0.1% and output the current stress value. s i ), represented as: .

[0035] S32, when predicting the current stress based on the second loading step e i-1 and s i-1 Take 0.1% and the previous stress prediction value respectively. Y 1, e i-2 and s i-2 All values ​​are set to 0, current strain ( e i Take 0.2%, and output the current stress value of the second loading step. s i ), represented as: .

[0036] S33, when predicting the current stress based on the third loading step. e i-2 and s i-2 Take 0.1% and the stress prediction values ​​from the previous two steps respectively.Y 1, e i-1 and s i-1 We take 0.2% and the stress value predicted in the previous step, respectively. Y 2. Current Situation ( e i Take 0.3%, and output the current stress value of the third loading step. s i ), represented as: .

[0037] S34, when predicting the circumferential strain-stress of the specimen, the strain values ​​are successively taken as 0, -0.1%, -0.2%, and so on, until the specimen fails under load, as shown in the following table: (5) S4. Input the test set data into the implicit evolutionary damage model of coal and rock for prediction. Compare and analyze the prediction results with the results of indoor experiments and numerical simulations to verify the accuracy of the model.

[0038] S41, the prediction set is used to test the generalization applicability of the model. The hidden layers of the meta-learner CNN are set to use the ReLU function and the max pooling function, the loss function is the mean squared error, and the learning rate is set to 0.005. When the model's error convergence curve is less than 0.1, the iteration operation ends.

[0039] S42, plot the stress-strain curves obtained from model prediction, laboratory experiments, and numerical simulation, respectively, and evaluate the accuracy of the prediction model based on the curve overlap. Figure 6 and Figure 7 As shown.

[0040] S5. Call the above model and input the non-destructive testing parameters and environmental index parameters of coal and rock under different aqueous solutions to obtain the full stress-strain curves of coal and rock damage under different working conditions, and evaluate the mechanical properties of coal and rock.

[0041] like Figure 8 As shown, the coal and rock implicit evolutionary damage constitutive model construction system based on ensemble learning is used to implement the above-mentioned coal and rock implicit evolutionary damage constitutive model construction method based on ensemble learning, and includes the following modules: Data acquisition module: Through indoor simulation experiments and related tests, acquire basic coal and rock parameter data, and combine coal and rock damage evolution mechanism and failure characteristics to establish a multidimensional coal and rock damage characteristic index system and sample dataset; Data processing module: preprocesses the sample dataset, divides it into training set, test set and prediction set, trains and compares various neural network algorithms to determine the base learner and meta learner, and globally optimizes the key parameters of the learner to build an integrated learning framework. Model training module: The integrated learning framework is iteratively trained and debugged by inputting a multidimensional coal and rock damage feature index system to construct a coal and rock implicit evolution damage model based on an integrated neural network; Performance evaluation module: The accuracy of the implicit evolutionary damage model of coal and rock is verified by using a test set. The test results are compared and analyzed with the results of indoor experiments and numerical simulations to obtain the prediction accuracy of the implicit evolutionary damage model of coal and rock. Damage prediction module: Input the non-destructive testing parameters and environmental index data of coal and rock under different working conditions into the implicit evolutionary damage model of coal and rock to obtain the full stress-strain curve of coal and rock damage and evaluate the mechanical properties of coal and rock.

[0042] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0043] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for constructing a implicit evolutionary damage constitutive model for coal and rock based on ensemble learning, characterized in that, Includes the following steps: S1. Through indoor simulation experiments and related tests, basic parameters of coal and rock are obtained. Combined with the coal and rock damage evolution mechanism and failure characteristics, a multidimensional coal and rock damage characteristic index system and sample dataset are established. The multidimensional coal and rock damage characteristic index system includes coal and rock intrinsic property indexes, groundwater environmental factor indexes, and stress-strain process indexes. The intrinsic properties of the coal and rock are expressed as follows: ； in, ρ For the density of the rock mass, v p For the speed of sound waves, Porosity D w Damage caused by water factors, For load damage, D T Total damage, D It is the fractal dimension; The groundwater environmental factor indicators are expressed as follows: ； in, T Soaking time, Na is Na + Concentration, Ca is Ca 2+ Concentration, Mg is Mg 2+ Concentration, pH refers to pH value. v The velocity of the water flow; The stress-strain process index is expressed as follows: ； in, ε i-2 For the first two strains, σ i-2 For the first two stresses, ε i-1 As a contingency plan, σ i-1 For the previous stress, ε i To prepare for the current situation; The density of the rock mass ρ、 sound wave speed v p and porosity Obtained through indoor simulation experiments; The water factor damage D w Represented as: ； in, S T0 This represents the initial NMR peak area. S T The area of ​​the nuclear magnetic resonance peaks after water-rock interaction. a and b These are the conversion coefficients between water content and NMR peak area, respectively. V 总 For the volume of the rock; The load damage Represented as: ； in, ε and ε p These are strain and peak strain, respectively. m For shape factor, exp It is an exponential function; The total damage D T Represented as: ； The fractal dimension D Represented as: ； in, T 2 is the NMR peak. T 2max The largest NMR peak, W v For the lateral relaxation time to be less than T 2% of the cumulative pore volume. lg It is a logarithmic function; S2, the sample dataset is preprocessed and divided into training set, test set and prediction set. By training and comparing various neural network algorithms, the base learner and meta learner are determined, and the key parameters of the learner are globally optimized to build an integrated learning framework. S3, by inputting a multidimensional coal and rock damage characteristic index system, the integrated learning framework is iteratively trained and debugged to construct a coal and rock implicit evolutionary damage constitutive model based on an integrated neural network; S4. The accuracy of the implicit evolutionary damage constitutive model of coal and rock was verified by using the test set. The test results were compared and analyzed with the results of indoor experiments and numerical simulations to obtain the prediction accuracy of the implicit evolutionary damage constitutive model of coal and rock. S5 inputs the non-destructive testing parameters and environmental index data of coal and rock under different working conditions into the implicit evolutionary damage constitutive model of coal and rock to obtain the full stress-strain curve of coal and rock damage and evaluate the mechanical properties of coal and rock.

2. The method for constructing a implicit evolutionary damage constitutive model of coal and rock based on ensemble learning according to claim 1, characterized in that, S1 includes: S11. Collect representative coal and rock samples from coal seam 16 in the mining area and their corresponding groundwater samples. The groundwater samples include Ordovician limestone water, Lithium 14 limestone water, and Jurassic sandstone water. Supplement the collection of water samples from the water inrush area and the goaf area. Determine the mineral composition of the coal and rock by XRD diffraction analysis. Process the coal and rock samples into standard specimens of Ф50×100 mm and test their physical properties and water quality parameters. S12, coal and rock immersion tests were conducted using solutions of different concentrations, pH values, flow rates, temperatures, and water pressures, as well as mine water. Uniaxial compression, nuclear magnetic resonance testing, and acoustic detection were used to collect the mechanical and microstructural parameters of coal and rock under different aqueous solutions and immersion times. S13. Based on the collected coal and rock mechanical parameters and microstructure parameters, and combined with PFC discrete element numerical simulation, a multidimensional coal and rock damage characteristic index system is constructed.

3. The method for constructing a implicit evolutionary damage constitutive model of coal and rock based on ensemble learning according to claim 2, characterized in that, S2 includes: S21, Set the input and output samples according to the sample dataset; The input sample is represented as follows: ； The output sample is represented as: ； in, For the first i One input sample, y i For the first i The current stress corresponding to each sample; S22, using the training set Five-fold cross-validation was used to train various neural network algorithms to obtain prediction results. And by using BP, RF, SVM and LSTM as base learners, new training sets are generated. As training samples for the meta-learner, among which... ; S23, use the trained base learner to compute the predicted values ​​on the test set. P te A new test set is constructed by combining the sample dataset. As test samples for meta-learners By using Taylor charts to compare the performance of base learners, the base learners and meta-learners are determined, and an ensemble learning framework is established.

4. The method for constructing a implicit evolutionary damage constitutive model of coal and rock based on ensemble learning according to claim 3, characterized in that, S3 includes: S31 employs a strain-controlled stress method, setting the strain increment for each loading step to 0.1%. In the first loading step, the stress and strain of the first two steps are... All are 0, current strain The current stress value is 0.1%, and it is expressed as follows: ； in, σ i The current stress; S32, when predicting the current stress value in the second loading step ε i-1 and σ i-1 The stress values ​​predicted in the first loading step and 0.1% were taken respectively. Y 1. Current Response The value is 0.2%, and the current stress value of the second loading step is output as follows: ； S33, when predicting the current stress value in the third loading step. ε i-2 and σ i-2 The stress values ​​predicted in the first loading step and 0.1% were taken respectively. Y 1, ε i-1 and σ i-1 The stress values ​​predicted in the second loading step were taken as 0.2% and 2% respectively. Y 2. Current Response The value is 0.3%, and the current stress value of the third loading step is output as follows: ； S34, through continuous iteration, the strain value is gradually increased until the specimen is loaded to failure, thus completing the stress-strain path prediction and constructing a constitutive model of implicit evolutionary damage in coal and rock.

5. The method for constructing a implicit evolutionary damage constitutive model of coal and rock based on ensemble learning according to claim 4, characterized in that, The stress-strain path prediction is expressed as follows: 。 6. The method for constructing a implicit evolutionary damage constitutive model of coal and rock based on ensemble learning according to claim 5, characterized in that, S4 includes: S41, set the meta-learner to a CNN model, in which convolution, ReLU activation function and max pooling function are used in sequence in the hidden layer, the loss function is mean squared error, and the learning rate is set to 0.

005. When the error convergence curve of the coal and rock implicit evolutionary damage constitutive model is less than 0.1, the iterative training is terminated. S42. Stress-strain curves predicted by the implicit evolutionary damage constitutive model of coal and rock, obtained from laboratory tests and numerical simulations, are plotted respectively. The accuracy of the implicit evolutionary damage constitutive model of coal and rock is evaluated based on the overlap of the stress-strain curves.

7. A system for constructing a coal and rock implicit evolutionary damage constitutive model based on ensemble learning, used to implement the method for constructing a coal and rock implicit evolutionary damage constitutive model based on ensemble learning as described in any one of claims 1-6, characterized in that, Includes the following modules: Data acquisition module: Through indoor simulation experiments and related tests, acquire basic coal and rock parameter data, and combine coal and rock damage evolution mechanism and failure characteristics to establish a multidimensional coal and rock damage characteristic index system and sample dataset; Data processing module: preprocesses the sample dataset, divides it into training set, test set and prediction set, trains and compares various neural network algorithms to determine the base learner and meta learner, and globally optimizes the key parameters of the learner to build an integrated learning framework. Model training module: The integrated learning framework is iteratively trained and debugged by inputting a multidimensional coal and rock damage feature index system to construct a coal and rock implicit evolutionary damage constitutive model based on an integrated neural network; Performance evaluation module: The accuracy of the implicit evolutionary damage constitutive model of coal and rock is verified by using a test set. The test results are compared and analyzed with the results of indoor experiments and numerical simulations to obtain the prediction accuracy of the implicit evolutionary damage constitutive model of coal and rock. Damage prediction module: Input the non-destructive testing parameters and environmental index data of coal and rock under different working conditions into the implicit evolutionary damage constitutive model of coal and rock to obtain the full stress-strain curve of coal and rock damage and evaluate the mechanical properties of coal and rock.

Images