Mine water inrush point and simulation model parameter identification method based on artificial intelligence

Abstract

The present application relates to a kind of artificial intelligence-based mine water inrush point and simulation model parameter identification method, belong to coal mine water inrush disaster prevention technical field.Step is: step 1: according to observation data, establish numerical model, determine the prior information of the to-be-identified parameter including water inrush point position coordinate;Step 2: based on numerical model and parameter prior information, generate training sample data set and test sample data set of surrogate model;Step 3: construct and train surrogate model neural network;Step 4: verify surrogate model accuracy;Step 5: execute simulated annealing algorithm, identify water inrush point position and simulation model parameter.The method provided by the present application can solve the problem of accurate positioning of coal mine water inrush point position, which is of great significance for timely scientific disaster management and rescue work.

Description

Technical Field

[0001] This invention belongs to the field of coal mine water inrush disaster prevention and control technology, and particularly relates to an artificial intelligence-based method for identifying the location and model parameters of mine water inrush. Background Technology

[0002] With the increasing depth of coal mining, hydrogeological conditions are becoming increasingly complex, leading to frequent coal mine water inrush disasters that cause serious casualties and property losses. Groundwater in aquifers is the primary source of mine water inrushes. Coal mining activities induce rock strata fracturing, forming water-conducting channels that allow large amounts of water from the aquifer to rush into the mine shafts in a short period, resulting in severe consequences such as mine flooding, trapped personnel, and damage to production equipment. To effectively carry out rescue operations for coal mine water inrush disasters and prevent secondary accidents, it is necessary to quickly determine the specific location of the water inrush point in the mine, and also to clarify the key model parameters involved in water inrush simulation and prediction. This will enable accurate analysis of the underground flooding range and is of great guiding significance for developing rescue and prevention plans.

[0003] Currently, the main method of water inrush detection relies on underground workers reporting the location of the inrush point to the dispatch center via underground telephone after the inrush. However, the reported location is often a regional location without precise coordinates. If the inrush damages underground communication equipment or traps personnel unable to contact surface personnel, the location of the inrush point cannot be determined. Existing research proposes methods for identifying the source of inrush water by collecting water samples after the inrush and analyzing the hydrochemical data. Based on the water quality characteristics of different sources, algorithms such as cluster analysis or support vector machines are used to determine which aquifer the inrush originated from. However, this method cannot pinpoint the location coordinates of the inrush point. Furthermore, the values ​​of the parameters in the numerical model involved in inrush simulation and prediction directly determine the reliability of the simulation and prediction results. However, in most cases, the parameters of many groundwater numerical models cannot be directly obtained using existing measurement methods.

[0004] Inversion simulation is a technique that, based on a forward numerical model, uses available observational data to inversely identify model parameters that are difficult to obtain directly. Currently, the main approach to inversion simulation is to transform the research problem into an optimization problem and solve it through optimization methods. In this technical field, based on nonlinear optimization theory, the research problem can be transformed into a parameter optimization problem. Then, an artificial intelligence method combining deep learning and simulated annealing algorithms can be used to efficiently and accurately solve the optimization problem, thereby simultaneously identifying the location of mine water inrush points and the parameters of the simulation model. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based method for identifying mine water inrush points and simulation model parameters. This method can determine the spatial coordinates of underground water inrush points, solving the problem of rapid and synchronous identification of underground water inrush point locations and simulation model parameters.

[0006] To achieve the above objectives, the technical solution adopted in this invention is: a method for identifying mine water inrush points and simulation model parameters based on artificial intelligence. The overall approach involves establishing a groundwater numerical model of the aquifer, and then, based on previously known groundwater level observation data from wells monitoring the aquifer, using the location of the water inrush point along with other unknown parameters such as permeability and inflow rate as decision variables to establish a nonlinear optimization model. The location of the water inrush point, permeability, and inflow rate are then simultaneously identified by solving the nonlinear optimization model based on simulated annealing. To ensure computational efficiency during this process, a deep learning-based alternative modeling method is also employed. The specific implementation of this technical solution includes the following steps:

[0007] Step 1: Establish a numerical model of groundwater flow in the aquifer of the mine where water inrush occurs. In the model, set the output points for the water level simulation results based on the location and observation time of the actual hydrological observation wells. Input the x-coordinate and y-coordinate of the water inrush point location, the water inrush volume Q, and other unknown model parameters (p1,…,p…). n (where n represents the number of other unknown parameters in the model) are used together as decision variables, expressed as: m=[X,Y,Q,p1,…,p n Simultaneously, based on the preliminary data collection, the value range of each decision variable in m is determined, with its upper and lower limits expressed as follows: m U =[X U ,Y U Q U ,p 1U ,…,p nU ] and m L =[X L ,Y L Q L ,p 1L ,…,p nL ].

[0008] Step 2: Based on the upper limit m of the decision variable determined in Step 1 U and lower limit m L Two sets of parameter datasets were randomly sampled using the Latin hypercube sampling method and used as input parameters for the training sample dataset. and the input parameters of the test sample dataset Using the groundwater flow numerical model from step 1, calculate M one by one. Train and M TestThe simulation results of water level at the observation time at the location of each hydrological observation well corresponding to each model parameter sample are stored, and all observation data are stored in vector data format y. i Finally, the model response results from the training dataset and the test sample dataset can be obtained, denoted as follows: and In the following explanation, the training sample dataset will be represented as D. Train ={M Train ,Y Train} and D Test ={M Test ,Y Test}

[0009] Step 3: Construct a deep convolutional neural network model (DNN for short), with the input layer and output layer being the model parameter vector m, respectively. i and model response vector y i The model is represented as A deep convolutional neural network is constructed based on L1 norm constraints to implement the loss function for alternative model predictions. Then, θ is updated with the goal of minimizing the loss function. DNN This completes the training of the DNN. At this point, the trained F... DNN (m i ,θ DNN () as an alternative model to the groundwater flow numerical model in step 1.

[0010] Step 4: Transfer the test sample dataset D obtained in Step 2 Test Input parameter M Test Substitute each item into the substitution model F DNN (m i ,θ DNN In the process, the corresponding prediction results are obtained. According to F DNN (m i ,θ DNN The convergence loss function value L during training and the value based on Y Test and The calculated coefficient of determination R 2 The value is used to determine whether the prediction accuracy of the alternative model meets the requirements. If it does, proceed to step 5; otherwise, return to step 2 to increase the number of samples in the training dataset.

[0011] Step 5: Apply the alternative model F obtained in Step 4 DNN (m i ,θ DNNAs equality constraints, the upper and lower limits of the decision variable m in step 1 are used as inequality constraints. Combined with the least squares constraint, a nonlinear optimization inversion model is constructed, which serves as the overall constraint condition for each decision variable in step 1. Then, the simulated annealing algorithm is used to optimize the decision variable m, finding the optimal solution of decision variable m that satisfies the constraint conditions of the nonlinear optimization inversion model constructed in this step. Thus, the location coordinates X and Y of the water inrush point, as well as other key simulation prediction parameters Q and p, are finally obtained. i .

[0012] The numerical model of groundwater flow in the mine water inrush aquifer in step 1 is established using the groundwater numerical simulation program TOUGHREACT. The water inrush volume Q is generalized to a constant value, and the n other unknown parameters represent the permeability values ​​of n permeability parameter zones within the simulation area.

[0013] In step 2, the upper limit m of the decision variable determined in step 1 is used. U and lower limit m L The Latin hypercube sampling method follows the principle of uniform distribution.

[0014] n in step 2 Train and n Test The value of n follows Train >n Test And n Test Set according to the principle of ≥50.

[0015] The DNN model in step 3 is an improvement on the ResNet-18 deep residual two-dimensional convolutional neural network. First, the vector data of the input decision variables is mapped to a 6400-dimensional vector using a fully connected neural network. Then, this vector is reshaped into an 80×80 rectangular data structure, which is used as the input to the ResNet-18. The output layer of the DNN is set to a vector with the same dimension as the observed data y.

[0016] The expression for the loss function used in step 3 to construct a deep convolutional neural network based on L1 norm constraints to achieve alternative model prediction is as follows:

[0017]

[0018] In the formula: θ DNN The m represents the weight parameters of a deep neural network. i and y i Let represent the model parameters and model output of the i-th sample in the training dataset, respectively; N represents the total number of samples in the training dataset; w d This represents a regularization term used during neural network training to prevent overfitting.

[0019] In step 3, the objective is to minimize the loss function (1) on θ. DNN The update process is implemented using the deep learning framework PyTorch.

[0020] The certainty coefficient R in step 4 2 The calculation formula is as follows:

[0021]

[0022] In the formula, Represents all y Train(i) The mean.

[0023] In step 4, the smaller the value of L, the more R... 2 The closer the value is to 1, the better the alternative model F is. DNN (m i ,θ DNN The higher the prediction accuracy, the better. In this technique, thresholds (L0 and L2) are pre-set. Then, by determining L≤L0, and Does it meet the accuracy requirements?

[0024] The basic form of the inversion nonlinear optimization model in step 5 is as follows:

[0025]

[0026] In the formula: F represents the objective function based on least squares constraints; y obs Represents the observed data vector; y obs [i] represents the i-th variable element in the observation data vector; m L and m U Let N represent the upper bound and lower bound vectors of the model parameter vector m, respectively; obs This indicates the number of observation data.

[0027] The simulated annealing algorithm in step 5 is implemented through the following steps:

[0028] Step 5-1: Set the initial iteration temperature T0 and the initial solution m of the decision variable m for the hyperparameters of the simulated degradation algorithm. i ;

[0029] Step 5-2: In m i A new solution m is randomly generated in the vicinity of [the solution]. j The specific method is to use m i Multiplied by a coefficient e(m) of a random perturbation j =e×m i Where e is based on the Gaussian distribution N~(1,σ). 2A randomly generated number with the same dimension as m. σ has a default value of 0.01, which can be adjusted in different application scenarios, ranging from 0 to 0.1.

[0030] Step 5-3: Place m i and m j Substituting these values ​​into formula (3), we can calculate m. i and m j The corresponding inversion optimization objective function value: F i and F j .

[0031] Step 5-4: Determine if F i ≥F j If true, then the current solution m will be... i Updated to m j Otherwise, calculate m according to the following formula. i Updated to m j The probability of:

[0032]

[0033] In the formula: a represents the decay coefficient in the simulated annealing algorithm, which is 0.99 here; t represents the current time and the current number of iterations.

[0034] In a specific computer program, the probability selection in formula (4) is determined by generating a random number rand(x) between 0 and 1. When rand(x) ≤ P(m) i →m j When ), then m i Updated to m j Otherwise, it will not be updated.

[0035] Step 5-5: Under the current temperature conditions, repeat steps 5-2 to 5-4 until the preset number of simulated annealing inner loop iterations is reached. Then update the temperature and time respectively: t = t+1 and T t =a t T0, then proceed to the next step.

[0036] Step 5-6: Return to step 5-2, and based on T obtained in step 5-5... t Update t until the preset number of outer loop iterations has been completed.

[0037] Beneficial Effects: Based on groundwater inversion theory, this invention proposes an artificial intelligence-based method for identifying mine water inrush points and simulation model parameters. This technique integrates a deep learning-based alternative model method and a parameter optimization strategy based on simulated annealing algorithms. It can utilize directly obtainable on-site water level change monitoring data to simultaneously identify the specific locations of water inrush points that are difficult to obtain directly, as well as key parameters of the simulation model. This provides technical support for mine water inrush disaster rescue and accurate simulation of the disaster's impact range. Attached Figure Description

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0039] Figure 1 This is a flowchart of the method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to the present invention.

[0040] Figure 2 This is a schematic diagram of the simulation region model according to an embodiment of the present invention;

[0041] Figure 3 This is a comparison chart of the simulated water level difference results and the actual observation results of the correction model of this invention;

[0042] Figure 4 This is a schematic diagram of the DNN model structure generated based on the ResNet-18 model results of the present invention. Detailed Implementation

[0043] The method of the present invention will be described in detail below with reference to specific examples.

[0044] like Figure 1 As shown, the present invention provides a method for identifying mine water inrush points and simulation model parameters based on artificial intelligence, comprising the following steps:

[0045] Step 1: Based on the basic hydrogeological data of the study area, establish a numerical model of groundwater flow in the aquifer of the mine water inrush zone. The numerical model of groundwater flow in the aquifer of the mine water inrush zone includes hydrological observation wells for monitoring the changes in water level drawdown within the study area. These hydrological observation wells are defined as water level observation points. Based on the mine mining engineering plan, preliminarily determine the coordinate range of the water inrush point, the water inrush volume, and the a priori interval range of other unknown parameters in the numerical model of the coal seam water inrush zone.

[0046] The x-coordinate and y-coordinate of the unknown water inrush point location, the water inrush volume Q, and n unknown parameters are used together as decision variables, expressed as: Total decision variables m = [X, Y, Q, p1, ..., p n ],p1,…,p nLet m represent the n unknown parameters in the groundwater model, including permeability parameters for different zones and boundary condition parameters; simultaneously, based on collected preliminary data, determine the value range of each decision variable in m, with the upper and lower limits of the decision variables expressed as: m U =[X U ,Y U Q U ,p 1U ,…,p nU ] and m L =[X L ,Y L Q L ,p 1L ,…,p nL ];

[0047] Step 2: Based on the determined upper limit m of the decision variable... U and lower limit m L Using the Latin hypercube sampling method, two sets of parameter datasets were randomly sampled from the groundwater flow numerical model. These two sets of parameter datasets were then used as input parameters for the training sample dataset. and the input parameters of the test sample dataset n Train M represents the training sample dataset. Train The sample, n Test M represents the test sample dataset. Test ;

[0048] Calculate M one by one using a groundwater flow numerical model Train and M Test The simulation results of water level at the observation time at the location of each hydrological observation well corresponding to each model parameter sample are stored, and all observation data are stored in vector data format y. i Finally, the model response results of the training dataset can be obtained. Model response results and test sample dataset Let the training sample dataset be represented as D. Train ={M Train ,Y Train} and D Test ={M Test ,Y Test};

[0049] Upper limit m of decision variable U and lower limit m L Sampling was performed using the Latin hypercube sampling method, following the principle of uniform distribution; n Train The number of samples n Train >n Test The number of samples, and satisfying n Test ≥50.

[0050] Step 3: Construct a deep convolutional neural network model (DNN). The input and output layers of the DNN are the parameter vectors m of the groundwater numerical model, respectively. i and model response vector y i The DNN model is represented as θ DNN The weight parameters of the deep neural network are represented; a loss function is constructed based on L1 norm constraints to implement an alternative model prediction for the groundwater numerical model; then, with the goal of minimizing the loss function, θ is updated through the error backpropagation algorithm. DNN To complete the training of the DNN; at this point, the trained DNN model F will be... DNN (m i ,θ DNN (This can be used as an alternative model to the groundwater flow numerical model in step 1.)

[0051] The loss function for constructing a deep convolutional neural network to implement alternative model predictions under the L1 norm constraint is calculated as follows:

[0052]

[0053] In the formula: θ DNN The m represents the weight parameters of a deep neural network. i and y i Let represent the model parameters and model output of the i-th sample in the training dataset, respectively; N represents the total number of samples in the training dataset; w d This represents a regularization term used during neural network training to prevent overfitting.

[0054] like Figure 4 The diagram shows the structure of a DNN model, which is an alternative model built based on the results of the ResNet-18 model. The DNN model is an improvement on the deep residual two-dimensional convolutional neural network of ResNet-18. First, the vector data of the input decision variables is mapped to a 6400-dimensional vector using a fully connected neural network. Then, it is reshaped into an 80×80 rectangular data structure, which is used as the input layer of ResNet-18. The output layer is a vector with the same dimension as the observed data y. The rest are the original structure of ResNet-18.

[0055] Step 4: Transfer the test sample dataset D obtained in Step 2 Test Input parameter M Test Substitute each item into the substitution model F DNN (m i ,θ DNN In the process, the corresponding prediction results are obtained. According to FDNN (m i ,θ DNN The convergence loss function value L during training and the value based on Y Test and The calculated coefficient of determination R 2 The value determines whether the prediction accuracy of the alternative model meets the requirements. If it does, proceed to step 5; otherwise, return to step 2 to increase the number of samples in the training dataset. The smaller the convergence loss function value L, the better the coefficient of determination R. 2 The closer the value is to 1, the better the alternative model F is. DNN (m i ,θ DNN The higher the prediction accuracy, the better. In this technique, the threshold L0 of the loss function and the threshold of the deterministic coefficient are pre-set. Then, by determining L≤L0, and Does it meet the accuracy requirements?

[0056] Coefficient of determination R 2 The calculation formula is as follows:

[0057]

[0058] In the formula, Represents all y Train(i) The mean.

[0059] Step 5: Replace the substitute model F that meets the accuracy requirements from Step 4. DNN (m i ,θ DNN As an equality constraint, the upper limit of the value of the decision variable m in step 1 is set to m. U and lower limit m L As an inequality constraint, combined with the least squares constraint, a nonlinear optimization inversion model is constructed, which serves as the population of decision variables m = [X,Y,Q,p1,…,p] in step 1. n The constraints of the nonlinear optimization inversion model constructed in this step are then determined, and the simulated annealing algorithm is used to optimize the overall decision variable m. The optimal solution for the overall decision variable m under these constraints is found, ultimately yielding the location coordinates X and Y of the water inrush point, as well as other key simulation and prediction parameters Q and p1,…,p. n .

[0060] The basic form of the nonlinear optimization inversion model is as follows:

[0061]

[0062] In the formula: F represents the objective function based on least squares constraints; y obs Represents the observed data vector; y obs[i] represents the i-th variable element in the observation data vector; m L and m U Let N represent the upper bound and lower bound vectors of the model parameter vector m, respectively; obs This indicates the number of observation data.

[0063] The simulated annealing algorithm is implemented through the following steps:

[0064] Step 5-1: Set the initial iteration temperature T0 and the initial solution m of the decision variable m for the simulated annealing algorithm. i ;

[0065] Step 5-2: In m i A new solution m is randomly generated in the vicinity of [the solution]. j The specific method is to use m i Multiplied by a coefficient e(m) of a random perturbation j =e×m i ), where e is a Gaussian distribution N~(1,σ), 2 ) Randomly generated numbers with the same dimension as m. The default value of σ is 0.01, which can be adjusted in different application scenarios. The adjustment range is 0 to 0.1.

[0066] Step 5-3: Place m i and m j Substituting these values ​​into formula (3) respectively, we obtain m. i and m j The corresponding inversion optimization objective function value: F i and F j ;

[0067] Step 5-4: Determine if F i ≥F j If true, then the current solution m will be... i Updated to m j Otherwise, calculate m according to the following formula. i Updated to m j The probability of:

[0068]

[0069] In the formula: a represents the attenuation coefficient in the simulated annealing algorithm, with a value of 0.99; t represents the current time, and T represents the current iteration number; T0 represents the temperature at the initial iteration time, with a default value of 100; the probability selection in formula (4) is determined by generating a random number rand(x) between 0 and 1, and when rand(x) ≤ P(m i →m j When ), then m i Updated to m j Otherwise, no update will be made.

[0070] Step 5-5: Under the current temperature conditions, repeat steps 5-2 to 5-4 until the preset number of simulated annealing inner loop iterations is reached; then update the temperature and time respectively: t = t+1 and T t =a t T0;

[0071] Step 5-6: Return to step 5-2, and based on T obtained in step 5-5... t Update t until the preset number of outer loop iterations has been completed.

[0072] Example

[0073] A coal mine water inrush scenario is constructed. The specific aquifer has been identified; further determination of the exact water inrush locations is needed. A two-dimensional groundwater flow model is obtained using TOUGHREACT modeling. The model range is 10000m × 10000m, with the east and west boundaries assumed to be constant head boundaries of equal value, and the north and south boundaries assumed to be zero flow boundaries. There are two known water inrush points within the area, with inflow rates of 72 m³ / s at point I1. 3 / h and I2 point 54m 3 / h. Assume a water inrush disaster occurs at a certain working face, but the location of the inrush point is unknown. The inrush begins after 360 days of model operation, with a water inrush volume of 720 m³. 3 / h(I3 point). There are 10 known water level change monitoring wells (#1 to #10) in the study area. In the TOUGHREACT numerical calculation process, the entire area in the model is divided into 80×80 discrete grids. Among them, the central 3000m×3000m range is subdivided into 60×60 grids. Assume that the permeability parameter in the model is divided into three zones. According to the water inrush scenario, there are 6 parameters that need to be identified, namely the x-coordinate of the water inrush point location, the y-coordinate of the water inrush location, the water inrush volume (Q), and the permeability of the three zones (k1, k2, k3). In this case, k1, k2, and k3 correspond to the other model parameters besides X, Y, and Q, corresponding to p1 to p3 in step 1. The specific information of the above model is as follows. Figure 2 As shown.

[0074] To verify the feasibility of this patented technology, water level change observation data were obtained from 10 observation points every two months over a simulated period of two years. An observation noise disturbance following a Gaussian distribution N(1,0.01) was then added to this data to represent the actual water level observation data obtained from this hypothetical case. Based on this observation data, the patented technology was then used to invert and identify six unknown model parameters, including the location of the water inrush point.

[0075] The prior interval values ​​of the six parameters introduced in step 1 are shown in Table 1.

[0076] The number of samples in the training sample dataset and the test sample dataset in step 2 are 300 and 50, respectively.

[0077] Alternative model prediction accuracy metrics in step 4: loss function and R0 2 The values ​​were 0.0066 and 0.9918, respectively. To further improve the accuracy of the substitution model, we returned to step 2 and increased the training sample dataset to 500. The substitution model was then retrained, with the loss function and R... 2 The values ​​were increased to 0.0040 and 0.9968 respectively. At this point, the prediction accuracy met the requirements, and subsequent steps were executed.

[0078] The key parameter settings in step 5 of the simulated annealing algorithm implementation process are as follows:

[0079] In step 5-1, T0 = 100;

[0080] The temperature decay constant a = 0.99 in steps 5-4 and 5-5;

[0081] The inner loop in step 5-5 has 150 iterations;

[0082] The outer loop in steps 5-6 has 300 iterations.

[0083] The inversion identification results of the six identification parameters obtained through this patented technology and their relative errors with the true values ​​are shown in Table 1.

[0084] Table 1. True values, prior intervals, inversion estimates, and relative errors of the parameters to be inverted.

[0085]

[0086]

[0087] As can be seen from the table, the relative errors of the X and Y coordinates of the water inrush point are both within 0.04. The error range of the X coordinate has been reduced from the original 850m (4875~5725m) to about 20m (5625m-5604.44m); the error range of the Y coordinate has been reduced from the original 150m (4875~5025m) to within 10m (4975m-4966.11m).

[0088] In addition, apart from k2, whose relative error is slightly larger (0.295), the relative errors of the other model parameters are all within 0.05. Nevertheless, the inversion value of k2 is 6.599 × 10⁻⁶. -13 m 2Compared with the actual value of 5.097×10 -13 m 2 They are on the same order of magnitude. For example... Figure 3 The comparison chart shown here, which shows the simulation results of the water level difference from the corrected model with the actual observed data, corresponds to... Figure 2 Information on measurement points #1-#10 in the diagram (points represent actual observation data, and lines represent simulated curves after model correction). Figure 3 The fitting between the simulation results and observed values ​​of the corrected model shows that the two-year observation data from the 10 observation points basically fit the corrected simulation curve. Based on the above information, the values ​​of Q, k1, k2, and k3 obtained from the inversion solution are all close to their true values, and they achieve good fitting and correction of the numerical model; therefore, their inversion results are reliable.

[0089] Therefore, the method for identifying mine water inrush points and simulation model parameters based on artificial intelligence provided by this patent can determine the specific location coordinates of the water inrush point, achieve accurate positioning of the underground water inrush point, and simultaneously identify the water inrush volume and permeability parameter values, providing key information for the prevention and control of water inrush disasters.

Claims

1. A method for identifying mine water inrush points and simulation model parameters based on artificial intelligence, characterized in that, The steps are as follows: Step 1: Collect preliminary data of the mine, establish a numerical model of groundwater flow in the aquifer of the mine, and set the water level simulation results in the groundwater flow numerical model according to the actual location of the hydrological observation well and the observation time to output the water inrush point; The x-coordinate and y-coordinate of the water inrush point location, the water inrush volume Q, and n unknown parameters are used together as decision variables, expressed as: total decision variables m = [X, Y, Q, p1, ..., p n ],p1,…,p n Let m represent the n unknown parameters in the groundwater model, including permeability parameters for different zones and boundary condition parameters; simultaneously, based on collected preliminary data, determine the value range of each decision variable in m, with the upper and lower limits of the decision variables expressed as: m U =[X U ,Y U Q U ,p 1U ,…,p nU ] and m L =[X L ,Y L Q L ,p 1L ,…,p nL ]; Step 2: Based on the determined upper limit m of the decision variable... U and lower limit m L Using the Latin hypercube sampling method, two sets of parameter datasets were randomly sampled from the groundwater flow numerical model. These two sets of parameter datasets were then used as input parameters for the training sample dataset. and the input parameters of the test sample dataset n Train M represents the training sample dataset. Train The sample, n Test M represents the test sample dataset. Test The sample; Calculate M one by one using a groundwater flow numerical model Train and M Test The simulation results of water level at the observation time at the location of each hydrological observation well corresponding to each model parameter sample are stored, and all observation data are stored in vector data format y. i Finally, the model response results of the training dataset can be obtained. Model response results and test sample dataset Let the training sample dataset be represented as D. Train ={M Train ,Y Train } and D Test ={M Test ,Y Test }; Step 3: Construct a deep convolutional neural network model (DNN). The input and output layers of the DNN are the parameter vectors m of the groundwater numerical model, respectively. i and model response vector y i The DNN model is represented as θ DNN The weight parameters of the deep neural network are represented; a loss function is constructed based on L1 norm constraints to implement an alternative model prediction for the groundwater numerical model; then, with the goal of minimizing the loss function, θ is updated through the error backpropagation algorithm. DNN To complete the training of the DNN; at this point, the trained DNN model F will be... DNN (m i ,θ DNN (This is used as an alternative model to the groundwater flow numerical model in step 1.) Step 4: Transfer the test sample dataset D obtained in Step 2 Test Input parameter M Test Substitute each item into the substitution model F DNN (m i ,θ DNN In the process, the corresponding prediction results are obtained. According to F DNN (m i ,θ DNN The convergence loss function value L during training and the value based on Y Test and The calculated coefficient of determination R 2 The value determines whether the prediction accuracy of the alternative model meets the requirements. If it does, proceed to step 5; otherwise, return to step 2 to increase the number of samples in the training sample dataset. Step 5: Replace the substitute model F that meets the accuracy requirements from Step 4. DNN (m i ,θ DNN As an equality constraint, the upper limit of the value of the decision variable m in step 1 is set to m. U and lower limit m L As an inequality constraint, combined with the least squares constraint, a nonlinear optimization inversion model is constructed, which serves as the population of decision variables m = [X,Y,Q,p1,…,p] in step 1. n The constraints of the nonlinear optimization inversion model constructed in this step are then determined, and the simulated annealing algorithm is used to optimize the overall decision variable m. The optimal solution for the overall decision variable m under these constraints is found, ultimately yielding the location coordinates X and Y of the water inrush point, as well as other key simulation and prediction parameters Q and p1,…,p. n .

2. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 1, characterized in that, The numerical model of the coal seam water inrush aquifer in step 1 was established using the groundwater numerical simulation software TOUGHREACT.

3. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 1, characterized in that, In step 2, the upper limit m of the decision variable determined in step 1 is used. U and lower limit m L The Latin hypercube sampling method was used for sampling, following the principle of uniform distribution. n Train The number of samples n Train >n Test The number of samples, and satisfying n Test ≥50.

4. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 1, characterized in that, The DNN model in step 3 is an improvement on the deep residual two-dimensional convolutional neural network of ResNet-18; First, the vector data of the input decision variables is mapped to a 6400-dimensional vector using a fully connected neural network. Then, it is reshaped into an 80×80 rectangular data structure, which is used as the input of ResNet-18. The output layer is a vector with the same dimension as the observed data y.

5. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 1, characterized in that, The formula for calculating the loss function of the deep convolutional neural network constructed based on the L1 norm constraints in step 3 to achieve the prediction of the alternative model is as follows: In the formula: θ DNN The weight parameters of a deep neural network; m i and y i Let represent the model parameters and model output of the i-th sample in the training dataset, respectively; N represents the total number of samples in the training dataset; w d This represents a regularization term used during neural network training to prevent overfitting.

6. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 5, characterized in that... In step 3, the objective is to minimize the loss function expressed by formula (1) on θ. DNN The update is performed using the deep learning framework PyTorch.

7. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 1, characterized in that... The certainty coefficient R in step 4 2 The calculation formula is as follows: In the formula, Represents all y Train(i) The mean.

8. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 7, characterized in that... In step 4, the smaller the value of the convergence loss function L, the higher the certainty coefficient R. 2 The closer the value is to 1, the better the alternative model F is. DNN (m i ,θ DNN The higher the prediction accuracy, the better; pre-setting the threshold L0 of the loss function and the threshold of the coefficient of determination. Then, by determining L≤L0, and Does it meet the accuracy requirements? 9. The method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 1, characterized in that... The basic form of the nonlinear optimization inversion model in step 5 is as follows: In the formula: F represents the objective function based on least squares constraints; y obs Represents the observed data vector; y obs [i] represents the i-th variable element in the observation data vector; m L and m U Let N represent the upper bound and lower bound vectors of the model parameter vector m, respectively; obs This indicates the number of observation data.

10. A method for identifying mine water inrush points and simulation model parameters based on artificial intelligence according to claim 9, characterized in that... The simulated annealing algorithm in step 5 is implemented through the following steps: Step 5-1: Set the initial iteration temperature T0 and the initial solution m of the decision variable m for the simulated annealing algorithm. i ; Step 5-2: In m i A new solution m is randomly generated in the vicinity of [the solution]. j The specific method is to use m i Multiplied by a coefficient e(m) of a random perturbation j =e×m i ), where e is a Gaussian distribution N~(1,σ), 2 ) Randomly generated numbers with the same dimension as m. The default value of σ is 0.01, which can be adjusted in different application scenarios. The adjustment range is 0 to 0.

1. Step 5-3: Place m i and m j Substituting these values ​​into formula (3) respectively, we obtain m. i and m j The corresponding inversion optimization objective function value: F i and F j ; Step 5-4: Determine if F i ≥F j If true, then the current solution m will be... i Updated to m j Otherwise, calculate m according to the following formula. i Updated to m j The probability of: In the formula: a represents the attenuation coefficient in the simulated annealing algorithm, with a value of 0.99; t represents the current time, indicating the current iteration number; T0 represents the temperature at the initial iteration time, with a default value of 100; the probability selection in formula (4) is determined by generating a random number rand(x) between 0 and 1, when rand(x) ≤ P(m i →m j When ), then m i Updated to m j Otherwise, no update will be made. Step 5-5: Under the current temperature conditions, repeat steps 5-2 to 5-4 until the preset number of simulated annealing inner loop iterations is reached; then update the temperature and time respectively: t = t+1 and T t =a t T0; Step 5-6: Return to step 5-2, and based on T obtained in step 5-5... t Update t until the preset number of outer loop iterations has been completed.

Images