Multi-factor comprehensive analysis-based construction method for water flowing fractured zone height prediction model

Abstract

The present invention relates to the technical field of water flowing fractured zone height prediction, and disclosed is a multi-factor comprehensive analysis-based construction method for a water flowing fractured zone height prediction model. According to the present invention, on the basis of measured data of the heights of water flowing fractured zones in coal seam roofs of a plurality of mines in the Huaibei and Huainan mining areas, a water flowing fractured zone height prediction index system including factors such as roof type, mining method, coal seam dip angle, mining thickness, working face inclination length, and the presence of faults in a working face is established; taking an SVR model as the core, factor analysis (FA) is used to optimize a raw data structure, an ant lion optimizer (ALO) is used to perform parameter optimization on a penalty factor C and a kernel function parameter g of the SVR model, and a water flowing fractured zone height prediction model based on an FA-ALO-SVR is established; and then, prediction performance of the model is validated by means of new samples, further improving the prediction performance of the water flowing fractured zone height prediction model based on the FA-ALO-SVR is further improved, and achieving good practicality.

Description

Construction method of water-conducting fracture zone height prediction model based on multi-factor comprehensive analysis Technical Field

[0001] The present invention belongs to the technical field of water-conducting fracture zone height prediction, and in particular relates to a method for constructing a water-conducting fracture zone height prediction model based on comprehensive analysis of multiple factors. Background Art

[0002] During coal mining, accurate prediction of the water-conducting fracture zone (WZ) in the coal seam roof is crucial for mine water hazard prevention and production safety. The height of WZs is influenced by multiple factors, including seam characteristics, geological conditions, and mining technology. Its prediction has long been a key focus in mining engineering research. Although various prediction methods have been developed, such as empirical formulas, theoretical analysis, simulation experiments, and numerical simulations, these methods still have limitations, such as the limited nature of empirical formulas and the high cost of field measurements. In recent years, statistical analysis and machine learning methods have made progress in predicting WZ height, as they can account for multiple factors. In particular, support vector machine (SVR) models have been widely used due to their advantages in handling small sample sizes, high dimensions, and nonlinear problems. However, the performance of SVR models is highly dependent on their parameter settings. Commonly used parameter optimization methods, such as grid search and genetic algorithms, suffer from complex and time-consuming operations, which limits the practicality and efficiency of these models. Therefore, finding a more efficient and accurate parameter optimization method is crucial for improving the accuracy and reducing the cost of WZ height prediction.

[0003] Summary of the Invention

[0004] In response to the problems existing in the prior art, the present invention provides a method for constructing a water-conducting fracture zone height prediction model based on comprehensive analysis of multiple factors.

[0005] The present invention is implemented as follows: a method for constructing a water-conducting fracture zone height prediction model based on comprehensive analysis of multiple factors includes:

[0006] Step 1: Based on measured data on the height of water-conducting fractured zones in coal seam roofs from multiple mines in the Lianghuai mining area, a water-conducting fractured zone height prediction index system was established, which included factors such as roof type, mining method, mining depth, coal seam inclination, mining thickness, working face oblique length, and the presence of faults in the working face.

[0007] Step 2: Taking the SVR model as the main body, factor analysis was used to optimize the original data structure. The ant lion optimization algorithm was used to optimize the penalty factor C and kernel function parameter g of the SVR model. A water-conducting fracture zone height prediction model based on FA-ALO-SVR was established.

[0008] Step 3: The prediction performance of the model was tested using new samples, using the mean absolute error (MAE), root mean square error (RMSE), and mean relative error (RMSE). The five indicators, including the Wilmot consistency index (IWA) and the Hill inequality coefficient (TIC), were used to comprehensively evaluate the model from the perspectives of prediction accuracy, prediction ability, and generalization ability, and compared with the traditional SVR model, FA-SVR model, and ALO-SVR model.

[0009] Furthermore, the factor analysis:

[0010] According to the correlation matrix of variable X, the original p variables can be expressed as a linear combination of m (m < p) new variables. Its mathematical model is:

[0011] Expressed in matrix form as X = AF + aε (2)

[0012] Among them, f j is a common factor of pairwise orthogonality; e i is a special factor; a ij is the load of the common factor, A is the load matrix of the common factor;

[0013] The specific steps of factor analysis are as follows:

[0014] Calculate its covariance matrix based on the matrix X, that is, the correlation matrix R, R = (r ij ) p*p ;

[0015] According to the covariance matrix, calculate its characteristic root λ i and its corresponding eigenvector;

[0016] The number of common factors q is determined based on the criterion that the cumulative percentage of variance of the first q eigenvalues ​​is greater than 80%;

[0017] Perform factor rotation and calculate factor loading matrix A;

[0018] Build and solve the factor score model.

[0019] Furthermore, the SVR is:

[0020] The SVR mathematical model is as follows:

[0021] Assume that the sample set is {(x i ,y i ),i=1,2,…,l}; where x i is the input variable of the i-th sample, x i ∈R ω ;yi is the output variable of the i-th sample, y i ∈R; ω represents the dimension of the input variable; the linear function of SVR in high-dimensional feature space can be expressed as:

[0022] Where: w is the normal vector; b is the bias term; is the feature mapping function;

[0023] For the regression problem, an insensitive loss function ε is introduced, and error-free fitting is performed at the levels of positive and negative relaxations ξ and ξ′ respectively. The minimum sum of the complexity of the regression function and the fitting error is taken as the objective function, and the optimization model is

[0024] The constraints are:

[0025] Where: ||w|| 2 is the complexity of the model, which is used to improve the generalization ability of the model; C is the penalty factor, which is used to change the penalty degree for exceeding the fitting error (ε>0); ξ and ξ′ are the introduced slack variables;

[0026] In order to solve the constrained optimization problem, Lagrange multipliers are introduced and the partial derivatives of variables such as w, b, ξ and ξ′ are set to zero, then the dual form of SVR can be obtained:

[0027] The constraints are:

[0028] Where: K(x i ,x j ) is the kernel function, which means x i 、x j After mapping to the high-dimensional feature space, the inner product calculation is performed; α i , α i ′ is the Lagrange multiplier;

[0029] At this time, the SVR regression function can be expressed as

[0030] The RBF kernel function is used, and its expression is: K(x i ,x j )=exp(-g||x i -y i || 2 ) (10)

[0031] Where: is the kernel function parameter, where σ is the width of the Gaussian kernel function;

[0032] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0033] 1. Improved comprehensive analysis capabilities for multi-dimensional factors: By integrating multiple key factors such as roof type, mining method, depth, and inclination, the model can comprehensively capture the factors affecting the height of the water-conducting fracture zone. Compared with empirical formulas based on a single factor, it provides more accurate and comprehensive prediction results.

[0034] 2. Enhanced data optimization and processing capabilities: Factor analysis is used to optimize the original data structure, effectively reducing data redundancy and noise, improving the quality of model input data, and thus enhancing the accuracy and stability of the prediction model.

[0035] 3. Innovative Application of Parameter Optimization Algorithms: We used the Ant Lion Optimization algorithm to optimize key parameters of the Support Vector Regression (SVR) model, improving the model's predictive performance. Compared to traditional parameter selection methods such as grid search, the Ant Lion Optimization algorithm provides a more efficient and accurate parameter optimization solution.

[0036] 4. Comprehensiveness of model evaluation indicators: By using multiple evaluation indicators such as MAE and RMSE to comprehensively evaluate model performance from different perspectives, the objectivity and comprehensiveness of model evaluation are ensured, which helps to accurately judge the practical application value of the model.

[0037] 5. Improved generalization ability of the prediction model: Through testing and verification of new samples, it is proved that the model has good generalization ability, that is, it can maintain high prediction accuracy even on unknown data. This is of great significance for practical applications facing ever-changing mine conditions.

[0038] 6. Dual improvement in safety and economic benefits: Accurately predicting the height of the water-conducting fracture zone helps to reasonably set the thickness of the waterproof coal rock pillar, which not only ensures mine safety and avoids the occurrence of water inrush accidents, but also reduces resource waste and improves the efficiency of coal resource mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] FIG1 is a flow chart of a method for constructing a water-conducting fracture zone height prediction model based on multi-factor comprehensive analysis provided by an embodiment of the present invention.

[0040] FIG2 is a heat map of correlation coefficients among the main control factors provided by an embodiment of the present invention.

[0041] FIG3 is a convergence curve diagram of the FA-ALO-SVR model provided by an embodiment of the present invention.

[0042] FIG4 is a diagram showing the fitting results of the training samples provided by an embodiment of the present invention.

[0043] FIG5 is a diagram showing the training results of three comparison models provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] As shown in FIG1 , the present invention provides a method for constructing a water-conducting fracture zone height prediction model based on comprehensive analysis of multiple factors, which includes the following steps:

[0045] S101, based on measured data on the height of water-conducting fractured zones in coal seam roofs from multiple mines in the Lianghuai mining area, established a water-conducting fractured zone height prediction index system that includes factors such as roof type, mining method, mining depth, coal seam inclination, mining thickness, working face oblique length, and the presence of faults in the working face;

[0046] S102, based on the SVR model, factor analysis was used to optimize the original data structure, and the ant lion optimization algorithm was used to optimize the penalty factor C and kernel function parameter g of the SVR model. A water-conducting fracture zone height prediction model based on FA-ALO-SVR was established.

[0047] In S103, the prediction performance of the model was tested using new samples, using the mean absolute error (MAE), root mean square error (RMSE), and mean relative error (MRSE). The five indicators, including the Wilmot consistency index (IWA) and the Hill inequality coefficient (TIC), were used to comprehensively evaluate the model from the perspectives of prediction accuracy, prediction ability, and generalization ability, and compared with the traditional SVR model, FA-SVR model, and ALO-SVR model.

[0048] The present invention adopts factor analysis, and the SVR provided by the present invention is:

[0049] The mathematical model of the SVR algorithm is as follows:

[0050] Assume that the sample set is {(x i ,y i ),i=1,2,…,l}; where x i is the input variable of the i-th sample, x i ∈R ω ;y i is the output variable of the i-th sample, y i ∈R; ω represents the dimension of the input variable; the linear function of SVR in high-dimensional feature space can be expressed as:

[0051] Where: w is the normal vector; b is the bias term; is the feature mapping function;

[0052] For the regression problem, an insensitive loss function ε is introduced, and error-free fitting is performed at the levels of positive and negative relaxations ξ and ξ′ respectively. The minimum sum of the complexity of the regression function and the fitting error is taken as the objective function, and the optimization model is

[0053] The constraints are:

[0054] Where: ||w|| 2 is the complexity of the model, which is used to improve the generalization ability of the model; C is the penalty factor, which is used to change the penalty degree for exceeding the fitting error (ε>0); ξ and ξ′ are the introduced slack variables;

[0055] In order to solve the constrained optimization problem, Lagrange multipliers are introduced and the partial derivatives of variables such as w, b, ξ and ξ′ are set to zero, then the dual form of SVR can be obtained:

[0056] The constraints are:

[0057] Where: K(x i ,x j ) is the kernel function, which means x i 、x j After mapping to the high-dimensional feature space, the inner product calculation is performed; α i , α i ′ is the Lagrange multiplier;

[0058] At this time, the SVR regression function can be expressed as:

[0059] The present invention adopts RBF kernel function, which is expressed as: K(x i ,x j )=exp(-g||x i -y i || 2 ) (10)

[0060] Where: is the kernel function parameter, where σ is the width of the Gaussian kernel function.

[0061] The present invention provides an ant lion optimization algorithm: To determine the optimal parameters of the SVR model and effectively reduce the parameter adjustment time, the present invention uses the ant lion optimization algorithm to optimize the parameters C and g of the SVR model to further improve the prediction accuracy and generalization ability of the model;

[0062] The mathematical model and specific steps of the Ant Lion optimization algorithm are described as follows:

[0063] (1) Antlion builds a trap

[0064] To simulate the hunting process of ant lions, during the ALO algorithm optimization process, ant lions are randomly selected to build traps based on their fitness values ​​through a roulette wheel operation. This method ensures that ant lions have a greater chance of hunting ants.

[0065] (2) Ant random walk

[0066] The process of ants randomly wandering in nature to find food can be regarded as the process of each search agent searching for a feasible domain; the random walk process can be mathematically expressed as: X(t) = [0, cumsum(2r(t1)-1),…, cumsum(2r(t n )-1)] (11)

[0067] Where: X(t) is the set of steps of the ant's random walk; cumsum is the cumulative sum; t is the number of steps of the random walk, and the maximum number of iterations in this invention is T; r(t) is a random function defined as

[0068] Where: rand is a random number in [0,1];

[0069] The positions and fitness of the ants during the optimization period are saved, which can be expressed as Equations (13) and (14);

[0070] Where: M Ant Represents the position of each ant lion; M OA Represents the fitness of each antlion;

[0071] In order to ensure the randomness of ants’ movement in the search space, the positions of ants need to be standardized according to formula (15):

[0072] Where: a i and b i Respectively represent the minimum and maximum values ​​of the random walk of the i-th variable; and They represent the minimum and maximum values ​​of the random walk of the i-th variable at the t-th iteration respectively;

[0073] (3) Ants enter the trap

[0074] The random walk of the ant is affected by the ant lion trap and can be expressed as:

[0075] Where: represents the position of the jth antlion in the tth iteration;

[0076] (4) Ants fall into the center of the trap

[0077] When the ant lion finds an ant entering the trap, it throws sand out of the trap, causing the ant to slide towards the center of the trap. At this time, the ant's random walk is adaptively reduced, which is also reflected in the algorithm and can be defined as follows:

[0078] Where: I is the ratio; t is the current iteration; T is the maximum number of iterations; w is a constant defined based on t and T;

[0079] (5) Reconstruction trap

[0080] After capturing prey, the ant lion evaluates the fitness of the ant's new position. If an individual with a higher fitness than the ant lion appears in the wandering ant population, the individual is captured. The ant lion's position can be updated as follows:

[0081] Where: t is the current iteration number; represents the position of the i-th ant at the t-th iteration;

[0082] (6) Elite

[0083] The ant's random walk is affected by both the antlion randomly selected by the roulette wheel and the elite antlion; therefore, the ant's position is updated according to the average of the roulette wheel chance and the elite antlion, which can be simulated as follows:

[0084] Where: Indicates the random walk of the ant lion selected by the roulette wheel at the tth iteration; represents the random walk of the elite ant lion at the tth iteration;

[0085] The overall steps of the FA-ALO-SVR-based prediction model for the height of the water-conducting fracture zone of the coal seam roof established by the present invention can be divided into several steps: the first step is to determine the main control factors affecting the height of the water-conducting fracture zone of the coal seam roof and collect sample data as original data; the second step is to divide the training sample set and the test sample set; the third step is to perform factor analysis on the original data of the training sample set, and extract a set of new variables that can reflect most of the information of the original data of the training sample set to reduce redundant information and noise, eliminate the interference of the correlation between various influencing factors on the prediction results, and use them as input variables of the optimization model; the fourth step is to use the ant lion optimization algorithm to optimize the SVR model for training the input variables, determine the penalty factor C and kernel function parameter g of the SVR model, and obtain the optimized model; the fifth step is to verify the prediction effect of the trained optimization model by using the reserved test sample set data as a new sample, and comprehensively evaluate the model from three perspectives of prediction accuracy, prediction ability and generalization ability, and compare it with other models.

[0086] According to the research data information, the prediction index matrix X is established: X=(x ij ) n*p (twenty one)

[0087] Among them, n is the number of measured sample groups of water-conducting fracture zone height; p is the number of main controlling factors affecting the development height of water-conducting fracture zone.

[0088] In order to eliminate the influence of different dimensions of each indicator and the errors caused by the variation of each indicator itself or the large difference in values, it is necessary to standardize the original data of each indicator [7]. The standardization formula is shown in the formula.

[0089] in, is the average value of the jth main control factor; is the standard deviation of the jth main control factor.

[0090] In order to facilitate the standardization of subsequent test samples, the present invention standardized 40 groups of training samples using SPSS26 software and obtained the standardization formula for each main control factor:

[0091] The present invention performs factor analysis using SPSS26 software. First, the KMO (Kaiser-Meyer-Olkin) test statistic and Bartlett's sphericity test are used to determine whether the research data are suitable for factor analysis. Among them, the KMO test statistic is an indicator for testing the simple correlation coefficient and partial correlation coefficient between variables. The value is between 0 and 1. The closer it is to 1, the more common factors there are between the variables, indicating that the correlation between the variables is stronger and the original variables are more suitable for factor analysis. When the KMO value is >0.5, factor analysis can be performed. The Bartlett's sphericity test is obtained based on the determinant of the correlation coefficient matrix. When the approximate chi-square value is large and its corresponding significance is <0.05, the null hypothesis should be rejected, that is, the correlation coefficient matrix is ​​not a unit matrix, there is correlation between the original variables, and factor analysis can be performed. The KMO and Bartlett test results of the data in this study are shown in Table 2. The KMO measure value is 0.588, and the significance in the Bartlett's sphericity test result is 0, indicating that the selected indicators and measurement data are suitable for factor analysis.

[0092] Table 2 KMO and Bartlett test

[0093] The variance contribution rate and cumulative contribution rate of each component were calculated (Table 3). Table 3 shows that the cumulative variance contribution rate of the first four components is 82.252%, which is greater than 80%, meeting the principle of principal component extraction and reflecting most of the information of the original data. Therefore, the first four components were selected as new predictive indicators and rotated using the maximum variance method (as shown in Table 4). Then, the regression method was used to calculate the factor scores, which were recorded as F1, F2, F3, and F4 respectively. The resulting component score coefficient matrix is ​​shown in Table 5. Finally, the score model of the four new components was obtained, as shown in Equation (24). Table 4 shows that for principal component F1, the load contribution of the mining method and mining thickness components is relatively large, which can be summarized as the human mining factor; principal component F2 has a large load on mining depth and working face oblique length, mainly representing the mining depth and working face oblique length factors; principal component F3 has a large load on roof type and coal seam inclination, mainly representing the roof type and coal seam inclination characteristics; principal component F4 has a large load on the presence or absence of faults in the working face, mainly representing the fault factor.

[0094] Table 3 Total variance explained

[0095] Table 4 Rotated component matrix

[0096] Table 5 Component score coefficient matrix

[0097] Where: F1, F2, F3, F4 are the scores of the four components; Xi ′(i=1,2,3,4,5,6,7) is the value after the original data is standardized.

[0098] This reduces the original 7-dimensional factors to 4 dimensions, reducing the model size; at the same time, the four new components generated can reflect and explain the complex relationship between the original variables, eliminate the correlation between the influencing factors, and optimize the input structure of the subsequent SVR model.

[0099] The four principal components {F1, F2, F3, F4} extracted from 40 training sample data sets through factor analysis were used as input features, and the water-conducting fracture zone height {y} of the 40 training sample data sets was used as the model output. A radial basis function was selected as the kernel function for the SVR model, and the ALO algorithm was used to optimize the parameters of the SVR model, searching for the optimal penalty factor C and kernel function parameter g. The ALO algorithm parameters were set as follows: number of ants and ant lions pop = 20, number of variables dim = 2, maximum number of iterations T = 500, lower bound lb = [0, 0], and upper bound ub = [100, 100]. The mean square error (MSE) between the measured and predicted water-conducting fracture zone heights of the samples was used to establish the fitness function [38, 53, 69].

[0100] After initializing each parameter and setting the fitness function, the ALO algorithm can be used to optimize the SVR parameters. The fitness value of each ant is calculated based on the fitness function. If the updated fitness value is better than the fitness value of the previous position, it is replaced and used as the starting point for the next iteration. Through continuous iterations, the optimal parameters are output until the obtained extreme value is less than the set threshold or the maximum number of iterations is reached. The optimal parameters are then used to construct the ALO-SVR prediction model. The entire optimization process was implemented using the MATLAB 2018b platform. The fitness curve during training is shown in Figure 3, and the fit results of the trained model to the training samples are shown in Figure 4. As the number of iterations increases, the fitness value decreases. After iteration 254, the algorithm reaches a plateau, and the fitness value remains unchanged in subsequent iterations, indicating that the model's error rate has reached its lowest level. At this point, the MSE value for the training samples is 9.367e-05, and the goodness-of-fit R² is 0.99957. The support vector machine parameters optimized by the ALO algorithm are C = 32.8628 and g = 18.5497.

[0101] In order to further verify the effectiveness of the prediction model constructed by the present invention, a comparative analysis was conducted with three other models, including the traditional SVR model, the FA-SVR model after factor analysis processing, and the ALO-SVR model without factor analysis processing. Among them, the traditional SVR model uses the original data of 7 main control factors as the model input, and the height of the water-conducting fracture zone as the output of the model. The penalty factor C and the kernel function parameter g in the model are set by trial and error. After many tests, C=10 and g=1 are taken. At this time, the goodness of fit between the predicted value and the actual value of the training sample is R2=0.95668, indicating that the model training effect is good; the FA-SVR model after factor analysis processing uses the scores of the extracted 4 principal components as the model input, and the height of the water-conducting fracture zone as the output of the model. The penalty factor C and the kernel function parameter g in the model are set by trial and error. The number g was also set using trial and error. After multiple experiments, C = 10 and g = 8 were selected. The goodness-of-fit R² of the predicted and actual values ​​of the training samples was 0.96062, indicating good model training results. The ALO-SVR model, without factor analysis, used raw data from seven key control factors as input and the height of the water-conducting fracture zone as output. The model parameters were: population of ants and ant lions, pop = 20, number of variables, dim = 2, maximum number of iterations, T = 500, lower bound, lb = [0, 0], and upper bound, ub = [100, 100]. The penalty factor, C, and kernel function parameter, g, were automatically optimized using the ALO algorithm. The training results were C = 44.8737 and g = 15.4776. The goodness-of-fit R² of the predicted and actual values ​​of the training samples was 0.99968, indicating good model training results.

[0102] The prediction results of the model for 6 groups of test samples are shown in Table 7.

[0103] Table 7 Prediction results of each model for test samples

[0104] The evaluation index values ​​of the prediction performance of different models are shown in Table 8.

[0105] Where: is the predicted value of the model; yi is the actual value; is the average value of the actual sample value; n is the number of test samples, here n = 6.

[0106] Table 8 Comparison of evaluation indicators of prediction performance of different models

[0107] As shown in Table 8, the above four models all show a certain ability to predict new samples, but the degree of prediction performance of each model is different. In terms of prediction accuracy, the FA-ALO-SVR model has the smallest mean absolute error (MAE), root mean square error (RMSE), and mean relative error (RMSE). The FA-SVR model, ALO-SVR, and traditional SVR models performed next, indicating that the FA-ALO-SVR model had the best prediction accuracy for new samples among the four models. In terms of predictive ability, the FA-ALO-SVR model had the smallest TIC value, followed by the FA-SVR model, ALO-SVR model, and traditional SVR model, indicating that the FA-ALO-SVR model had the strongest predictive ability among the four models. In terms of generalization ability, the IWA values ​​of all four models were greater than 0.6, indicating that all four models had good generalization ability and practical predictive value. Among them, the FA-ALO-SVR model had the largest IWA value, followed by the FA-SVR model, traditional SVR model, and ALO-SVR, indicating that the FA-ALO-SVR model had the best generalization ability among the four models.

[0108] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for constructing a water-conducting fracture zone height prediction model based on multi-factor comprehensive analysis, characterized in that: The following steps are involved: Step 1: Establish a prediction index system based on measured data, covering roof type, mining method, depth, inclination, thickness, working face length and fault factors; Step 2: Optimize the data structure using factor analysis, tune the SVR model parameters using the Ant Lion Optimization Algorithm, and construct the FA-ALO-SVR water-conducting fracture zone height prediction model; Step 3: Test the model prediction performance through new samples, and use MAE and RMSE indicators to comprehensively evaluate the model effect and compare it with the traditional model.

2. The method for constructing a water-conducting fracture zone height prediction model according to claim 1, wherein: The factor analysis optimization of step one is characterized by converting the original variables into linear combinations of a smaller number of new variables through the correlation matrix of the variables to simplify the data structure and eliminate redundant information, including calculating the correlation matrix, eigenroots, eigenvectors, determining the number of common factors, performing factor rotation and factor loading calculation, and establishing a factor score model.

3. The method for constructing a water-conducting fracture zone height prediction model according to claim 1, wherein: The SVR model construction and optimization in step 2 is characterized by adopting the SVR regression function, minimizing the sum of model complexity and fitting error under a given error level, introducing the Lagrange multiplier method to solve the dual problem, using the RBF kernel function, and optimizing the penalty factor C and the kernel function parameter g through the ant lion optimization algorithm.

4. The method for constructing a water-conducting fracture zone height prediction model according to claim 1, wherein: The antlion optimization algorithm application is characterized by simulating the antlion hunting process, selecting antlions to build traps through a roulette wheel operation, and ants randomly wandering to find antlions. Antlions capture ants through traps and update their positions based on fitness. At the same time, an elitist strategy is considered to improve algorithm performance.

5. The method for constructing a water-conducting fracture zone height prediction model according to claim 1, wherein: The model evaluation in step 3 is characterized by using five indicators, namely MAE, RMSE, relative error, IWA and TIC, to evaluate the prediction accuracy, capability and generalization ability of the model from different perspectives, and verifying the superiority of the proposed model by comparing with other models.

6. The method for constructing a water-conducting fracture zone height prediction model according to claim 1, wherein: A comprehensive analysis is conducted by combining multiple factors, reducing data complexity and noise through factor analysis, and using the ant lion optimization algorithm to accurately tune the SVR model parameters, thereby improving the model's prediction accuracy and practicality. It is suitable for decision support in mine water hazard prevention and safe mining.

Images