Method for predicting damage depth of coal seam floor based on SSA-RF
By combining the Sparrow Search algorithm with the Random Forest algorithm, the instability problem caused by the random setting of hyperparameters in the Random Forest model is solved, and higher accuracy and faster speed are achieved in predicting the depth of coal seam floor failure.
Patent Information
- Application Number
- CN202410979066.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-15
AI Technical Summary
Existing random forest algorithms, when predicting the depth of coal seam floor failure, suffer from unstable results due to the random setting of hyperparameters such as the number of trees and the minimum number of leaf points, which affects prediction accuracy and speed.
By combining the Sparrow Search algorithm with the Random Forest algorithm, and optimizing the number of trees and the minimum number of leaf points, an SSA-RF model is established to improve the stability and accuracy of predictions.
It improves the accuracy and reliability of predicting the depth of base plate failure, reduces cost and time consumption, and significantly improves the model's generalization ability and iteration speed.
Smart Images

Figure CN118964978A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of coal mining and relates to a method for predicting the damage depth of a coal seam floor based on a machine learning algorithm. Background Art
[0002] Coal mining will lead to the formation of "three lower zones" in the floor rock strata, which are the floor damage zone, the middle protection zone and the pressure-bearing lifting zone. The floor damage depth refers to the deepest distance of the water-conducting fissures in the floor water-conducting damage zone. The study of the floor damage law is of great significance to the prevention and control of mine water hazards.
[0003] At present, the main methods for calculating the depth of floor damage are: drilling water injection method, empirical formula method, numerical model establishment, acoustic CT detection test, mine electrical profile method, water inrush coefficient method, multivariate nonlinear model, etc. Among them, the drilling water injection method is difficult to implement, time-consuming and costly; the empirical formula method is easy to use, although the formula is constantly revised, it still has certain limitations; although the numerical simulation method predicts that the results are not much different from the theoretical calculation results, it is easy to ignore other influencing factors such as structural surfaces and faults; using acoustic CT detection tests, the dynamic change law of floor damage within a certain range of the working face can be grasped, and the results are true and reliable, but this method requires the arrangement of instruments in two boreholes around the working face, which is costly and has a limited observation range and limitations.
[0004] Machine learning is economical, convenient, and accurate, and is widely used. At present, machine learning has formed unique advantages in predicting the depth of floor damage. Among them, Random Forest is an ensemble learning algorithm based on decision trees. Since a single decision tree is more sensitive to the noise of the training set, RF uses the Bagging algorithm in ensemble learning to combine multiple decision trees together. On the basis of a given data set, the data set is selected with replacement, and a part of the features are randomly selected as input, thereby reducing the correlation between the trained multiple decision trees; RF can reduce overfitting compared to decision trees, bring better performance to the model, and have a faster learning speed; the specific process of the RF algorithm can be seen in Figure 1 However, the prediction accuracy and speed of the RF model are greatly affected by hyperparameters such as the number of trees and the minimum number of leaf points. The hyperparameters of the random forest, such as the number of trees and the minimum number of leaf points, are randomly set, which will lead to unstable results. Summary of the invention
[0005] In order to solve the defects existing in the RF machine learning method, the present invention is based on the RF model and combines the sparrow search algorithm (SSA) with RF. Aiming at the disadvantages of hyperparameters such as the number of trees and the minimum number of leaf points in RF, the present invention provides a prediction method for the coal seam floor damage depth based on SSA-RF, and uses SSA to optimize the RF algorithm to obtain a prediction model with stronger generalization ability.
[0006] Concept of the present invention: The present invention adopts machine learning technology, especially the random forest (RF) algorithm, combined with the sparrow search algorithm (SSA) for optimization, and proposes an improved method for predicting the depth of bottom plate damage. The random forest algorithm reduces the overfitting of the model by integrating multiple decision trees and improves the generalization ability of the prediction, while the sparrow search algorithm optimizes the key hyperparameters of these decision trees, such as the number of trees and the minimum number of samples of leaf nodes, so that the model prediction is more stable and accurate. By analyzing key parameters such as the bottom plate burial depth, inclination, mining height, inclined length and bottom plate anti-destruction ability, the present invention establishes an optimized bottom plate damage depth prediction model, and verifies its prediction accuracy and efficiency through experiments. The implementation of this technical solution not only improves the accuracy and reliability of the prediction of the bottom plate damage depth, but also significantly reduces the cost and time consumption, and has significant practical application value for mine safety production and water hazard prevention and control.
[0007] Based on the RF model, the present invention takes advantage of the global search capability of SSA to optimize the two parameters of the number of trees and the minimum number of leaf points in the RF model, so that the fitting result in the prediction of the bottom plate damage depth is closer to the actual value.
[0008] The present invention provides a prediction method for coal seam floor damage depth based on SSA-RF, comprising the following steps: combining the influencing factors of coal seam floor water inrush such as mining depth, thickness, coal seam inclination, working face inclined length, coal seam mining method and impermeable layer thickness, collecting and collating to obtain a relatively complete data set of each factor; then dividing the data set after preliminary processing into a test set and a training set; selecting all training sample data, establishing an RF model, and outputting an unoptimized floor damage depth result after determining relevant parameters; then optimizing the RF model by using a sparrow search algorithm to obtain an RF-SSA model, outputting the optimized floor damage depth result through the model, and comparing the relative error and mean square error between the predicted values and the true values of the two models to determine whether the SSA optimization is effective.
[0009] In the present invention, the sparrow search algorithm (SSA) is used to optimize the key hyperparameters of the random forest (RF) model, including the number of decision trees and the minimum number of leaf nodes of the tree. The specific optimization process is as follows: (1) First, the position of each sparrow is defined in the n-dimensional parameter space, and these positions represent different decision tree configurations. Using the initial random forest model, the mean square error (MSE) of the model is calculated on a given data set as the fitness function of the SSA model. Based on this fitness function, the initial fitness value of each sparrow is calculated to determine the initial individual optimal and global optimal solutions. (2) In each iteration, the position of the sparrow is updated according to the strategies of the "discoverer" and "joiner". Each sparrow re-evaluates the fitness value according to its new position. If the fitness of the new position is better than the current position, it is updated to the optimal position. During the iteration process, the position of the entire sparrow population is continuously adjusted according to the individual historical optimal solution and the group historical optimal solution, and the parameter settings are gradually refined. (3) After setting an appropriate number of iterations, the iteration is continued until the position of the sparrow colony is stable. At this time, the number of trees and the minimum number of leaf nodes achieved are the optimal solution, thereby optimizing the performance and prediction accuracy of the random forest model.
[0010] The above-mentioned method for predicting the failure depth of coal seam floor based on SSA-RF specifically includes the following steps:
[0011] S1: Data collection: Based on previous research conclusions and coal seam mining conditions, the floor failure depth and related influencing factor data sets corresponding to the mine working face in the mining area are collected;
[0012] S2: Data classification: The dimension-eliminated data is divided into a training set and a test set. The training set is 80%-90% of the entire data set, and the rest is the test set.
[0013] S3: Establish RF model: After defining the model including the number of trees, the minimum number of leaf points and the depth of the tree, the relationship between each influencing factor and the bottom plate damage depth is obtained through the training set data, and the model is obtained;
[0014] S4: Establish SSA-RF model: optimize the number of trees and the minimum number of leaf points of each influencing factor in the RF model through the SSA sparrow search algorithm, and obtain the relationship between each influencing factor and the coal seam floor failure depth through the training set data to obtain the optimized model;
[0015] S5: Calculation of prediction value: The test set data is brought into the two models RF and SSA-RF (the number of trees and the minimum number of leaf points obtained by the sparrow algorithm are transferred into the RF model, and then the test set is input into the SSA-RF model to obtain the optimized prediction result), and the prediction values of the floor damage depth of the two models are obtained respectively;
[0016] S6: Evaluation and analysis: Compare the predicted values of RF and SSA-RF with the actual values, and calculate the relative error and mean square error between the predicted values and the actual values, so as to determine whether the optimized prediction model is effective.
[0017] The specific steps of S3: establishing the RF model are as follows:
[0018] S31: Perform random sampling without replacement N times on a data set of size K, and select a sample each time; then, use the selected N training samples to form a training sample set from the sampled data;
[0019] S32: When splitting each node of the decision tree, randomly select m attributes from the M attributes of each sample, satisfying the condition m<<M, which helps to avoid excessive attention to noise during model training, thereby improving the model's prediction ability for new data; select a feature A from the randomly selected m attributes for splitting. Feature A refers to the specific attribute used to split the data set into two subsets; for the selected feature A, select the best split point from all possible split points s; then select a split point s from the m attributes as the node splitting attribute according to the following formula (1);
[0020] min[min∑(y i -v1) 2 +min∑(y i -v2) 2 ] (1)
[0021] Among them, y i Represents the specific value of sample i on a certain attribute; v1 represents the average value of all samples in the subset to the left of the current split point on this attribute; v2 represents the average value of all samples in the subset to the left of the current split point on this attribute.
[0022] S33: During the formation of the decision tree, each node is split according to S32 until it cannot be split any more. It is worth noting that no pruning is performed during the formation of the entire decision tree.
[0023] S34: Establish a large number of decision trees according to steps S31 to S33, thereby forming a random forest, and average the results of all decision trees to determine the final result;
[0024]
[0025] Where: T is the final result of random forest; T1, T2, …T N is the result of N decision trees.
[0026] The number of trees and the minimum number of leaf points in the RF random forest are used as the parameters to be optimized for the SSA algorithm in S4.
[0027] The specific steps of the above S4: establishing the SSA-RF model are as follows:
[0028] S41: Initialize the population
[0029] Set up a sparrow group in the d-dimensional search space, with the number of n. Use a matrix to represent the positions of the sparrows:
[0030]
[0031] At the same time, set the number of discoverers, the number of sparrows aware of danger, and the maximum number of iterations iter max Alarm threshold R2.
[0032] S42: Take the Mean Squared Error (MSE) in the RF model as the fitness function. Represent the fitness values of all sparrows using the following vector Fx:
[0033]
[0034] Among them: Each value in each row of Fx represents the fitness of an individual; find the best fitness value and the worst fitness value, and define them as f g and f w .
[0035] S43: Update the positions of the discoverers in each iteration using the following formula (5):
[0036]
[0037] In the formula: t is the current iteration number; j is the current dimension; is the j-th dimension position of the i-th sparrow at the t-th iteration; iter max is the maximum number of iterations; α ∈ [0, 1] is a random number; R2 and ST are the alarm threshold and safety threshold respectively. Usually, R2 ∈ [0, 1] and ST ∈ [0.5, 1]; Q is a random number obeying the normal distribution; L is a d-dimensional unit vector, where d is the dimension of the variable to be optimized and each element is 1.
[0038] If R2 < ST, that is, there is no predator nearby, the discoverers enter the wide-area search mode. If R2 ≥ ST, it means that there are predators near some sparrows, and all sparrows need to fly to other safe areas quickly.
[0039] Update the positions of the joiners in each iteration using the following formula (6):
[0040]
[0041] In the formula: is the position of the i-th sparrow in the j-dimension in the t+1-th iteration;
[0042] is the worst position in the entire population at the tth iteration; is the best position of the finder in the t+1th iteration; A is a 1×d-dimensional matrix with the same dimension as the input, in which each element is randomly assigned 1 or -1, and A + =A T (AA T ) -1 ; n is the number of sparrows;
[0043] when When , it means that the i-th hungry joiner with a poor fitness value is most likely to starve; When , the joiner follows the discoverer's position to update.
[0044] The initial position of the sentinel is randomly generated, accounting for 10% to 20% of the population. The position of the sentinel is updated in each iteration using the following formula (7):
[0045]
[0046] Where: is the global optimal position in the entire population at the tth iteration; λ is the step length control parameter, which obeys a normal distribution with a mean of 0 and a variance of 1; J∈[-1, 1] is a random number, J represents the direction in which the sparrow moves, that is, the step length control coefficient; f i is the fitness value of the current sparrow; f g and f w is the global best fitness value and the worst fitness value; ε is defined as a very small constant in order to prevent f i -f w =0.
[0047] When f i >f g When , it indicates the position change of the sparrows at the edge of the group; f i =f g , it means that the sparrows in the middle of the population are aware of the danger and need to get closer to other sparrows.
[0048] The position of each sparrow is continuously updated through iteration. After setting a suitable number of iterations, the position of each sparrow will change with the iteration of the sparrow group according to the individual historical optimal value and the global optimal value, so that the sparrow population gradually approaches the optimal solution.
[0049] Beneficial effects of the present invention:
[0050] (1) In the prediction of the bottom plate damage depth, the present invention uses the SSA sparrow search algorithm to optimize the two parameters of the number of trees and the minimum number of leaf points in the RF model, so that the model has stronger generalization ability while ensuring the iteration speed and is not prone to overtraining;
[0051] (2) By using the sparrow search algorithm to improve the RF model, the problem of unstable results caused by the random setting of hyperparameters such as the number of trees and the minimum number of leaf points in RF was improved;
[0052] (3) Compared with RF, SSA-RF converges faster, and both the relative error and the mean square error are significantly reduced when compared with the actual values. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 Represents the model structure of RF;
[0054] Figure 2 Represents the algorithm flow of SSA-RF;
[0055] Figure 3 It represents the iteration result of the SSA-RF model in the embodiment;
[0056] Figure 4 Indicates the degree of fit of the actual values of the two models;
[0057] Figure 5 The factors affecting the damage depth of the floor of No. 3 coal seam in the embodiment are as follows: (a) burial depth; (b) inclination angle; (c) mining height; (d) anti-damage capability of the floor;
[0058] Figure 6 The figure shows the prediction result of the floor damage depth of the No. 3 coal seam using the SSA-RF model in a specific embodiment. DETAILED DESCRIPTION
[0059] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.
[0060] The method for predicting the failure depth of coal seam floor based on SSA-RF provided by the embodiment of the present invention has the following process: Figure 2 As shown, the following steps are included:
[0061] Step 1: Based on previous research conclusions and coal seam mining conditions, a set of influencing factors related to the floor failure depth of the mine working face in the mining area is collected.
[0062] Step 2: Eliminate the dimension of the data collected in step 1 and divide it into training set D train and the test set D test .
[0063] Step 3: Build RF model
[0064] S31: Perform random sampling without replacement N times on a data set of size K, and select a sample each time. Then, use the selected N training samples to form a training sample set from the sampled data.
[0065] S32: When splitting each node of the decision tree, randomly select m attributes from the M attributes of each sample, satisfying the condition m<<M, which helps to avoid excessive attention to noise during model training, thereby improving the model's prediction ability for new data; select a feature A from the randomly selected m attributes for splitting. Feature A refers to the specific attribute used to split the data set into two subsets; for the selected feature A, select the best split point from all possible split points s; then select a split point s from the m attributes as the node splitting attribute according to the following formula (1);
[0066] min[minΣ(y i -v1) 2 +minΣ(y i -v2) 2 ] (1)
[0067] S33: During the formation of the decision tree, each node is split according to S32 until it cannot be split. It is worth noting that no pruning is performed during the formation of the entire decision tree.
[0068] S34: Establish a large number of decision trees according to S31 to S33, and then form a random forest, and take the average of the results of all decision trees to determine the final result.
[0069]
[0070] Where: T is the final result of random forest; T1, T2, ...T N is the result of N decision trees.
[0071] The number of trees and the minimum number of leaf points in the RF random forest are used as the parameters to be optimized for the SSA algorithm in step 4.
[0072] Step 4: Establish the SSA algorithm model and bring it into the RF model to optimize the number of trees and the minimum number of leaf points.
[0073] The specific optimization process is:
[0074] S41: Initialize the population
[0075] Set up a group of sparrows in the d-dimensional search space, the number is n, and use the matrix to represent the position of the sparrows:
[0076]
[0077] At the same time, set the number of discoverers, the number of sparrows aware of the danger, and the maximum number of iterations iter max , alarm threshold R2.
[0078] S42: Using the mean squared error (MSE) in the RF model as the fitness function, the fitness values of all sparrows are represented by the following vector.
[0079]
[0080] The value of each row of Fx represents the fitness of the individual. And find the best fitness and the worst fitness value, which are defined as f g and f w .
[0081] S43: Update the position of the finder in each iteration using the following formula:
[0082]
[0083] Where: is the j-th dimension position of the i-th sparrow at the t+1-th iteration; t is the current iteration number; j is the current dimension; is the j-th dimension position of the i-th sparrow at the t-th iteration; iter max is the maximum number of iterations; α∈[O, 1] is a random number; R2 and ST are the alarm threshold and safety threshold respectively, usually R2∈[0, 1], ST∈[0.5, 1]; Q is a random number that follows a normal distribution; L is a d-dimensional unit vector, where d is the dimension of the variable to be optimized and each element is 1.
[0084] If R2 < ST, that means there are no predators nearby, the discoverer enters wide-area search mode. If R2 ≥ ST, it means there are predators near some sparrows, and all sparrows need to fly quickly to other safe areas.
[0085] The position of the joiner is updated in each iteration using the following formula:
[0086]
[0087] Where: is the worst position in the entire population at the tth iteration; is the best position of the finder in the t+1th iteration; A is a 1×d-dimensional matrix with the same dimension as the input, in which each element is randomly assigned 1 or -1, and A + =A T (AAT ) -1 ; n is the number of sparrows;
[0088] when When , it means that the i-th hungry joiner with a poor fitness value is most likely to starve; When , the joiner follows the discoverer's position to update.
[0089] The initial position of the sentinel is randomly generated, accounting for 10% to 20% of the population, and the position of the sentinel is updated in each iteration using the following formula:
[0090]
[0091] Where: is the global optimal position in the whole population at the tth iteration; λ is the step length control parameter, which obeys the normal distribution with mean 0 and variance 1; J∈[-1,1] is a random number, J represents the direction of the sparrow's movement, that is, the step length control coefficient; f i is the fitness value of the current sparrow; f g and f w is the global best fitness value and the worst fitness value; ε is defined as a very small constant in order to prevent f i -f w =0.
[0092] When f i >f g When , it indicates the position change of the sparrows at the edge of the group; f i =f g , it means that the sparrows in the middle of the population are aware of the danger and need to get closer to other sparrows.
[0093] The position of each sparrow is continuously updated through iteration. When a suitable number of iterations is set, the position of each sparrow will change with the iteration of the sparrow group according to the individual historical optimal value and the global optimal value, so that the sparrow population gradually approaches the optimal solution. Figure 2 .
[0094] Step 5: The parameters such as the number of trees and the minimum number of leaf points obtained by the sparrow algorithm are transferred into the RF model, and then the test set is input into the SSA-RF model to obtain the optimized prediction results.
[0095] Step 6: Perform error analysis on the final result and the measured value to evaluate the performance of the SSA-RF model. The relative error and mean square error are used as error analysis indicators. The formula is as follows:
[0096]
[0097]
[0098] The performance of the prediction model is evaluated based on the above two indicators and verified in the following specific examples. Specific embodiment:
[0100] The mine field is located in Mazhuang Village, Qinchi Town, 25km northwest of Yangcheng County, Jincheng City, Shanxi Province. It is located on the southern edge of the Qinshui Basin on the west side of the Taihang Mountains and is a low mountain and hilly area. The mine field area is 9.2582km 2 , in an irregular polygon, with a maximum length of 4051m from east to west and a maximum width of about 2812m from north to south. The structural form of the mine field is mainly folded, and the strata generally dip to the north with a dip angle of 4° to 14°. No. 3 coal is the current coal seam of the mine, located in the middle and lower part of the Shanxi Formation. The coal seam is 3.04 to 6.48m thick, with an average thickness of 4.48m, and is a stable and mineable coal seam in the entire area ( Figure 1 ). The structure of the coal seam is simple to complex; the pseudo-roof of the coal seam is carbonaceous mudstone, the thickness of the direct roof is 7.92 to 8.49m, with an average of 8.12m, mainly mudstone or sandy mudstone. The thickness of the bottom plate is 6.47 to 8.55m, with an average of 7.80m, mainly mudstone and silty mudstone. Based on the working face profile and geological conditions of the coal mine, this method is used to predict the depth of bottom plate damage. The specific steps are as follows:
[0101] Step 1) The main factors affecting the damage depth include: mining depth, thickness, coal seam inclination, working face oblique length, coal seam mining method and waterproof layer thickness. According to the selected working face and previous research conclusions, this patent selects 5 factors as the main control factors, namely coal seam burial depth, coal seam inclination, mining height, working face oblique length and floor anti-destruction ability. A total of 44 sets of sample data of water-conducting fracture zone height were collected. Among them, 37 sets of data were used as training sets, and the remaining 7 sets of data were used as test sets. Some training data are shown in Table 1 below:
[0102] Table 1. Some samples of base plate damage depth
[0103]
[0104]
[0105] Step 2) RF random forest model.
[0106] First, the RF model structure is initialized, the number of trees is set to 20, the minimum number of leaf points is randomly generated, and the mean square error between the predicted value and the true value is used as the fitness of the optimization algorithm.
[0107] Step 3) Use the SSA algorithm to optimize the number of trees and the minimum number of leaf points. Taking the coal seam depth as an example, the specific operations are as follows:
[0108] a) Initialization of the sparrow population
[0109] In the 2D search space, a particle swarm is set up with a number of 20. The number of joiners, discoverers and guards is randomly generated in proportion. The position of the i-th sparrow can be expressed as x i ={x i1 , x i2 , x i3 , ..., x iD}, i=1, 2, 3,..., N.
[0110] Each solution is represented by the current position of the sparrow, and the corresponding fitness value is calculated through the fitness function.
[0111] The initial position and identity of the sparrows are randomly assigned within a given range.
[0112] b) Update the location of the sparrow
[0113]
[0114]
[0115]
[0116] The above formula is used to iteratively update the positions of the joiner, discoverer, and alerter to obtain their new positions.
[0117] c) Calculate the updated fitness. For each sparrow, compare its fitness value with the best position f it has passed. g Compare and if it is better, take it as the current best position f g .
[0118] d) Repeat the process of calculating fitness ~ individual historical optimal solution and group historical optimal solution, and constantly adjust the position of each sparrow until 30 iterations are reached or the fitness value reaches the termination threshold. After the termination condition is met, the position of the current optimal sparrow is output as the number of trees of this influencing factor and the estimated position of the minimum number of leaf points, and the optimal value is output. The solution process of other influencing factors is the same;
[0119] e) Substitute the training set into the optimized SSA_RF model, output 7 water-conducting fracture zone height values, and inversely normalize the results to obtain the final results.
[0120] Step 4) Compare the iteration speed and fitting degree of RF and SSA_RF, and evaluate the two models by relative error and mean square error, as shown in Table 2.
[0121] Table 2. Relative error between model test value and actual value
[0122]
[0123] From Table 2, we can see that the relative error of the unoptimized RF algorithm is as low as 3.77% and as high as 20.68%. After SSA optimization, the relative error is as low as 1.16% and as high as 9.06%, indicating that the relative error is significantly reduced after optimization, and the predicted value is closest to the measured value. The mean square error of SSA-RF is 0.3908, and the mean square error of RF is 2.428, which further proves the effectiveness of the particle swarm optimization algorithm. Figure 4 It can be concluded that the maximum absolute error of the SSA-RF result is only 0.8789, the prediction result is more accurate and stable, the SSA-RF curve is very close to the measured results, and is significantly better than the RF prediction result.
[0124] Based on the SSA-RF model established above, using the mine geological exploration data, the five factors of coal seam burial depth, coal seam inclination, mining height and floor anti-destruction ability of No. 3 coal in Zhulinshan Coal Mine were first analyzed respectively, and the inclined length of the mine working face was taken as 267m.
[0125] First, the mining area was divided into 75,317 grid pixels based on ArcGis, and the distribution thematic maps of burial depth, inclination, mining height and floor anti-destruction capacity were made in turn. Figure 5 It can be seen that the floor depth of No. 3 coal gradually increases from the south to the north of the mine, and the maximum can reach 468.12m. The inclination of No. 3 coal is basically less than 10°, and the average mining height is about 4.48m. The floor destruction capacity of the western part of the mine is stronger than that of the eastern part. By extracting the values of the main control factors of each grid, based on the above prediction model, the floor destruction depth of 75,317 grids in the mine is predicted, and finally the prediction distribution results of the floor destruction depth of No. 3 coal in Zhulinshan Coal Mine are formed, as shown in the figure. Figure 6 According to the prediction results, the depth of bottom plate damage in the mining area gradually decreases from southwest to northeast, from about 60m to about 40m, with a maximum value of 60.35m, a minimum value of 37.35m, and an overall average value of 50.63m. This prediction result can provide an important theoretical basis for the prevention and control of mine water hazards in the No. 3 coal production of the mine.
Claims
1. A method for predicting the failure depth of coal seam floor based on SSA-RF, characterized in that The following steps are involved: Combined with the influencing factors of water inrush from coal seam floor, a relatively complete data set of each factor is collected and sorted; then the preliminarily processed data set is divided into a test set and a training set; All training sample data are selected to establish the RF model. After determining the relevant parameters, the unoptimized bottom plate damage depth results are output. The RF model is then optimized using the sparrow search algorithm to obtain the RF-SSA model, and the optimized bottom plate damage depth results are output through this model.
2. The method for predicting coal seam floor failure depth based on SSA-RF according to claim 1 is characterized in that The steps include: S1: Data collection: Based on previous research conclusions and coal seam mining conditions, the floor failure depth and related influencing factor data sets corresponding to the mine working face in the mining area are collected; S2: Data classification: The dimension-eliminated data is divided into a training set and a test set. The training set is 80%-90% of the entire data set, and the rest is the test set. S3: Establishing RF model: After defining the model including the number of trees, the minimum number of leaf points and the depth of the tree, the relationship between each influencing factor and the bottom plate damage depth is obtained through the training set data, and the model is obtained; S4: Establish SSA-RF model: optimize the number of trees and the minimum number of leaf points of each influencing factor in the RF model through the SSA sparrow search algorithm, and obtain the relationship between each influencing factor and the coal seam floor failure depth through the training set data to obtain the optimized model; S5: Calculation of predicted value: The test set data is brought into the SSA-RF model to obtain the predicted value of the bottom plate damage depth of the model.
3. The method for predicting coal seam floor failure depth based on SSA-RF according to claim 2, It is characterized in that: S3: The specific steps of establishing the RF model are as follows: S31: Perform random sampling without replacement N times on a data set of size K, and select a sample each time; then, use the selected N training samples to form a training sample set from the sampled data; S32: When splitting each node of the decision tree, randomly select m attributes from the M attributes of each sample, satisfying the condition m<<M, which helps to avoid excessive attention to noise during model training, thereby improving the model's prediction ability for new data; select a feature A from the randomly selected m attributes for splitting; feature A refers to a specific attribute used to split the data set into two subsets; for the selected feature A, select an optimal split point from all possible split points s; then select a split point s from the m attributes as the node splitting attribute according to the following formula (1); min[min∑(y i -v1)2 + min∑(y i -v2) 2 ] (1) Among them, y i Indicates the specific value of sample i on a certain attribute; v1 indicates the average value of all samples in the subset to the left of the current split point on this attribute; v2 indicates the average value of all samples in the subset to the left of the current split point on this attribute; S33: During the formation of the decision tree, each node is split according to S32 until it cannot be split any more. It is worth noting that no pruning is performed during the formation of the entire decision tree. S34: Establish a large number of decision trees according to steps S31 to S33, thereby forming a random forest, and average the results of all decision trees to determine the final result; Where: T is the final result of random forest; T1, T2, …T N is the result of N decision trees; The number of trees and the minimum number of leaf points in the RF random forest are used as the parameters to be optimized for the SSA algorithm in S4.
4. The method for predicting coal seam floor failure depth based on SSA-RF according to claim 2 is characterized in that: S4: the specific steps of establishing the SSA-RF model are as follows: S41: Initialize the population Set up a group of sparrows in the d-dimensional search space, the number is n, and use the matrix to represent the position of the sparrows: At the same time, set the number of discoverers, the number of sparrows aware of the danger, and the maximum number of iterations iter max , alarm threshold R2; S42: The mean square error in the RF model is used as the fitness function, and the fitness values of all sparrows are represented by the following vector Fx: in: The value of each row of Fx represents the fitness of the individual; find the best fitness value and the worst fitness value, which are defined as f g and f w ; S43: Update the position of the discoverer in each iteration using the following formula (5): Where: t is the current number of iterations; j is the current dimension; is the j-th dimension position of the i-th sparrow at the t-th iteration; iter max is the maximum number of iterations; α∈[0,1] is a random number; R2 and ST are the alarm threshold and safety threshold respectively, usually R2∈[0,1], ST∈[0.5,1]; Q is a random number that follows a normal distribution; L is a d-dimensional unit vector, where d is the dimension of the variable to be optimized, and each element is 1; If R2<ST, that is, there is no predator nearby, the discoverer enters the wide-area search mode; if R2≥ST, it means that there are predators near some sparrows, and all sparrows need to fly quickly to other safe areas; The position of the joiner is updated in each iteration using the following formula (6): Where: is the position of the i-th sparrow in the j-dimension in the t+1-th iteration; is the worst position in the entire population at the tth iteration; is the best position of the finder in the t+1th iteration; A is a 1×d-dimensional matrix with the same dimension as the input, in which each element is randomly assigned 1 or -1, and A + =A T (AA T ) -1 ; n is the number of sparrows; when When , it means that the i-th hungry joiner with a poor fitness value is most likely to starve; When , the joiner follows the discoverer's position to update; The initial position of the sentinel is randomly generated, accounting for 10% to 20% of the population. The position of the sentinel is updated in each iteration using the following formula (7): Where: is the global optimal position in the entire population at the tth iteration; λ is the step length control parameter, which obeys a normal distribution with a mean of 0 and a variance of 1; J∈[-1,1] is a random number, J represents the direction in which the sparrow moves, that is, the step length control coefficient; f i is the fitness value of the current sparrow; f g and f w is the global best fitness value and the worst fitness value; ε is defined as a very small constant in order to prevent f i -f w =0; When f i >f g When , it indicates the position change of the sparrows at the edge of the group; f i =f g , it means that the sparrows in the middle of the population are aware of the danger and need to get closer to other sparrows.
Citation Information
Cited By
Coal mining water flowing fractured zone development height prediction method based on SSA-RF model
CN120046793A
Coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment
CN120107511A
Frozen soil zone thickness evolution analysis method based on sparrow search algorithm optimization
CN121412959A
