Method for predicting water inrush risk of coal mining working face floor
Through the white whale optimization algorithm and genetic algorithm, the BP neural network is optimized, and the BWO-GA-BP neural network model is constructed, which solves the problems of low prediction accuracy and slow speed in the existing technology, and realizes higher accuracy and faster water outburst prediction of coal seam bottom plates, supporting bottom plate water damage prevention and control.
Patent Information
- Application Number
- CN202510363121.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2025-07-04
AI Technical Summary
When using neural networks to predict water burstability of coal seam base plates, the prior art has problems such as low prediction accuracy, traditional algorithm models are prone to local convergence and slow prediction speed.
The weights and thresholds of the BP neural network are optimized in primary and secondary ways by using beluga optimization algorithm and genetic algorithm to construct a BWO-GA-BP neural network model, and combined with the standardized processing data sample points to predict the water burst risk.
The accuracy and convergence speed of the bottom plate water inrush prediction model are improved, more scientific prediction results are provided, and theoretical basis for bottom plate water damage prevention and control work.
Smart Images
Figure CN120258523A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of prevention and control of mine water hazards, and particularly relates to a method for predicting the water inrush risk of the floor of a coal seam mining face. Background Art
[0002] China is rich in coal resources and has always been and will long remain the mainstay in China's energy structure. Due to the geological conditions at high altitudes in the western region, the mining depth is gradually increasing, and the mine working faces are also facing an increasingly serious threat of water inrush from the floor. The fractures formed in the coal seam floor under the influence of mining activities damage the integrity of the floor water-resisting layer. When the fractures are connected to the confined water, the confined water will rise along the fractures, causing water inrush accidents from the floor, which may pose serious threats to the safety of mine production personnel and economic property losses. Therefore, further improving the support of science and technology for mine safety and reducing the occurrence of water inrush accidents have become major strategic needs for economic construction and development.
[0003] For the multi-factor composite superposition non-linear problem and small sample characteristic problem of coal mine floor water inrush, the artificial neural network can rely on its independent training and learning ability and high-efficiency data processing and analysis ability, and plays an increasingly important role in guiding mine safety production. Compared with other methods, the powerful self-learning ability of the neural network directly hits the non-linear relationship between the data source and the evaluation target, and buffers to a certain extent the impact of the water inrush mechanism on the water inrush risk evaluation work. However, the existing methods for predicting the water inrush of the floor using neural networks still have problems such as low prediction accuracy, easy local convergence of traditional algorithm models when processing mine sample data, and slow prediction speed. How to solve the above problems is the research direction required by this industry. Summary of the Invention
[0004] Aiming at the problems existing in the above-mentioned prior art, the present invention provides a method for predicting the water inrush risk of the floor of a coal seam mining face. By standardizing the collected data and outputting the prediction results through the floor water inrush prediction model, the prediction speed is not only fast, but also the prediction accuracy is high, providing a scientific reference and theoretical basis for the prevention and control of floor water hazards.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a method for predicting the water inrush risk of the floor of a coal seam mining face, and the specific steps are as follows: Step 1: Obtain the geological conditions and borehole data of the mining area to be predicted; Step 2: Based on the geological conditions of the mining area, borehole data and on-site data during the working face mining, determine the main control factors of the water inrush from the coal seam floor; Step 3: Conduct data statistics and standardization processing on each main control factor obtained in Step 2, and establish special topic charts for each main control factor of the water inrush from the coal seam floor; Step 4: Based on the drilling data collected in Step 2 and the on-site data during the working face mining, combined with the water inrush coefficient method, determine the water inrush risk assessment levels of each data sample point in the special topic chart of Step 3; Step 5: Divide each data sample point processed in Step 4, where a part is used as the training set and the remaining part is used as the test set; Step 6: Based on the BP neural network, first use the Beluga Whale Optimization Algorithm (BWO) to perform the first optimization on the weights and thresholds randomly obtained initially by the BP neural network, and then perform the second optimization on the obtained weights and thresholds through the Genetic Algorithm (GA), so as to construct the BWO-GA-BP neural network prediction model as the floor water inrush prediction model; Step 7: Use the training set in Step 5 to train the floor water inrush prediction model in Step 6. After the training is completed, input the test set in Step 5 into the prediction model, so as to realize the water inrush prediction of each data sample point in the test set, and then conduct prediction error analysis in combination with the on-site data to verify the prediction accuracy of the model.
[0006] Furthermore, the main control factors in Step 2 include: thickness of the water-resisting layer, complexity of the structure, fault density, water pressure of the aquifer, burial depth of the coal seam, thickness of the coal seam, and depth of floor failure.
[0007] Furthermore, each special topic chart of the main control factors in Step 3 consists of multiple data sample points (that is, each main control factor determined at the same position forms a data sample point), and each data sample point contains all the main control factors determined in Step 2.
[0008] Furthermore, the specific standardization process in Step 3 is as follows: Use the Mapminmax function in Matlab to perform standardization processing on each data sample point, and the data interval of each main control factor after standardization processing of each data sample point is [0,1]. And since each main control factor is divided into positive factors (that is, the larger the value, the better) and negative factors (that is, the smaller the value, the better), it is necessary to make a distinction according to the situation of each main control factor during the standardization process.
[0009] Furthermore, in Step 5, 80% of all data sample points are used as the training set, and the remaining 20% are used as the test set.
[0010] Furthermore, the specific process of the first optimization by the Beluga Whale Optimization Algorithm in Step 6 is as follows: ① Exploration stage: According to the social behaviors of beluga whales in different postures, perform the exploration stage of the algorithm. For example, when beluga whales swim, they swim in a mirror or synchronous manner, and their position update (1) is as follows: (1) where T is the current iteration number, X i,Pjis the new position of the i-th individual in the j-th dimension at the next iteration; p is a random integer within the data dimension range; X i,Pj and X r,Pj are the positions of the i-th individual and the r-th individual in the random dimension p at the current iteration; r1 and r2 are random numbers in (0, 1); and the transition from the exploration stage to the development stage is achieved through the balance factor B f as follows. The calculation formula (2) is: (2) where B f is a random value in (0, 1) that changes at each iteration, T is the current iteration number, and T max is the maximum iteration number; when the balance factor B f > 0.5, the algorithm is in the exploration stage; when the balance factor B f ≤ 0.5, the algorithm is in the development stage; as the number of iterations increases, the fluctuation range of the balance factor B f shrinks from (0, 1) to (0, 0.5), indicating that the probabilities of the exploration stage and the development stage have changed significantly, that is, the probability of the development stage gradually increases as the number of iterations increases; ② Development stage: The development stage of the beluga whale optimization algorithm comes from the foraging behavior of beluga whales. Beluga whales can cooperate to forage and move according to the positions of nearby beluga whales. Considering the influence of the best individual and other individuals on position update during the iteration process and adding the Levy flight strategy to improve the convergence of the algorithm, the mathematical model (3) of the algorithm is expressed as follows: (3) where T is the current iteration number; X i and X r are the positions of the i-th individual and a random individual; X best is the position of the best individual; C1 is the random jump intensity, used to measure the intensity of the Levy flight; r3 and r4 are random numbers in (0, 1); L F is a random number that conforms to the Levy distribution; ③ Whale fall stage: Beluga whale individuals either migrate to other places or perform whale fall behavior. To ensure that the population size remains unchanged, the current position of the beluga whale and the step length of the beluga whale's fall are used to establish the updated position. Its mathematical model (4) is: (4) where r5, r6, and r7 are random numbers in (0, 1); X step is the step length of the beluga whale's fall, and its calculation formula (5) is: (5) where ub and l b are the upper and lower bounds of the optimization problem respectively; C2 is the step factor, which is determined by the probability of whale fall and the population size: C2 = 2W f ×N, where N is the population size and the probability of whale fall W f The function (6) is: (6) Through the above process, the initial optimization of the weights and thresholds is finally achieved.
[0011] Furthermore, the process of secondary optimization by the genetic algorithm in step six is as follows: population initialization, calculation of initial fitness, selection, crossover, mutation, calculation of individual fitness, judgment of whether the iteration is completed, and finally output of the optimal weights and thresholds.
[0012] Compared with the prior art, the present invention combines the beluga optimization algorithm, the genetic algorithm and the BP neural network to form a floor water inrush prediction model to predict the water inrush danger level of the standardized data sample points; because the beluga algorithm has good global search ability, it can quickly search out all the solutions in the solution space without falling into the local optimal solution trap. The global search performance of the genetic algorithm is general. A pure genetic algorithm is time-consuming and has low search efficiency in the later stage, and is prone to the problem of premature convergence. And the genetic algorithm has a certain dependence on the selection of the initial population. In view of the advantages and disadvantages of the genetic algorithm, the beluga optimization algorithm is selected to perform the initial optimization of the initial weights and thresholds, and the genetic algorithm uses the weights and thresholds after the initial optimization for secondary optimization, which greatly reduces the possibility of the genetic algorithm falling into the local optimal trap. Combining the two with the BP neural network to construct the BWO-GA-BP neural network as the floor water inrush prediction model, which improves the optimization ability and iterative convergence speed of the floor water inrush prediction model, as well as the nonlinear dynamic mapping ability of the model, and improves the learning accuracy of the model. Finally, the floor water inrush prediction model has higher accuracy and faster convergence speed. Then, the test set is used to verify the prediction results of the trained model. After the error is within the allowable range, the water inrush danger is predicted; therefore, the prediction results output by the coal seam floor water disaster prediction model constructed by the present invention can provide a scientific reference and theoretical basis for the prevention and control of floor water disasters. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 is the overall flowchart of the present invention; Figure 2 is the logical flowchart of constructing the floor water inrush prediction model of the present invention; Figure 3 is the comparison chart of the prediction results and prediction errors of the present invention and other neural network models for the same test set; Among them, (a) is the water inrush risk level predicted by each model; (b) is the prediction error of each model. Specific implementation mode
[0014] The present invention will be further described below.
[0015] As Figure 1 shown, the specific steps of the present invention are as follows: Step 1: Collect geological data, borehole data of the mining area to be predicted, and on-site data during the mining of each working face.
[0016] Step 2: Based on the data in Step 1, determine the main controlling factors for water inrush from the coal seam floor, including: thickness of the water-resisting layer A1, complexity of the structure A2, fault density A3, water pressure of the aquifer A4, buried depth of the coal seam A5, thickness of the coal seam A6, and depth of floor failure A7.
[0017] Step 3: Conduct data statistics and standardization processing on the main controlling factors obtained in Step 2. Specifically: Use the Mapminmax function in Matlab to perform standardization processing on each data sample point. The data range of each main controlling factor after standardization processing for each data sample point is [0, 1]; and since each main controlling factor is divided into positive factors (i.e., the larger the value, the better) and reverse factors (i.e., the smaller the value, the better), it is necessary to distinguish according to the situation of each main controlling factor during the standardization process; establish thematic charts for each main controlling factor of water inrush from the coal seam floor; in this embodiment, the thematic charts for each main controlling factor are composed of 60 data sample points, and each data sample point includes all the main controlling factors determined in Step 2; Step 4: According to the borehole data collected in Step 2 and the on-site data during the mining of the working face, combined with the water inrush coefficient method, determine the water inrush risk assessment levels of the 60 data sample points in the thematic chart in Step 3. The specific level values are divided into 0, 1, and 2; where 0 means no risk, 1 means low risk, and 2 means high risk, as shown in Table 1; Step 5: Divide each data sample point processed in Step 4, where 50 data sample points are used as the training set, and the remaining part is used as the test set, as shown in Table 1; Step 6: As Figure 2 shown, based on the BP neural network, first use the beluga whale optimization algorithm (BWO) to perform primary optimization on the weights and thresholds randomly obtained initially by the BP neural network. Specifically: ① Exploration stage: According to the social behaviors of beluga whales in different postures, conduct the exploration stage of the algorithm. For example, when beluga whales swim, they swim in a mirror or synchronous manner, and their position update (1) is as follows: (1) Where T is the current iteration number, and X i,Pj is the new position of the i-th individual in the j-th dimension at the next iteration; p is a random integer within the data dimension range; X i,Pj and X r,Pj are the positions of the i-th individual and the r-th individual in the random dimension p at the current iteration; r1 and r2 are random numbers in (0, 1); and the transition from the exploration stage to the development stage is achieved through the balance factor B f The calculation formula (2) is as follows: (2) Where B f is a random value in (0, 1) that changes continuously at each iteration, T is the current iteration number, and T max is the maximum iteration number; when the balance factor B f > 0.5, the algorithm is in the exploration stage; when the balance factor B f ≤ 0.5, the algorithm is in the development stage; as the number of iterations increases continuously, the fluctuation range of the balance factor B f shrinks from (0, 1) to (0, 0.5), indicating that the probabilities of the exploration stage and the development stage have changed significantly, that is, the probability of the development stage gradually increases as the number of iterations increases; ② Development stage: The development stage of the beluga whale optimization algorithm comes from the foraging behavior of beluga whales. Beluga whales can cooperate to forage and move according to the positions of nearby beluga whales. Considering the influence of the best individual and other individuals on position update during the iteration process and adding the Levy flight strategy to improve the convergence of the algorithm, the mathematical model (3) of the algorithm is expressed as follows: (3) Where T is the current iteration number; X i and X r are the positions of the i-th individual and the random individual; X best is the position of the best individual; C1 is the random jump intensity, which is used to measure the intensity of the Levy flight; r3 and r4 are random numbers in (0, 1); L F is a random number that conforms to the Levy distribution; ③ Whale fall stage: Beluga whale individuals either migrate to other places or perform whale fall behavior. To ensure that the population size remains unchanged, the current position of the beluga whale and the step length of the beluga whale's fall are used to establish the updated position, and its mathematical model (4) is: (4) Where r5, r6, and r7 are random numbers in (0, 1); X step is the step length of the beluga whale's fall, and its calculation formula (5) is: (5) Wherein, u b and l b are respectively the upper bound and the lower bound of the optimization problem; C2 is a step factor, which is determined by the whale fall probability and the population size: C2 = 2W f ×N, where N is the population size, and the function of the whale fall probability W f is as follows: (6) Through the above process, the initial optimization of the weights and thresholds is finally achieved.
[0018] Then, the obtained weights and thresholds are further optimized by the genetic algorithm (GA). The specific process is as follows: population initialization, calculation of the initial fitness, selection, crossover, mutation, calculation of the individual fitness, determination of whether the iteration is completed, and finally output of the optimal weights and thresholds; thus, a BWO-GA-BP neural network prediction model is constructed as the floor water inrush prediction model.
[0019] In this embodiment, the beluga whale optimization algorithm, the genetic algorithm, and the BP neural network can all be implemented by coding with Matlab software. Among them, when establishing the BWO-GA-BP neural network prediction model, it is necessary to optimize the number of hidden layer nodes of the required prediction mining area sample dataset. After multiple verifications of the BP neural network, when the number of hidden layer nodes is 12, the error of the neural network can be minimized. At this time, the number of weights is 96, the number of thresholds is 13, and the total number is 109. The parameters of the neural network are: the maximum number of training times is 200, the learning rate is 0.2, and the target error is less than 10 -8 . The set parameters of the beluga whale algorithm are: the scale is 80, and the number of iterations is 120. Then, the genetic algorithm performs a secondary optimization on the obtained parameters. The set parameters of the genetic algorithm are: the population size is 50, the number of iterations is 120, the crossover probability is 0.6, and the mutation rate is 0.1.
[0020] Step 7: Use the training set in Step 5 to train the floor water inrush prediction model in Step 6. After the training is completed, input the test set in Step 5 into the prediction model, so as to realize the water inrush prediction of each data sample point in the test set as shown in Table 2; in order to verify the prediction accuracy of the model, this embodiment introduces the BWO-BP, GA-BP, and BP neural network prediction models, which are also trained with the same training set in this embodiment and then predict the same test set. The specific performance comparisons are shown in Table 2, Figure 3 a and Figure 3 b, as shown in Table 2 and Figure 3It can be seen that the neural network prediction model after the initial optimization by the beluga optimization algorithm and the secondary optimization by the genetic algorithm of the present invention has smaller errors compared with other existing prediction models, verifying the accuracy of this model. Therefore, the floor water inrush prediction model of the present invention can provide a scientific reference and theoretical basis for the development and research of floor water disaster prevention and control technologies.
[0021] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.
Claims
1. A prediction method for the water inrush risk of the floor in a coal seam mining face, characterized in that, The specific steps are as follows: Step 1: Obtain the geological conditions and borehole data of the required predicted mining area; Step 2: Based on the geological conditions of the mining area, borehole data, and on-site data during the working face mining, determine the main controlling factors of water inrush from the coal seam floor; Step 3: Conduct data statistics and standardization processing on each main controlling factor obtained in Step 2, and establish thematic charts of each main controlling factor for water inrush from the coal seam floor; Step 4: According to the borehole data collected in Step 2 and the on-site data during the working face mining, and in combination with the water inrush coefficient method, determine the water inrush risk assessment levels of each data sample point in the thematic charts of Step 3; Step 5: Divide each data sample point processed in Step 4, with a part of them as the training set and the remaining part as the test set; Step 6: Based on the BP neural network, first use the beluga optimization algorithm to perform primary optimization on the randomly obtained initial weights and thresholds of the BP neural network, and then perform secondary optimization on the obtained weights and thresholds through the genetic algorithm, so as to construct a BWO-GA-BP neural network prediction model as the floor water inrush prediction model; Step 7: Use the training set in Step 5 to train the floor water inrush prediction model in Step 6. After the training is completed, input the test set in Step 5 into the prediction model, so as to realize the water inrush prediction of each data sample point in the test set.
2. The prediction method for the water inrush risk of the floor of a coal seam mining face according to claim 1, wherein, The main controlling factors in Step 2 include: thickness of the water-resisting layer, complexity of the structure, fault density, water pressure of the aquifer, burial depth of the coal seam, thickness of the coal seam, and depth of floor failure.
3. The prediction method for the water inrush risk of the floor of a coal seam mining face according to claim 1, characterized in that, Each thematic chart of the main controlling factors in Step 3 consists of multiple data sample points, and each data sample point contains all the main controlling factors determined in Step 2.
4. The prediction method for the water inrush risk of the floor of a coal seam mining face according to claim 1, wherein The specific standardization processing in Step 3 is as follows: Use the Mapminmax function in Matlab to perform standardization processing on each data sample point, and the data range of each main controlling factor after standardization processing of each data sample point is [0,1].
5. The prediction method for the water inrush risk of the floor of a coal seam mining face according to claim 1, characterized in that, In Step 5, 80% of all data sample points are used as the training set, and the remaining 20% are used as the test set.
6. The prediction method for the water inrush risk of the floor of a coal seam mining face according to claim 1, characterized in that, The specific process of the primary optimization by the beluga optimization algorithm in Step 6 is as follows: ① Exploration stage: Explore the weights and thresholds based on the social behaviors of beluga whales in different postures, and achieve the transition from the exploration stage to the development stage through the balance factor B f Realize the transition from the exploration stage to the development stage; ② Development stage: Use the predation behavior of the beluga to iterate the positions of the weights and thresholds, consider the influence of the best individual and other individuals on the position update during the iteration process, and add the Levy flight strategy to improve the convergence of the algorithm; ③ Whale fall stage: In order to ensure that the population size remains unchanged, use the current position of the beluga and the step size of the beluga's fall to establish the updated position, and finally realize the primary optimization of the weights and thresholds.
7. The prediction method for the water inrush risk of the floor of a coal seam mining face according to claim 1, wherein, The process of the secondary optimization by the genetic algorithm in Step 6 is as follows: population initialization, calculation of the initial fitness, selection, crossover, mutation, calculation of the individual fitness, judgment of whether the iteration is completed, and finally output the optimal weights and thresholds.