Coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment
By employing search space pruning and multi-group dynamic adjustment methods in CT reconstruction of underground coal mine working faces, the problem of grid scale selection under incomplete projection conditions was solved, achieving efficient and accurate coal mine working face reconstruction and improving the stability and efficiency of the reconstruction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNIV OF MINING & TECH
- Filing Date
- 2025-03-07
- Publication Date
- 2026-06-09
AI Technical Summary
In CT reconstruction of underground coal mine working faces, the difficulty in selecting the grid scale due to incomplete projection conditions makes it difficult for existing technologies to achieve efficient and accurate reconstruction, especially in the case of exploration areas and uneven ray density, resulting in large reconstruction errors and low efficiency.
A method based on search space pruning and multi-population dynamic adjustment is adopted. The exploration area of the working face is divided into multi-scale grids, and the grid scale is optimized by dynamic multi-population genetic algorithm. A multi-scale reconstruction objective function is constructed, and a fine search is performed to improve the reconstruction accuracy and efficiency.
It achieves efficient and accurate reconstruction of coal mine working faces under incomplete projection conditions, improves the stability and efficiency of reconstruction results, reduces the number of grids that are not passed through by the observed rays, and improves calculation speed and accuracy.
Smart Images
Figure CN120107511B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coal mine working face reconstruction technology under incomplete projection conditions, and in particular to a coal mine working face reconstruction method based on search space pruning and multi-group dynamic adjustment. Background Technology
[0002] Intelligent mining in coal mines is a trend towards high-quality development in the industry. Refined reconstruction and prediction of the working face are the foundation and prerequisite for reducing, minimizing, or even eliminating manpower in mines. Electromagnetic CT (EMC) technology for working faces, as a non-contact detection method, is widely used in coal mining enterprises due to its low cost and high efficiency. However, due to the limitations of the enclosed space and narrow, elongated distribution of the working face in underground coal mines, the scanning angle of the detection and observation space is limited. The effective CT projection data collected is often incomplete and highly sparsity. Reconstruction technology under these incomplete projection conditions has always been a research hotspot in the industry.
[0003] Currently, when performing CT reconstruction calculations on coal mine working faces, the detection and reconstruction area needs to be discretized into regular rectangular units. The scale of the subdivided grid units directly affects the efficiency and resolution of the inversion calculation. Generally, the grid subdivision scale is determined based on the scale of the exploration area and the distribution of emission and reception points. A suitable grid subdivision scale can accurately reflect the coal seam occurrence state within the working face. Due to limitations in exploration conditions and underground structure conditions, the coverage of emission points, reception points, and ray density is often uneven, making it difficult to select a grid scale suitable for the entire calculation area. In 2001, E. Kisslin et al. proposed a three-grid method, which uses a smaller grid scale for ray tracing and selects the most suitable scale as the velocity structure of the detection medium. Using a larger scale for inversion calculation is the optimal model parameterization method. In 2003, Hu Wei, in the selection of grid scale for seismic travel time inversion, used multiple grid scales from small to large for inversion calculations, and took the average of the calculation results of each grid scale as the final result according to the set weights. This method solves the problem of uneven ray coverage to some extent, but it increases the computation time significantly. In 2005, Ma Detang et al. also proposed a dual-grid seismic tomography method, but for the sake of computational convenience, the imaging grid scale is required to be an integer multiple of the ray tracing grid.
[0004] Existing research has found that achieving good reconstruction results hinges on determining the optimal grid scale for both forward and inversion calculations before computation and performing calculations accordingly. However, in practical problems, selecting the optimal scale for forward and inversion is not easy, and how to finely match the grid subdivision scale to the reconstruction needs of different working face observation systems has not yet been effectively improved.
[0005] Therefore, it is necessary to provide a multi-scale adaptive grid partitioning strategy that solves the grid scale partitioning problem by pruning the reconstruction search space, so as to make full use of the ray coverage information of the working face CT exploration observation system. Summary of the Invention
[0006] The purpose of this invention is to provide a coal mine working face reconstruction method based on search space pruning and multi-group dynamic adjustment, which can perform fine grid division of the working face exploration area, and obtain the optimal solution in terms of overall reconstruction accuracy, stability and efficiency, so as to achieve effective reconstruction of coal mine working faces under incomplete CT projection conditions.
[0007] To achieve the above objectives, this invention provides a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment, comprising the following steps:
[0008] S1. Perform preliminary grid division on the exploration area of the working face, and perform multi-scale grid division by searching space pruning based on the sum of the ray intercepts in each network.
[0009] S2. Based on the divided multi-scale grid, a multi-scale reconstruction objective function is constructed, and a super population is obtained by performing population elimination and dynamic adjustment through a dynamic multi-population genetic algorithm. Then, a fine search is performed on the super population to iteratively solve the multi-scale reconstruction objective function and obtain the electromagnetic wave absorption coefficient corresponding to the multi-scale grid.
[0010] Preferably, step S1 includes: first, dividing the exploration area of the working face into square grids according to its size; second, dividing each square grid into four subdivisions; then, determining the grids that need to be subdivided again based on the ray intercepts within the subdivided grids, and iterating in this way until the minimum scale limit is reached.
[0011] Preferably, in step S2, the multi-scale reconstruction objective function is expressed as follows:
[0012]
[0013] In the formula, A is the total number of rays, B is the number of grids, and X = (x1, x2, ..., x...). j ,···,x B H is the vector of absorption coefficients to be inverted from B grids. i Let r be the electromagnetic field strength data collected for the i-th ray. i Let d be the length of the i-th ray, H0 be the initial field strength value, and d be the length of the ray. ij Let be the intercept of the i-th ray in the j-th grid.
[0014] Preferably, the solution is obtained using a dynamic multi-population genetic algorithm, including:
[0015] S21. Initialize the population. Each population independently performs selection, crossover, and mutation operations to explore different regions of the solution space and record the elite individuals of all populations.
[0016] S22. By periodically introducing the best individuals from various populations into the target population through migration operators, co-evolution among populations is achieved.
[0017] S23. Set a fixed evaluation period, calculate the average fitness of each subpopulation, sort the subpopulations according to fitness, and merge the sorted subpopulations and their corresponding elite individuals to obtain a superpopulation.
[0018] S24. Using smaller crossover and mutation probabilities, perform a fine search on the superpopulation until the preset maximum number of iterations is reached.
[0019] Preferably, different control parameters are used for each population.
[0020] Therefore, the coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment described above has the following technical effects:
[0021] (1) This invention effectively integrates the grid through which the unobserved rays pass and reconstructs the search space for effective pruning, making it applicable to different working face exploration scales and ray distribution characteristics.
[0022] (2) Based on the dynamic multi-population genetic algorithm, this invention significantly improves the accuracy, stability and efficiency of the reconstruction results by dynamically adjusting the population size.
[0023] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0024] Figure 1 This is a numerical model of the CT exploration system for a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment.
[0025] Figure 2 This is a flowchart of a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment;
[0026] Figure 3 This is a schematic diagram of traditional grid division in an embodiment of a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment;
[0027] Figure 4 This is a distribution map of the number of rays at different scales in a traditional grid division of a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment.
[0028] Figure 5 This is a stability analysis diagram of reconstruction operations at different scales using traditional mesh generation in an embodiment of a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment. Detailed Implementation
[0029] The present invention will be explained in more detail through the following embodiments. The purpose of disclosing the present invention is to protect all changes and modifications within the scope of the present invention. The present invention is not limited to the following embodiments.
[0030] like Figure 1 As shown, the numerical model of the CT exploration and observation system for the working face has a length L of 200m and a width W of 100m. Two common geological anomaly areas exist in this numerical model, with an electromagnetic wave absorption coefficient β1 = 1.0dB / m in the anomaly areas and β2 = 0.5dB / m in the other normal areas. In the observation system layout, the spacing between transmitting points is 30m, and the spacing between corresponding receiving points is 10m, for a total of 420 rays (GS = 420).
[0031] like Figure 2 As shown, this invention provides a coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment, including the following steps:
[0032] S1. Due to the complex environment and electromagnetic wave propagation characteristics of coal mine working faces, electromagnetic wave CT projection data often exhibits incompleteness, leading to errors and uncertainties in the CT reconstruction results of the working face. Furthermore, mesh generation of the working face reconstruction region is a crucial step in the reconstruction calculation, and the mesh size directly affects the efficiency and accuracy of the inversion operation. For example... Figure 3 As shown, before the working surface reconstruction calculation, the target detection area needs to be discretized into a grid, divided into B square grids. Each grid is called a pixel and is represented by x. j This represents the true absorption coefficient of the j-th grid, where j = 1, 2, 3, ..., B.
[0033] Currently, uniform grid-scale meshing is commonly used in CT reconstruction calculations for working faces. However, this method often results in dense meshing to match the dense ray distribution, leading to computationally expensive calculations. Furthermore, many mesh regions are not traversed by observed rays, which increases the stability and speed of the solution. Simply increasing the mesh spacing and reducing the number of meshes will reduce the inversion resolution. Specifically, in a CT exploration area of 200m*100m, using the traditional uniform grid-scale meshing method, the uniform meshing scales were set to 25m, 20m, 12.5m, 10m, 6.25m, 3.125m, 2m, and 1.5625m. It was found that as the uniform meshing scale decreases, the number of meshes not traversed by observed rays increases almost exponentially. Figure 4 As shown. These grids that were not traversed by the ray do not carry any valid information about the projection data and still need to participate in the reconstruction calculation process. This not only reduces the efficiency of reconstruction but also increases the difficulty of searching for the optimal solution.
[0034] To reduce the number of unobserved rays passing through the grid, invalid grids need to be integrated. This involves reducing the grid scale in areas with dense ray coverage and appropriately increasing the grid scale in areas with low ray coverage, ensuring that every grid carries ray information and valuable observation data. This embodiment employs a multi-scale adaptive grid partitioning method. It effectively prunes the target search space through the working face reconstruction to finely match the ray distribution characteristics of different working face exploration sizes and densities. This fully utilizes the valuable information in the incomplete projection data of the working face CT exploration observation system, reducing the number of invalid grids and improving the efficiency and accuracy of reconstruction, as follows:
[0035] First, based on the CT scan dimensions of the working face (200m x 100m), the greatest common divisor is determined to be 100m. This ensures the side length of the largest square to be divided, and the number of squares is determined. Ultimately, the working face is divided into two 100m x 100m squares. The specific operation is as follows: starting from one corner of the rectangle, squares are sequentially divided within the rectangle according to the calculated side lengths. After each division, the boundaries of the remaining rectangles are updated, and this process is repeated until the entire rectangle is completely divided.
[0036] Secondly, based on the intercept distribution of the observed rays in the grid, each of the two squares is subjected to an iterative quadrisection operation, and the sum of the intercepts passed through by all observed rays in each square grid after quadrisection is calculated. If there is a case where the intercept is 0 in any of the four grids after quadrisection, that is, no observed ray passes through the grid, then the grid will not be further processed by subsequent smaller-scale quadrisections.
[0037] Then, based on the size of the on-site exploration of the working face, and taking into full account the efficiency and accuracy requirements of the reconstruction operation, the minimum scale of the four-segmentation is restricted as an iteration stopping condition, resulting in the multi-scale mesh division result of the working face.
[0038] To verify the impact of the minimum resolution of multi-scale partitioning on reconstruction performance, three minimum partitioning scales were set: 6.25m, 3.125m, and 1.5625m. As shown in Table 1, multi-scale partitioning according to the above rules effectively reduced the number of grids. In particular, when the minimum scale was 1.5625m, the number of grids decreased from 8192 after uniform-scale partitioning to 1790, achieving effective pruning of the search space for reconstruction of the working surface. This laid a good foundation for subsequent reconstruction operations.
[0039] Table 1. Number of grids at different minimum scales
[0040]
[0041] S2. In the CT reconstruction calculation of the working face, the core problem lies in solving an ill-conditioned matrix equation. After multi-scale meshing of the working face reconstruction, the problem of solving the ill-conditioned matrix is transformed into a functional extremum problem. A multi-scale reconstruction objective function is constructed, and then a dynamic multi-population genetic algorithm with strong global search capabilities is used to solve the objective function.
[0042] The objective function is defined as follows:
[0043]
[0044] In the formula, A is the total number of rays, B is the number of grids, and X = (x1, x2, ..., x...). j ,···,x B H is the vector of absorption coefficients to be inverted from B grids. i Let r be the electromagnetic field strength data collected for the i-th ray. i Let d be the length of the i-th ray, H0 be the initial field strength value, and d be the length of the ray. ij Let be the intercept of the i-th ray in the j-th grid.
[0045] The geological anomaly tomography inversion model can be defined as: finding X'∈C such that f(X')=minf(X); where X'=(x'1,x'2,···,x' j ,···,x' B ) is the absorption coefficient value obtained for each grid when the objective function f(X) reaches its minimum value.
[0046] In this embodiment, the Dynamic Multipopulation Genetic Algorithm (DMPGA) is used to optimize the working face reconstruction, specifically including:
[0047] S21. Initialize the population, including the number of populations MP = 10, the number of individuals in each population NIND = 20, the number of bits in the variables PRECI = 20, and the generation gap GGAP = 0.95; each population independently performs selection, crossover, and mutation operations, and each population has different control parameters, such as the crossover probability P. c ∈[0.75,0.95], mutation probability P m ∈[0.001,0.05], maximum number of iterations MAXGEN=1500, explore different regions of the solution space, and record the elite individuals of all populations.
[0048] S22. By periodically introducing the best individuals from various populations into the target population through migration operators, information exchange is achieved, enabling co-evolution among populations.
[0049] S23. Population Elimination and Dynamic Adjustment: A fixed evaluation period is set, the average fitness of each subpopulation is calculated, and individuals from the lowest-ranked subpopulations are directly removed based on fitness. When the population size decreases to a certain threshold, all remaining subpopulations and recorded elite individuals are merged into a superpopulation with higher diversity and better gene combinations.
[0050] S24. By utilizing smaller crossover and mutation probabilities, a finer search is performed on the superpopulation to further improve the quality and accuracy of the solution. While maintaining the stability of the population, it can also accelerate convergence until the preset maximum number of iterations is reached. The reconstruction operation ends, and the optimal solution of the working surface reconstruction objective function and the electromagnetic wave absorption coefficient corresponding to the multi-scale grid of the working surface are output.
[0051] By employing the above strategies, the population size can be gradually reduced while maintaining population diversity and avoiding premature convergence, ultimately forming a single population containing the optimal individual. This strategy helps improve the search efficiency and optimization performance of genetic algorithms.
[0052] To verify the effectiveness of the method in this embodiment, two sets of comparative experiments were conducted. First, based on the Multi-Population Genetic Algorithm (MPGA), the impact of uniform and multi-scale mesh generation on the working face reconstruction operation was verified. Second, based on MPGA, the impact of dynamic adjustment of multiple populations (i.e., DMPGA) on the efficiency and accuracy of working face reconstruction was verified.
[0053] (1) When generating uniform-scale meshes, the stability of the working surface reconstruction operation at different scales was compared (as shown in Table 2), and the correspondence between the final average convergence result and the computational efficiency was also compared (as shown in Table 2). Figure 5As can be seen, when the uniform mesh size of the working face decreases from 25m to 6.25m, after 1500 generations of reconstruction iteration search evolution, the convergence result shows a significant decrease, from 2.7925dB / m to 0.5210dB / m. However, when the uniform mesh size of the working face continues to decrease from 6.25m to 1.5625m, the average convergence result actually increases from 0.5210dB / m to 5.0449dB / m. Simultaneously, as the mesh size decreases from 25m to 1.5625m, the maximum change in the search results from multiple repeated reconstructions increases from 0.0953dB / m to 5.6625dB / m, and the average computation time also increases from 17.0324s to 4642.7151s. This indicates that as the mesh size of the working face reconstruction decreases, the stability and efficiency of the reconstruction results gradually decrease. Therefore, under the condition of incomplete projection of the working face, the reasonable design of the reconstruction mesh size has an important impact on the performance of the reconstruction operation.
[0054] Table 2 Results of uniform-scale mesh generation and working face reconstruction calculations
[0055]
[0056] In addition, by Figure 4 It can be seen that when the uniform mesh division scale is reduced to 6.25m, a large number of meshes without effective projection information appear that are not passed through by the observed rays. Based on this, when using multi-scale mesh division, this embodiment uses 6.25m, 3.125m, and 1.5625m as the minimum mesh division scale, and performs multiple working surface reconstruction operations. The corresponding mesh subdivision numbers are 446, 1130, and 1790, respectively. The convergence results and reconstruction efficiency are shown in Table 3.
[0057] Table 3 Results of Multi-Scale Mesh Generation and Surface Reconstruction Calculation
[0058]
[0059] It can be seen that when the minimum grid scale is 6.25m, 3.125m, and 1.5625m, the average convergence results decrease from 0.5210dB / m, 0.9265dB / m, and 5.0449dB / m at the uniform scale to 0.3479dB / m, 0.8286dB / m, and 1.1425dB / m, respectively; the change in reconstruction results increases from 0.63... The accuracy, stability, and efficiency of the reconstruction results were significantly improved by effectively integrating the meshes through which the unobserved rays passed and effectively pruning the reconstruction search space. The values of 97 dB / m, 0.8070 dB / m, and 5.6625 dB / m were reduced to 0.4146 dB / m, 0.7058 dB / m, and 0.9681 dB / m, respectively. The reconstruction computation time was also reduced from 221.4356 s, 936.8193 s, and 4668.2367 s to 194.0055 s, 493.1762 s, and 833.2183 s, respectively. This demonstrates that by effectively integrating the meshes through which the unobserved rays passed and effectively pruning the reconstruction search space, the accuracy, stability, and efficiency of the reconstruction results were significantly improved.
[0060] (2) Based on MPGA, we set an interval of 100 generations to evaluate all populations and perform reconstruction operations according to the multi-population dynamic adjustment strategy. The results are shown in Table 4.
[0061] Table 4. Results of working surface reconstruction calculation based on DMPGA multi-scale mesh generation
[0062]
[0063] It can be seen that the average convergence value and change amplitude obtained by the DMPGA multi-scale reconstruction model are not much different from those of the MPGA algorithm, but the computation time is reduced from 194.0055s, 493.1762s and 833.2183s to 85.3674s, 219.434s and 351.7312s respectively, and its reconstruction efficiency is further greatly improved.
[0064] Therefore, the coal mine working face reconstruction method based on search space pruning and multi-group dynamic adjustment adopted in this invention can finely match different working face exploration sizes and different density ray distribution characteristics, realize multi-scale grid division, and improve the accuracy, stability and efficiency of working face reconstruction.
[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment, characterized in that, Includes the following steps: S1. The exploration area of the working face is initially divided into grids, and multi-scale grids are generated by pruning the search space based on the sum of the ray intercepts in each grid. Step S1 includes: First, the exploration area of the working face is initially divided into square grids according to its size; second, each initially divided square grid is iteratively divided into four segments; then, based on the ray intercepts in the four-segmented grids, the grids that need to be further divided are determined. If there is a grid with an intercept of 0 after four-segmentation, that is, no observed ray passes through the grid, then the grid will not be further divided into smaller scales; finally, based on the size of the on-site exploration of the working face and taking into full consideration the efficiency and accuracy requirements of the reconstruction operation, the minimum scale of the four-segmentation is restricted as an iteration stopping condition, thus forming the multi-scale grid division result of the working face. S2. Based on the partitioned multi-scale grid, a multi-scale reconstruction objective function is constructed, and a super population is obtained by eliminating and dynamically adjusting the population through a dynamic multi-population genetic algorithm. Then, a fine search is performed on the super population, and the multi-scale reconstruction objective function is iteratively solved to obtain the electromagnetic wave absorption coefficient corresponding to the multi-scale grid. Population elimination and dynamic adjustment are performed using dynamic multi-population genetic algorithms, including: S21. Initialize the population. Each population independently performs selection, crossover, and mutation operations to explore different regions of the solution space and record the elite individuals of all populations. S22. By periodically introducing the best individuals from various populations into the target population through migration operators, co-evolution among populations is achieved. S23. Set a fixed evaluation period, calculate the average fitness of each subpopulation, sort the subpopulations according to fitness, and merge the sorted subpopulations and their corresponding elite individuals to obtain a superpopulation. S24. Using smaller crossover and mutation probabilities, perform a fine search on the superpopulation until the preset maximum number of iterations is reached.
2. The coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment according to claim 1, characterized in that, In step S2, the multi-scale reconstruction objective function is expressed as follows: ； In the formula, The total number of rays, For the number of grid cells, for The absorption coefficient vector of each grid to be inverted For the first Electromagnetic field strength data collected by a single ray. For the first The length of the ray, This represents the initial field strength value at launch. For the first The first ray The intercept of each grid.
3. The coal mine working face reconstruction method based on search space pruning and multi-population dynamic adjustment according to claim 1, characterized in that, Different control parameters are used for each population.