An intelligent top coal caving method for fully mechanized caving working face based on transparent geology

Abstract

The present invention discloses an intelligent top coal caving method for a fully mechanized top coal caving working face based on transparent geology: S1. Based on the coordinates of the hydraulic support, the elevation of the roof above the hydraulic support, the elevation of the support top, and the molar coefficient of the top coal are extracted by searching a three-dimensional geological model database; S2. The hydraulic support tail beam is manually controlled to caving coal, and the corresponding caving time of the hydraulic support tail beam under different top coal thicknesses and the number of times the tail beam shakes the coal under different top coal molar coefficients are tested; S3. A trained model is obtained using the DBN-PSO-MSVR model, which serves as a relationship model between top coal thickness, top coal molar coefficient, caving time, and the number of times the tail beam shakes the coal; S4. Based on the relationship model, the hydraulic supports in all working sections are caving coal in sections according to the total amount of top coal. The present invention can reduce the number of top coal caving workers in the working face, reduce the labor intensity of the caving workers, and reduce the differences in coal caving effects caused by human factors. At the same time, it improves the efficiency of the tail beam in shattering the top coal and increases the coal output.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal mining, and relates to an intelligent top coal caving method for a fully mechanized caving working face based on transparent geology. Background Art

[0002] In order to achieve efficient mining of thick coal seams, coal mining enterprises currently generally adopt the top coal caving technology for coal mining, that is, a fully mechanized caving working face is arranged at the bottom of the thick coal seam, the bottom coal is crushed and mined by a coal mining machine, and the top coal is mined by controlling the free collapse of the top coal through hydraulic supports, so that the full height can be mined in one time. This is conducive to reducing the amount of tunnel engineering in the mining area, reducing the investment in working face equipment and reducing coal mining costs.

[0003] The implementation of the top coal caving process requires not only suitable hydraulic supports, reasonable mining and caving ratios, and caving step distances, but also the mastery of the rules of top coal caving and the formulation of reasonable top coal caving process procedures and safety measures to ensure high productivity and efficiency of the working face and construction safety.

[0004] The core issue of top coal caving is caving more coal and less gangue. Reference (1) studied key technologies of top coal caving, such as accurate detection of top coal thickness by ground penetrating radar, identification of gangue during caving by vibration frequency, and precise control of the caving mechanism by installing a travel sensor. Reference (2) addresses the problem of gangue identification during caving and constructs a "three-in-one" intelligent identification system for gangue inclusions through image processing. This system accurately identifies gangue inclusions that may occur during caving, thereby improving the intelligence level of the caving process. Reference (3) divides the coal caving process into three stages: before caving, during caving, and after caving. Different sensing technologies and equipment are used. Before caving, a ground-penetrating radar is installed at the front of the support top beam to measure the top coal thickness as the benchmark for the caving amount. During caving, a three-dimensional radar is installed at the junction of the support top beam and the cover beam to scan the space where the top coal is not caving. The remaining top coal volume is measured and compared with the caving amount benchmark before caving to determine when the caving process ends. After caving, the volume of coal caving and the proportion of coal gangue on the rear scraper conveyor are identified to improve the robustness of coal gangue detection. Reference (4) controls the caving amount based on the principle of high recovery rate and low gangue content. A memory caving time sequence control mode is adopted, and an intelligent decision-making mechanism is embedded to achieve scientific and continuous caving action. Reference (5) learns the coal caving control strategy from the perspective of the overall working face, and takes the control of the coal-rock interface morphology as the goal of intelligent learning. It establishes a top coal caving intelligent decision model based on the mean deviation reward function Q-learning to improve the coordination and synergy of the actions between the coal caving ports during the coal caving process of the fully-mechanized working face, so that it can better control the coal caving process and improve the coal caving effect of the working face. Reference (6) identifies the signal generated by the impact of coal gangue by installing a vibration sensor on the tail beam of the top coal caving hydraulic support, thereby improving the controllability of the coal gangue caving process. Reference (7) proposes an automatic memory coal caving control strategy and specifically designs the automatic memory coal caving control process, including the specific equipment used, control process and main control parameters, and points out the implementation method of automatic memory coal caving control. Reference (8) Based on the fact that the geological conditions of the actual working face are relatively stable, the production process of each coal cut does not change much, and top coal caving is a monotonous and repetitive task, the memory function is used to memorize the control process of the support by the coal caving operator, so that memory coal caving can be achieved in the subsequent coal caving process.

[0005] In summary, the existing top coal caving methods have the following problems: 1. When manually operating the tail beam of the support to caving coal, there are problems such as harsh working environment, large dust and gas, and large differences in individual coal caving; 2. When the tail beam of the intelligent control support is used to caving top coal, the top coal thickness detection precision is not high, large coal blocks cannot be caved, and the coal gangue morphology monitoring at the coal caving port has large errors.

[0006] The fully-mechanized caving (FPC) mining process involves mining thick and extra-thick coal seams (5-10 m) with a 2-4 m mining face along the seam floor or within a certain thickness range. Coal miners are used for advance mining. In the unmined seam above the shearer, the top coal is broken into loose pieces using mine pressure or assisted by loosening blasting. The loose coal is then released from a "coal window" behind or above the supports and transported out of the face by a face conveyor. A fully-mechanized caving face encompasses the stratigraphic information between the cut, return airway, haulage, and main return airway, as well as the equipment and its distribution, operating conditions, and environmental factors. A fully-mechanized caving face based on transparent geology involves collecting stratigraphic information between the cut, return airway, haulage, and main return airway through geophysical exploration, drilling, and geological realism. This information is then presented in a three-dimensional model using computer modeling and other techniques. Equipment and operating condition information is then presented in a scene-integrated manner using computer technologies such as data twins. Stratum information includes top coal thickness, roof and floor, roadway slope, geological structure and other information. Comprehensive caving equipment includes coal mining machine, scraper conveyor, hydraulic support, transfer machine, crusher, belt conveyor, advanced hydraulic support, water pump, mud pump, combination switch, electro-hydraulic control system, etc., including the operating position and posture of coal mining machine, scraper conveyor and hydraulic support (9). Reference (10) proposed a three-dimensional coal seam modeling method based on transparent geology for comprehensive caving working face, aiming at the low prediction accuracy of the roof and floor elevation of coal seams under complex geological conditions, which is difficult to meet the actual needs of coal mining. Based on the geological data of the air intake and return tunnels, drilling measurement data, working face cutting data and coal seam geological data obtained by three-dimensional seismic reinterpretation technology, channel wave seismic exploration technology and radio electromagnetic wave perspective technology, the discrete smooth interpolation (DSI) algorithm is applied to predict the elevation of the coal seam roof and floor, and a static three-dimensional coal seam model of the comprehensive caving working face is constructed. By dynamically updating the newly revealed geological information through eye-cutting and the DSI algorithm, a more accurate dynamic three-dimensional coal seam model of the working face is obtained, guiding the coal mining machine to perform automatic height control, thereby achieving adaptive coal cutting.

[0007] Transparent working faces are currently an important direction for the advancement of intelligent coal mines. They can provide information such as top coal thickness, top coal molar coefficient and top coal gas content for the top coal caving process. However, there is currently no specific implementation method for how to use transparent working face information to optimize the top coal caving process.

[0008] References:

[0009] [1] Zhang Xueliang, Liu Qing, Lang Ruifeng, et al. Research on key technologies of intelligent coal caving process and precise control in thick coal seams [J]. Coal Engineering, 2020, 52(9): 1-6.

[0010] [2] Wang Jiachen, Pan Weidong, Zhang Guoying, Yang Shengli, Yang Kehu, Li Lianghui. Principle and application of image recognition intelligent coal caving technology[J]. Journal of China Coal Society, 2022, 47(01): 87-101.

[0011] [3] Zhang Shouxiang, Zhang Xueliang, Liu Shuai, Xu Guoqing. Precise coal caving control technology for intelligent top coal caving mining[J]. Journal of China Coal Society, 2020, 45(06): 2008-2020.

[0012] [4] Ma Ying. Research on intelligent coal caving mode based on memory coal caving timing control[J]. Coal Mine Electromechanical, 2015(02):1-5.

[0013] [5] Luo Kaicheng, Gao Yang, Yang Yi, Chang Yajun, Yuan Ruifu. Research on coal caving control strategy based on mean deviation reward function[J]. Coal Engineering, 2022, 54(09): 105-111.

[0014] [6] Ma Ying. Research on intelligent coal caving method for gangue identification based on tail beam vibration signal acquisition [J]. Coal Mining, 2016, 21(04): 40-42+25.

[0015] [7] Wang Qixin. Precision coal caving control technology for intelligent top coal caving mining[J]. Contemporary Chemical Industry Research, 2021(14):67-68. [8] Xu Dongfei. Research on key technologies for intelligent mining of fully mechanized top coal caving working face[J]. Coal, 2019, 28(04):76-77.

[0016] [9] Lei Xiaorong, Li Mingxing, Yue Hui, An Lin. Key technologies and implementation of digital twin system for transparent working face[J]. Intelligent Mine, 2022, 3(07): 50-56.

[0017]

[10] Xue Guohua. Three-dimensional coal seam modeling of fully mechanized mining working face based on transparent geology[J]. Industrial and Mining Automation, 2022, 48(04): 135-141. Summary of the Invention

[0018] The purpose of the present invention is to provide an intelligent top coal caving method based on a transparent working surface to solve the following technical problems existing in the prior art:

[0019] (1) When manually operating the tail beam of the support to place coal, the support worker needs to observe the coal discharge and equipment operation status, operate the hydraulic support controller or solenoid valve to open the coal opening for top coal placement, which has problems such as harsh environment and high labor intensity. In addition, different operating habits will affect the coal placement effect.

[0020] (2) During the top coal caving process, when there is gangue in the top coal, the gangue on the scraper conveyor is easy to be misjudged by the support workers, who will think that the top coal has been caved and rock has been exposed and stop caving, resulting in coal caving losses;

[0021] (3) During the top coal caving process, the top coal may not be pre-cracked or pre-cracked insufficiently. At this time, the top coal will not fall down when the coal opening is opened. The scaffolding worker needs to operate the tail beam to shake the coal blocks to caving the coal. If the shaking time is too long, the efficiency of the top coal caving will be affected. If the shaking time is too short, the effect of the top coal caving will be affected.

[0022] (4) During the top coal caving process, there is a possibility that the amount of coal caving is too large in a short period of time or the gas content in the caving top coal is too high due to improper control of the coal caving amount, causing the gas content in the working face to exceed the standard and endangering the safety of the personnel on the working face.

[0023] In order to achieve the above object, the present invention adopts the following technical solutions:

[0024] An intelligent top coal caving method for a fully mechanized top coal caving working face based on transparent geology specifically comprises the following steps:

[0025] S1. Based on the XY coordinate position of the hydraulic support in the fully mechanized caving working face equipment, the data of the roof elevation above the hydraulic support, the support top elevation, the top coal molar coefficient, etc. are extracted by searching the data in the 3D geological model database, and the top coal thickness is calculated as follows: roof elevation above the hydraulic support - support top elevation;

[0026] S2, manually control the hydraulic support tail beam to release coal, test the corresponding coal release time of the hydraulic support tail beam under different top coal thicknesses, and the number of coal breaking vibrations of the support tail beam under different top coal molar coefficients, and obtain the data D = {(X i ,Y i )}, where X i is an m×2 matrix consisting of top coal thickness and top coal molar coefficient, Y i The m×2 matrix is ​​composed of the coal discharge time and the number of tail beam coal shaking; the data D is divided into the training set D train and the test set D test ; m represents the number of hydraulic supports tested;

[0027] S3, the training set D train The top coal thickness and top coal molar coefficient in the training set D are used as model input. train The coal caving time and tail coal shaking times in the model are used as model outputs, and the DBN-PSO-MSVR model is used for training to obtain a trained model as the relationship model between top coal thickness, top coal molar coefficient, coal caving time, and tail coal shaking times;

[0028] S4. Based on the relationship model established in S3, the coal discharge time and the number of shakes are predicted according to the top coal thickness and the top coal molar coefficient for the working face in which coal is being discharged. In accordance with the principles of not affecting the safe support of the support, the coal discharge volume not exceeding the carrying capacity of the scraper conveyor, and the gas content near the support not exceeding 0.3-0.9%, the hydraulic supports of the entire working section are discharged in sections according to the total amount of top coal. In each section, one coal discharge port is kept open for coal discharge in the order of the hydraulic supports.

[0029] Furthermore, S3 specifically includes the following sub-steps:

[0030] S31, the training set D train The top coal thickness and top coal molar coefficient in the DBN network are input for unsupervised pre-training. The training set D train The coal discharge time and tail coal jitter times are taken as output to obtain the trained DBN network;

[0031] S32, the training set D train The trained DBN network is then retrained to achieve supervised parameter fine-tuning. After the training is completed, the parameters of each layer of the DBN network are determined, and the output of the third RBM in the DBN network is used as the training data set;

[0032] S33, input the training data set obtained in S32 into the PSO-MSVR model for training. In this process, PSO is used to optimize the regularization parameter C and kernel function parameter σ in MSVR, initialize PSO to determine the maximum number of iterations K, the population size M, the regularization constant C∈[a,b], the kernel function parameter σ∈[c,d], the c1 and c2 group cognitive coefficients, initialize the positions and velocities of M particles, iteratively update the population fitness, and obtain the optimal parameters (C,σ); substitute the optimized regularization parameter C and kernel function parameter σ into MSVR to obtain the trained DBN-PSO-MSVR model as the relationship model.

[0033] Furthermore, in step S32, the first RBM is first trained to obtain the weights and biases of the visible layer and hidden layer, the state of the neurons in the hidden layer of the first RBM is used as the input vector of the second RBM, and the weights and biases of the second RBM are trained again until the training of the three RBMs is completed, and then the error is calculated through the last BP layer for reverse fine-tuning.

[0034] Furthermore, in step S33, the kernel function parameters adopt Gaussian kernel function.

[0035] Furthermore, in step S4, the top coal caving is divided into multiple rounds according to the thickness of the segmented top coal, and the top coal caving time of the corresponding support is evenly distributed to the multiple rounds of top coal caving; the corresponding tail beam coal crushing shaking times are evenly distributed to the multiple rounds of top coal caving.

[0036] The design concept of this invention is as follows: through extensive experiments, data on caving time and number of oscillations for different top-coal thicknesses and molar coefficients are obtained. A machine learning approach is then used to establish a coupling model between the four factors, studying the influence of these factors on caving time and number of oscillations. This is then achieved by remotely controlling the top-coal caving of a hydraulic support using a computer, achieving automatic control of both caving time and number of oscillations. This reduces the number of caving support workers on the working face, lowering their workload and minimizing the variability in caving results caused by human factors. Furthermore, this improves the efficiency of the tail beam in shattering the top coal, thereby increasing coal output.

[0037] Compared with the prior art, the present invention has the following beneficial technical effects:

[0038] (1) The present invention realizes the top coal caving by remotely controlling the hydraulic support through the upper computer, thereby reducing the number of support workers for top coal caving on the working face, reducing the labor intensity of the support workers, and reducing the differences in coal caving effects caused by human factors.

[0039] (2) The present invention uses a three-dimensional geological model to determine the top coal thickness according to the coordinate position of the coal caving port. Combined with the test coal caving time, the top coal thickness is strongly correlated with the coal caving time, and the coal caving time of each coal caving port is predicted to ensure that the top coal can be caved as much as possible, thereby improving the top coal recovery rate.

[0040] (3) The present invention predicts the optimal solution for the number of rear tail beam vibrations by learning the molar coefficient of the top coal and the support pressure data of the hydraulic support, and controls the rear tail beam to crush the top coal with the highest efficiency.

[0041] (4) The gas content of the top coal is searched in the gas model based on the caving port coordinates, and the gas release amount during the caving process is predicted. Under the principle of ensuring that the gas content in the working face does not exceed the standard, the top coal caving is carried out with maximum efficiency. This not only ensures that there are no gas hazards in the working face, but also achieves efficient top coal caving. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the top coal caving method of a fully mechanized caving working face based on transparent geology of the present invention;

[0043] Figure 2 This is a schematic diagram of top coal caving action;

[0044] Figure 3 This is a schematic diagram of the open / closed state of the coal vent at the working face;

[0045] Figure 4 This is a schematic diagram of the DBN-PSO-MSVR structure.

[0046] Figure 5 This is the structural diagram of the DBN network. DETAILED DESCRIPTION

[0047] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] The present invention provides an intelligent top coal caving method for a fully mechanized top coal caving working face based on transparent geology, which specifically includes the following steps:

[0049] S1. Based on the XY coordinate position of the hydraulic support in the fully mechanized caving working face equipment, the data of the roof elevation above the hydraulic support, the support top elevation, the top coal molar coefficient, etc. are extracted by searching the data in the 3D geological model database, and the top coal thickness is calculated as follows: roof elevation above the hydraulic support - support top elevation;

[0050] S2, manually control the hydraulic support tail beam to release coal, test the corresponding coal release time of the hydraulic support tail beam under different top coal thicknesses, and the number of coal breaking vibrations of the support tail beam under different top coal molar coefficients, and obtain the data D = {(X i ,Y i )}, where X i is an m×2 matrix consisting of top coal thickness and top coal molar coefficient, Y i The m×2 matrix is ​​composed of the coal discharge time and the number of tail beam coal shaking; the data D is divided into training set D according to the ratio of 4:1. train and the test set D test ; m represents the number of hydraulic supports tested, which is usually an integer multiple of 12.

[0051] S3, the training set D train The top coal thickness and top coal molar coefficient in the training set D are used as model input. train The coal-laying time and the number of tail coal shaking times in the model are used as the model outputs, and the DBN-PSO-MSVR model (such as the one that combines the Deep Belief Network (DBN), the Particle Swarm Optimization (PSO) and the Multi-output Support Vector Regression (MSVR)) is adopted. Figure 4 As shown in the figure, the relationship model between top coal thickness, top coal molar coefficient, coal caving time, and tail coal shaking times is as follows:

[0052] S31, the training set D train The top coal thickness and top coal molar coefficient in the DBN network are input for unsupervised pre-training. The training set D train The coal discharge time and tail coal jitter times are taken as output to obtain the trained DBN network;

[0053] Specifically, the DBN network consists of three restricted Boltzmann machines (RBMs) and a BP layer, such as Figure 5 As shown in the figure, during the pre-training process of the DBN network, the first RBM is trained to obtain the weights and biases of the visible and hidden layers. The state of the neurons in the hidden layer of the first RBM is used as the input vector of the second RBM. The weights and biases of the second RBM are then trained until the training of the three RBMs is completed. The error is then calculated through the last BP layer for reverse fine-tuning.

[0054] S32, the training set D train The trained DBN network is then retrained to achieve supervised parameter fine-tuning. After the training is completed, the parameters of each layer of the DBN network are determined, and the output of the third RBM in the DBN network is used as the training data set;

[0055] S33, input the training data set obtained in S32 into the PSO-MSVR model for training. In this process, PSO is used to optimize the regularization parameter C and kernel function parameter σ in MSVR. The kernel function parameter here adopts Gaussian kernel function. Initialize PSO to determine the maximum number of iterations K, the population size M, the regularization constant C∈[a,b], the kernel function parameter σ∈[c,d], the group cognition coefficients c1 and c2, initialize the positions and velocities of M particles, iteratively update the population fitness, and obtain the optimal parameters (C,σ); substitute the optimized regularization parameter C and kernel function parameter σ into MSVR to obtain the trained DBN-PSO-MSVR model as the relationship model.

[0056] S4. Based on the relationship model established in S3, the caving time and number of shakes are predicted for the working face undergoing caving based on the top coal thickness and the top coal molar coefficient. Based on the principles of not affecting the safe support of the supports, ensuring that the caving volume does not exceed the carrying capacity of the scraper conveyor, and that the gas content near the supports does not exceed 0.3-0.9%, the hydraulic supports in the entire working section are caving in sections according to the total amount of top coal. Each section maintains a caving opening open for caving according to the order of the hydraulic supports. Preferably, to maintain a gentle descent of the coal-rock interface, the caving process can be divided into multiple rounds according to the thickness of the segmented top coal. The caving time of the corresponding supports is evenly distributed over the multiple rounds; the corresponding number of shakes of the tail beam coal crusher is also evenly distributed over the multiple rounds.

[0057] Example:

[0058] Following the transparent geology-based top coal caving method of the fully mechanized caving working face of the present invention, the process of this embodiment is described in combination with the actual situation of the 4-2302 transparent working face of Jianzhuang Mining:

[0059] S1. Based on the XY coordinate positions of the hydraulic supports in the fully mechanized caving working face, data was retrieved from the 3D geological model database to extract data such as the roof elevation, support top elevation, and top coal molar coefficient for a section of the seam roof and floor in the 3D fine model of the 4-2302 working face. (Typically, the supports at both ends of the working face are less likely to be used, so data from the center of the working face is more accurate and effective.) The top coal thickness was calculated by comparing the roof elevations and hydraulic support top elevations for Nos. 11 to 22 shown in Table 1.

[0060] Top coal thickness = roof elevation above hydraulic support Z1 - support top elevation Z2

[0061] Table 1

[0062]

[0063] S2, such as Figure 3 As shown, the support workers sequentially opened the caving openings according to the hydraulic support number for top coal caving, implementing single-support caving. The working face supports were divided into 12 sections for simultaneous caving. To maintain a balanced caving time for each section, the sections were evenly divided according to the total amount of top coal. For example, if the total amount of top coal was 12 million tons, and each section averaged approximately 1 million tons, but the top coal distribution was uneven, based on the top coal data, top coal could be accumulated starting from the 11th support until the 22nd support had accumulated approximately 1 million tons. Sections 11-22 would then be designated as the first caving section, and so on. The caving times for different top coal thicknesses for hydraulic supports Nos. 11-22 were tested and recorded. As shown in Table 2, the more data recorded, the greater the diversity of the recorded data, the higher the data sample quality, and the more accurate the established model.

[0064] Table 2

[0065]

[0066] The number of times the tail beam vibrates and breaks coal under the conditions of the top coal molar coefficient of different support numbers was tested and recorded, as shown in Table 3. The more data recorded, the more significant the diversity of the recorded data, the higher the quality of the data sample, and the more accurate the established model.

[0067] Table 3

[0068]

[0069] S3. Model the data of the four parameters of top coal thickness, coal caving time, top coal molar coefficient, and tail beam shaking and broken coal in Table 2 and Table 3. The top coal thickness and top coal molar coefficient are used as model inputs, and the coal caving time and tail beam shaking times are used as outputs. The DBN-PSO-MSVR fusion method is used to establish a relationship model between top coal thickness, top coal molar coefficient, coal caving time, and tail beam shaking times.

[0070] S4. Based on the relationship model established in S3, the coal placement time and tail coal shaking times of different top coal thicknesses of No. 23-34 hydraulic supports are predicted. The prediction results are shown in Tables 4 and 5.

[0071] Each section can only have one coal discharge port for coal discharge, and the coal is discharged in a cyclic manner within the section in sequence. If the top coal thickness is less than 1 meter, the coal discharge time is evenly distributed according to two rounds for each bracket, and the number of times the tail beam shakes and breaks the coal is distributed according to the two rounds; if the top coal thickness is greater than 2 meters, the coal is discharged in a round of one meter according to the thickness corresponding to the top coal thickness, and the coal discharge time for each round is evenly distributed according to the total coal discharge time of the bracket, and the number of tail beam shaking times is similarly distributed to each round of coal discharge.

[0072] The principle is to not affect the safe support of the support; the principle is that the coal discharge volume does not exceed the carrying capacity of the scraper conveyor; do not open two adjacent supports to discharge coal at the same time, to avoid the discharge of coal blocks that are too large and exceed the local carrying capacity of the scraper conveyor; the principle is that the gas concentration of the working face does not exceed 0.5%. According to the current average gas content of the coal seam in the working face, the gas increase when discharging coal around a single coal discharge port, the number of coal discharge ports at the same time on the working face should not exceed 12.

[0073] Table 4

[0074]

[0075]

[0076] Table 5

[0077] Bracket number Top coal molar coefficient Number of tail beam coal crushing vibrations (s) 23 3 6 24 3.1 6 25 3 6 26 3 6 27 2.9 5 28 3 6 29 3 5 30 3 6 31 3 6 32 3 5 33 2.3 4 34 2.3 3

[0078] The evaluation indicators used are Mean Absolute Error (MAE) and Mean Squared Error (MSE). The calculation formula is as follows:

[0079]

[0080]

[0081] Where Y is the actual value of coal discharge time and number of times the tail beam coal is shaken. The predicted values ​​for the coal caving time and the number of tail beam coal crushing vibrations are shown in Table 6. The prediction results of the DBN-PSO-MSVR model are shown in Table 6. As can be seen, the model has a maximum prediction error of 5.83% for the coal caving time and a maximum prediction error of 20% for the number of tail beam coal crushing vibrations, meeting the error limits required for practical engineering projects. For comparison and verification, a comparison with the least squares method and support vector machine (SVR) is shown in Table 7. Table 7 shows that the DBN-PSO-MSVR model has a smaller prediction error than the traditional least squares method and support vector machine (SVR) prediction regression methods.

[0082] Table 6

[0083]

[0084] Table 7

[0085]

Claims

1. An intelligent top coal caving method for fully mechanized caving working face based on transparent geology, characterized in that: The specific steps include: S1. Based on the XY coordinate position of the hydraulic support in the fully mechanized caving working face equipment, the roof elevation above the hydraulic support, the support top elevation, and the top coal molar coefficient are extracted by searching the data in the 3D geological model database, and the top coal thickness is calculated as follows: roof elevation above the hydraulic support - support top elevation; S2, manually control the hydraulic support tail beam to release coal, test the corresponding coal release time of the hydraulic support tail beam under different top coal thickness, and the number of coal breaking vibrations of the support tail beam under different top coal molar coefficients, and obtain data ,in X i is composed of top coal thickness and top coal molar coefficient The matrix, Y i It is composed of the coal discharge time and the number of times the tail beam coal is shaken. Matrix; divide the data D into training set and test set ; m Indicates the number of hydraulic supports tested; S3, the training set The top coal thickness and top coal molar coefficient in the training set are used as model input. The coal caving time and tail coal shaking times in the model are used as model outputs, and the DBN-PSO-MSVR model is used for training to obtain a trained model as the relationship model between top coal thickness, top coal molar coefficient, coal caving time, and tail coal shaking times; S4. Based on the relationship model established in S3, the coal discharge time and the number of shakes are predicted according to the top coal thickness and the top coal molar coefficient for the working face in which coal is being discharged. In accordance with the principles of not affecting the safe support of the support, the coal discharge volume not exceeding the carrying capacity of the scraper conveyor, and the gas content near the support not exceeding 0.3-0.9%, the hydraulic supports of the entire working section are discharged in sections according to the total amount of top coal. In each section, one coal discharge port is kept open for coal discharge in the order of the hydraulic supports.

2. The intelligent top coal caving method for fully mechanized caving working face based on transparent geology according to claim 1 is characterized in that: S3 specifically includes the following sub-steps: S31, the training set The top coal thickness and top coal molar coefficient in the training set are input into the DBN network for unsupervised pre-training. The coal discharge time and tail coal jitter times are taken as output to obtain the trained DBN network; S32, the training set The trained DBN network is then retrained to achieve supervised parameter fine-tuning. After the training is completed, the parameters of each layer of the DBN network are determined, and the output of the third RBM in the DBN network is used as the training data set; S33, input the training data set obtained in S32 into the PSO-MSVR model for training. In this process, PSO is used to optimize the regularization parameters in MSVR. and kernel function parameters , initialize PSO to determine the maximum number of iterations , population size , regularization constant , kernel function parameters , and Group cognition coefficient, initialization The particle positions and velocities are iteratively updated to obtain the optimal parameters. ; The optimized regularization parameter and kernel function parameters Substitute into MSVR to obtain the trained DBN-PSO-MSVR model as the relational model.

3. The intelligent top coal caving method for fully mechanized caving working face based on transparent geology according to claim 2, characterized in that: In step S32, the first RBM is first trained to obtain the weights and biases of the visible layer and hidden layer. The state of the neurons in the hidden layer of the first RBM is used as the input vector of the second RBM. The weights and biases of the second RBM are then trained until the training of the three RBMs is completed. The error is then calculated through the last BP layer for reverse fine-tuning.

4. The intelligent top coal caving method for fully mechanized caving working face based on transparent geology according to claim 2, characterized in that: In step S33, the kernel function parameters adopt Gaussian kernel function.

5. The intelligent top coal caving method for fully mechanized caving working face based on transparent geology according to claim 1, characterized in that: In step S4, the top coal caving is divided into multiple rounds according to the thickness of the segmented top coal, and the top coal caving time of the corresponding support is evenly distributed to the multiple rounds of top coal caving; the corresponding tail beam coal crushing shaking times are evenly distributed to the multiple rounds of top coal caving.

Images