Ground surface subsidence prediction system and method considering influence of extremely thick and weakly cemented sandstone

Abstract

The invention provides a surface subsidence prediction system and method considering the influence of extremely thick and weakly cemented sandstone, belongs to the technical field of deformation monitoring and subsidence control, and solves the problems that the relationship between surface subsidence rule characterization parameters and the extremely thick and weakly cemented sandstone is less concerned, and the response rule between the surface subsidence rule characterization parameters and the extremely thick and weakly cemented sandstone is not clear. The method comprises the following steps: firstly, carrying out surface subsidence rule characterization parameter inversion based on a BKA-BP neural network; then, a PCA method is adopted to analyze main control factors of the surface subsidence characterization parameters, a mathematical model of the surface subsidence characterization parameters and the main control factors is established, and the mathematical model is a mathematical model of the surface subsidence characterization parameters fused with independent variables and considering the triple-arch control effect; and finally, correcting the surface subsidence characterization parameters based on the mining degree, and combining the surface subsidence characterization parameters and the main control factor mathematical model established in the step 2 to obtain a bidirectional random medium priori model considering the triple-arch control effect, and predicting the mining surface subsidence of the working face of the trench coal mine by using the priori model.

Description

Technical Field

[0001] The present invention belongs to the field of geological engineering technology, and in particular relates to a surface subsidence prediction system and method taking into account the influence of extremely thick and weakly cemented sandstone. Background Art

[0002] The deep mining area of ​​Ordos Coalfield has rich coal resources, simple geological structure and sparse population, which is very suitable for high-intensity large-scale mining of deep coal resources. However, with the continuous increase in mining intensity, the Cretaceous and Jurassic thick sandstones in the eastern part of Ordos Coalfield have a great impact on large-scale continuous mining due to their special rock movement and energy accumulation properties. Based on this method of deep wide-band mining in some mining areas of Ordos Coalfield, active subsidence reduction and loss control technology is used to effectively slow down surface subsidence while ensuring coal recovery rate. However, due to the complex movement mechanism of the thick sandstone of Zhidan Group, its control role in wide-band mining has greatly limited it. Therefore, clarifying the position of thick sandstone in mining subsidence in the deep mining area of ​​Ordos Coalfield and establishing a surface subsidence prediction method that takes into account the control role of thick sandstone are of great significance for subsidence reduction and loss control and environmental protection.

[0003] Since it was proposed by Zhang Huaxing in 2000, wide strip mining has been continuously applied in engineering practice and achieved good results. Daizhuang Coal Mine of Zibo Mining Group uses wide strip mining technology when mining under village groups, which effectively solves the problem of mining coal under buildings, greatly reduces surface subsidence, and improves economic benefits. Tianchen Mine of Zaozhuang Mining Group adopts strike-wide strip mining for coal under village pressure, with a mining width of 120m, which significantly improves production efficiency. Zibo Xuchang Coal Mine, Yanzhou Nantun Coal Mine, and Linyi Gucheng Coal Mine have also implemented the practice of wide strip mining, taking into account resource mining and surface movement control, and achieved good results. Common prediction methods for surface subsidence caused by strip mining include influence function method and theoretical model method. The influence function method is to assume that the mining of small units will cause small surface movement on the surface, and the surface movement caused by the entire mining area can be regarded as the sum of small surface movement caused by all small unit mining in the mining area. The probability integral method is a typical influence function method, which is the most widely used and mature mining subsidence prediction method in my country. Since strip mining involves simultaneous or continuous mining of multiple working faces, the probability integral prediction parameters need to be corrected, and the results obtained by this method are not only difficult to explain in a physical sense, but also relatively inaccurate. The theoretical model rule is to abstract the rock mass into a certain theoretical model (such as viscoelastic rheological model, elastic-plastic finite element model), and calculate the movement, deformation, and stress distribution of the rock mass affected by mining according to this model.

[0004] In general, the research on mining subsidence by scholars at home and abroad has basically matured, but most of them are still mainly aimed at parameter acquisition, especially the acquisition of parameters representing the law of surface subsidence of collapsed working faces. There is still less attention paid to their relationship with rock formations, and the response law between them and rock formations is still unclear. The author first uses the BKA-BP neural network model to invert the parameters representing the law of surface subsidence in deep mines in the Ordos coalfield, analyzes the main controlling factors affecting these parameters, summarizes the response law between the main controlling factors and the parameters representing the law of surface subsidence, establishes a function model between the two, and applies it to a specific engineering background. Summary of the invention

[0005] The deep mining area of ​​Ordos Coalfield has rich coal resources, simple geological structure and sparse population, which is very suitable for high-intensity large-scale mining of deep coal resources. However, with the continuous increase in mining intensity, the Cretaceous and Jurassic thick sandstones in the eastern part of Ordos Coalfield have a great impact on large-scale continuous mining due to their special rock movement and energy accumulation properties. Based on this method of deep wide-band mining in some mining areas of Ordos Coalfield, active subsidence reduction and loss control technology is used to effectively slow down surface subsidence while ensuring coal recovery rate. However, due to the complex movement mechanism of the thick sandstone of Zhidan Group, its control role in wide-band mining has greatly limited it. Therefore, clarifying the position of thick sandstone in mining subsidence in the deep mining area of ​​Ordos Coalfield and establishing a surface subsidence prediction method that takes into account the control role of thick sandstone are of great significance for subsidence reduction and loss control and environmental protection.

[0006] To solve the above problems, the present invention proposes a surface subsidence prediction system and method which takes into account the influence of extremely thick and weakly cemented sandstone. Firstly, the BKA-BP neural network model is used to invert the surface subsidence law characterization parameters of Ordos coal deep mines, and the main controlling factors affecting these parameters are analyzed. The response law between the main controlling factors and the surface subsidence law characterization parameters is summarized, and a function model between the two is established, which is applied to a specific engineering context.

[0007] The technical solution adopted by the present invention is:

[0008] In a first aspect, the present invention proposes a method for predicting surface subsidence taking into account the influence of extremely thick and weakly cemented sandstone, the prediction method comprising the following steps:

[0009] Step 1: Establish a BKA-BP neural network model and use it to invert the parameters characterizing the surface subsidence law;

[0010] Step 2: Use the PCA method to analyze the main controlling factors of surface subsidence characterization parameters, and establish a mathematical model of surface subsidence characterization parameters and main controlling factors. This mathematical model is a fusion independent variable surface subsidence characterization parameter mathematical model that takes into account the control effect of the triple arch;

[0011] Step 3: Modify the surface subsidence characterization parameters based on the mining degree, and combine the surface subsidence characterization parameters established in step 2 with the mathematical model of the main controlling factors to obtain a bidirectional random medium prior model that takes into account the control effect of the triple arch;

[0012] Step 4: The surface subsidence prediction of the working face in Yingpanhao Coal Mine was completed using the bidirectional random medium prior model considering the control effect of the triple arch.

[0013] In one implementation, the establishment process of the BKA-BP neural network model in step 1 is as follows:

[0014] (1) Obtaining measured surface data from an already mined mining area, normalizing the measured data, and randomly dividing the processed data into training samples and test samples;

[0015] (2) Determine the activation function of the BP neural network;

[0016] (3) Reset the network structure parameters of the BP neural network according to the training samples;

[0017] (4) Use the following formula to calculate the fitness and optimal individual of the population;

[0018] X i =f(X i )=BK lb +rand(BK ub -BK lb ) (12)

[0019] Among them, X i is the fitness of the i-th black-winged kite, BK lb and BK ub are the upper and lower bounds of each black-winged kite, rand is a random number between [0,1]; X L As the "leader" in the initial group, during the initialization process, BKA selects the individual with the best fitness value as the "leader" in the initial group. L , taking the position of the "leader bird" as the optimal position of the black-winged kite, the "leader bird" can be expressed as the following formula, taking the minimum value as an example:

[0020]

[0021] (5) Individuals are updated through foraging and migration behaviors, that is, the position of the updated individuals is calculated using the following formula;

[0022]

[0023] n=0.05×exp(-2×(t / T) 2 ) (15)

[0024]

[0025]

[0026] In the above formula, and denote the position of the jth dimension of the black-winged kite in the ith position at the tth and t+1th iterations, respectively; ra is a random number between 0 and 1; p is a constant, which is 0.9 here; T is the total number of iterations; t is the number of completed iterations; represents the “leader” of the black-winged kite in the jth dimension at the tth iteration, F i represents the current position of any black-winged kite in the jth dimension in the tth iteration, F ri represents the fitness value of any black-winged kite at a random position in the jth dimension in the tth iteration, and C(0,1) represents the Cauchy mutation;

[0027] (6) Determine whether the number of iterations has been reached. If the iteration is completed, proceed to the next step. Otherwise, return to step ④ and continue to find the optimal position.

[0028] (7) The obtained optimal position is assigned to the weight value and bias term of the BP neural network to obtain the trained BKA-BP neural network model, and then the test sample is input to solve and verify the accuracy of the BKA-BP neural network model.

[0029] In one implementation, the surface subsidence characterization parameters of the mining area obtained in step 1 include the subsidence coefficient, the main influencing angle tangent and the inflection point offset, which are represented by q, tanβ and S respectively.

[0030] In one implementation, the PCA method was used to analyze the main controlling factors of the surface subsidence characterization parameters. The main controlling factors include the thickness of extremely thick sandstone, mining depth and mining thickness. Then, the function models of the subsidence coefficient and the thickness of extremely thick sandstone, mining depth and mining thickness, the function models of the main influencing angle tangent and the thickness of the loose layer and mining depth, and the function models of the inflection point offset and the thickness of extremely thick sandstone, mining depth and mining thickness were established respectively.

[0031] In one implementation, the function model of the subsidence coefficient and the thickness, mining depth and thickness of the thick sandstone is expressed as:

[0032]

[0033] The function model of the main influencing angle tangent, loose layer thickness and mining depth is expressed as:

[0034] tanβ=-2.0277+0.0056H-0.0042h(R 2 =0.99) (22)

[0035] The function model of the inflection point offset, the thickness of the thick sandstone, the mining depth and the mining thickness is expressed as:

[0036]

[0037] In the above formula, H is the mining depth, J is the thickness of the thick sandstone, J / H is the sand depth ratio, h is the thickness of the loose layer, and m is the mining thickness.

[0038] In some implementations, the leveling data obtained from the coal mine working face mining in step 3 is used to correct the surface subsidence rate under extremely insufficient mining conditions, and the function model between the strike and dip mining degree and the surface subsidence rate is expressed as:

[0039]

[0040] In the above formula, q' is the subsidence rate, n1 and n2 are the strike and dip mining degrees respectively, and the calculation formulas are D1 is the strike length, D2 is the dip length, and K is the mining intensity coefficient.

[0041] In a second aspect, the present invention proposes a surface subsidence prediction system taking into account the influence of extremely thick and weakly cemented sandstone. The prediction system is used to implement the above-mentioned prediction method, comprising:

[0042] Model building module, used to build the BKA-BP neural network model and deploy the BKA-BP neural network model to the background;

[0043] The parameter acquisition module calls the BKA-BP neural network model to invert the parameters representing the surface subsidence law;

[0044] The mathematical model building module uses the PCA method to analyze the main controlling factors of the surface subsidence characterization parameters, and establishes a mathematical model of the surface subsidence characterization parameters and the main controlling factors. The mathematical model is a fusion independent variable surface subsidence characterization parameter mathematical model that takes into account the control effect of the triple arch, and the mathematical model is deployed to the background;

[0045] The prior model building module modifies the surface subsidence characterization parameters based on the mining degree, and combines the surface subsidence characterization parameters with the mathematical model of the main controlling factors to obtain a bidirectional random medium prior model that takes into account the control effect of the triple arch, and deploys the prior model to the background;

[0046] The surface subsidence prediction module calls the prior model to predict the surface subsidence of the coal mine working face after mining.

[0047] In a third aspect, the present invention also provides an electronic device comprising a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory to implement the steps of a surface subsidence prediction method that takes into account the influence of extremely thick and weakly cemented sandstone.

[0048] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a computer to execute the steps of a surface subsidence prediction method that takes into account the influence of extremely thick and weakly cemented sandstone.

[0049] Beneficial effects of the present invention:

[0050] 1. The present invention firstly establishes a prediction model of surface subsidence law characterization parameters of the BKA-BP neural network based on the BKA algorithm, selects surface measured data of 41 mining areas for model verification and analysis, and believes that the training accuracy of the BKA-BP model is significantly better than that of the BP neural network, with an average relative error of a minimum of 6.6% and a maximum of 9.6%, which can more accurately predict the parameters of the probability integral method.

[0051] 2. The present invention combines the principal component analysis method to establish a function model of the surface subsidence coefficient and the thickness of the extremely thick sandstone, mining depth and mining thickness, which is q=0.9214-0.2537J / H-0.1053m, a function model of the main influencing angle tangent and the thickness of the loose layer and mining depth, which is tanβ=-2.0277+0.0056H-0.0042h, and a function model of the inflection point offset and the thickness of the extremely thick sandstone, mining depth and mining thickness, which is S / H=-0.0081-0.198J / H+0.009m. The above function models can be used to calculate the parameters characterizing the surface subsidence law of the working face to be predicted.

[0052] 3. The present invention uses the leveling data obtained from the working face mining of Yingpanhao Coal Mine to correct the surface subsidence rate under extremely insufficient mining conditions, and obtains the function model between the strike and dip mining degree and the surface subsidence rate:

[0053]

[0054] The mathematical model and random medium theory were used to complete the prediction of surface subsidence during mining at the 2101, 2102 and 2201, 2202 working faces in Yingpanhao Coal Mine. The errors and relative errors of the predicted results compared with the leveling results of the same period were 21.4mm, 88.2mm, 8.5% and 8.7% respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a flow chart of the present invention.

[0057] Figure 2 The morphology of a black-winged kite; (a) flying; (b) hunting.

[0058] Figure 3 Two hunting behaviors of the black-winged kite; (a) the attacking state of the black-winged kite when hovering in the air; (b) the hovering state in the air.

[0059] Figure 4 The migration behavior of black-winged kites.

[0060] Figure 5 Build a pipeline for the BKA-BP neural network.

[0061] Figure 6 Comparison of the output results of the subsidence coefficient, main influencing angle tangent, and inflection point offset test set; (a) Comparison of the output results of the subsidence coefficient test set; (b) Comparison of the output results of the main influencing angle tangent test set; (c) Comparison of the output results of the inflection point offset test set.

[0062] Figure 7 is the contribution rate of each principal component and the cumulative contribution rate.

[0063] Figure 8 Fitting the relationship between subsidence coefficient, sand depth ratio and mining depth.

[0064] Fig. 9 It is the main influencing factor for fitting the relationship between the tangent angle and the thickness of the loose layer and the mining depth.

[0065] Fig.10 Fitting the relationship between the inflection point offset, sand depth ratio and mining thickness.

[0066] Fig.11 is the fitting result of q / q′ and n1 and n2.

[0067] Fig.12 This is the contour map of surface subsidence when the 210 working face was mined at a depth of 1,634 m.

[0068] Fig.13 This is the contour map of surface subsidence when the 2201 working face was mined at 1654m.

[0069] Fig.14This is the contour map of surface subsidence caused by mining at the 2201 working face and the 2101 working face.

[0070] Fig.15 This is a comparison chart between the predicted subsidence and the actual level measurement.

[0071] Fig.16 This is the predicted result of surface subsidence at the 2102 working face.

[0072] Fig.17 This is a comparison chart between the predicted subsidence and the actual level measurement.

[0073] Fig.18 This is the estimated result of surface subsidence at the 2202 working face.

[0074] Fig.19 The estimated results of surface subsidence after mining of the four working faces are shown in Figure 2.

[0075] Fig. 20 Schematic diagram of the triple arch effect. DETAILED DESCRIPTION

[0076] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0077] This embodiment takes Yingpanhao Coal Mine as an example to predict the surface subsidence caused by the deep wide band mining process currently being carried out in the coal mine. Figure 1 As shown in the figure, a surface subsidence prediction method considering the influence of extremely thick weakly cemented sandstone is proposed, which specifically includes the following steps:

[0078] Step 1: Establish a BKA-BP neural network model and use it to invert the parameters characterizing the surface subsidence law;

[0079] Step 2: Use the PCA method to analyze the main controlling factors of surface subsidence characterization parameters, and establish a mathematical model of surface subsidence characterization parameters and main controlling factors. This mathematical model is a fusion independent variable surface subsidence characterization parameter mathematical model that takes into account the control effect of the triple arch;

[0080] Step 3: Based on the modification of the surface subsidence characterization parameters of the mining degree, and combining the surface subsidence characterization parameters established in step 2 with the mathematical model of the main controlling factors, a bidirectional random medium prior model considering the control effect of the triple arch is obtained; among which, the control effect of the triple arch is as follows Fig. 20 As shown;

[0081] Step 4: The surface subsidence prediction of the working face in Yingpanhao Coal Mine was completed using the bidirectional random medium prior model considering the control effect of the triple arch.

[0082] The above four steps are described in detail below.

[0083] 1. Inversion of parameters characterizing surface subsidence law based on BKA-BP neural network

[0084] 1.1 Construction of BKA-BP neural network model

[0085] (1) Black Kite Algorithm

[0086] The black-winged kite is a bird of the genus Black-winged Kit, family Accipitridae, order Falconiformes, also known as the gray kite. Figure 2 As shown in Figure 2, their notable characteristics include migration and predation behavior. Wang et al. proposed a black kite optimization algorithm by deeply simulating the high adaptability of black kites to environmental changes and target locations. The imitation of this biological mechanism provides the algorithm with a strong dynamic search capability, enabling it to effectively respond to the ever-changing optimization environment. Based on the above characteristics, it can be seen that the BKA algorithm has the potential to solve some complex problems, and there is no research on applying the BKA algorithm to the field of mining subsidence. Therefore, this paper combines the BKA algorithm with the traditional BP neural network to establish a BKA-BP neural network model to invert the parameters characterizing the surface subsidence law.

[0087] For a black-winged kite population, there are i black-winged kites, each of which may appear in the j-dimensional search space. Then the positions of all black-winged kites can be represented by the following matrix:

[0088]

[0089] The fitness of each black-winged kite can be calculated according to the following formula:

[0090] X i =f(X i )=BK lb +rand(BK ub -BK lb ) (12)

[0091] In the above formula, BK lb and BK ub are the upper and lower bounds of each black-winged kite, and rand is a random number between [0,1].

[0092] During the initialization process, BKA selects the individual with the best fitness value as the "leader" X in the initial group. L , the position of the "leader" is taken as the optimal position of the black-winged kite. The "leader" can be expressed as the following formula, taking the minimum value as an example.

[0093]

[0094] In the above formula, f(X i ) is the fitness of the i-th black-winged kite. When hunting, the black-winged kite will rush towards the prey at a very fast speed. Figure 3 (a) and Figure 3 (b) shows the attack state and hovering state of the black-winged kite when it is hovering in the air. The mathematical model of the black-winged kite's predation behavior is as follows:

[0095]

[0096] n=0.05×exp(-2×(t / T) 2 ) (15)

[0097] In the above formula, and They represent the position of the jth dimension of the black-winged kite in the ith level at the tth and t+1th iterations respectively, ra is a random number between 0 and 1, p is a constant, which is 0.9 here; T is the total number of iterations, and t is the number of completed iterations.

[0098] Bird migration is a complex behavior influenced by environmental factors such as climate and food supply. Bird migration is to adapt to seasonal changes, and many birds migrate from north to south in winter to obtain better living conditions and resources. Migration is usually led by leaders, and their navigation skills are crucial to the success of the team. Based on this black-winged kite optimization algorithm, a bird migration hypothesis is proposed: if the fitness value of the current population is less than the fitness value of the random population, then the leader will give up the leadership and join the migrating population, indicating that it is not suitable to lead the population forward. On the contrary, if the fitness value of the current population is greater than the fitness value of the random population, it will guide the population to its destination. This strategy can dynamically select excellent leaders to ensure the success of migration. Figure 4 The mathematical model of the migration behavior of black-winged kites is as follows:

[0099]

[0100]

[0101] In the above formula, represents the “leader” of the black-winged kite in the jth dimension at the tth iteration, and denotes the position of the black-winged kite in the i-th dimension at the t-th and t+1-th iterations, respectively. i represents the current position of any black-winged kite in the jth dimension in the tth iteration, F ri represents the fitness value of any black-winged kite at a random position in the jth dimension in the tth iteration, and C(0,1) represents the Cauchy mutation.

[0102] The one-dimensional Cauchy distribution is a continuous probability distribution with two parameters. The probability density function of the one-dimensional Cauchy distribution is:

[0103]

[0104] When δ=1,μ=0, its probability density function is in standard form, as follows:

[0105]

[0106] (2) Model for obtaining parameters representing surface subsidence law based on BKA-BP neural network

[0107] BP neural network is a multi-layer feedforward network trained by the error back propagation algorithm. It is trained by obtaining a certain amount of sample data, and the weight values ​​and bias terms of the neural network are updated continuously so that the error function decreases along the negative gradient direction until the requirements are met. Compared with other algorithms, BP neural network has better self-learning and adaptive capabilities and stronger generalization. At the same time, when some neurons are damaged or there is noise or incompleteness in the data set, BP neural network can also have good fault tolerance. However, when performing nonlinear fitting, it is easy to fall into the problem of local optimal solution and slow convergence speed. To this end, this study uses the BKA optimization algorithm to continuously correct the number of hidden layer nodes during BP neural network training, obtain the optimal weight value and bias term, so as to reduce the error between the output value of the network and the expected value, thereby improving the convergence speed of the BP neural network. The flow chart of the BKA-BP neural network model is shown below. Figure 5 shown.

[0108] (1) Obtaining measured surface data from an already mined mining area, normalizing the measured data, and randomly dividing the processed data into training samples and test samples;

[0109] (2) Determine the activation function of the BP neural network. Common activation functions include sigmoid function, Tanh function, ReLU function and its variants. Try them one by one and get the best result.

[0110] (3) Reset the network structure parameters of the BP neural network according to the training samples;

[0111] (4) Use equations (12) and (13) to calculate the fitness and optimal individual of the population;

[0112] (5) Update individuals through foraging and migration behaviors, that is, use equations (14) and (16) to calculate the position of the updated individuals;

[0113] (6) Determine whether the number of iterations has been reached (the number of iterations will be pre-set when the algorithm is running, generally around 50 to 200). If the iteration is completed, execute the next step. Otherwise, return to step ④ and continue to search for the optimal position.

[0114] (7) The obtained optimal position is assigned to the weight value and bias term of the BP neural network to obtain the trained BKA-BP neural network model, and then the test sample is input to solve and verify the accuracy of the BKA-BP neural network model.

[0115] 1.2 Model Setting and Accuracy Comparison Method

[0116] The parameters characterizing the law of surface subsidence are closely related to geological mining conditions. This paper selected 41 sets of data from the guidelines for coal pillar retention and coal compression mining in buildings, water bodies, railways and main shafts and tunnels and the measured data of coal mining working faces in the western mining area to verify the model. The specific data are shown in Table 1. The five geological mining condition parameters, namely, average mining depth H, loose layer thickness h, depth ratio δ, mining thickness m, working face strike length D1, and working face dip length D2, are used as the input values ​​of the BKA-BP neural network, and the three parameters characterizing the law of surface subsidence, namely, q, tanβ, and S, are used as the output values ​​of the BKA-BP neural network.

[0117] Table 1 Measured data

[0118]

[0119]

[0120] Considering that the range of values ​​of input parameters and output parameters is quite different, direct input for training will cause large errors in the training results. Therefore, the Mapminmax function is used to normalize the sample data before training to improve the convergence speed and training accuracy. Since the optimal weights and bias terms of each probability integral method prediction parameter may be different, this embodiment adopts multiple input layers and a single output layer, that is, 5 input nodes and 1 output node, to establish 5 independent optimization models.

[0121] 1.3 Model Accuracy Evaluation

[0122] This paper uses Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) to measure the solution accuracy of the two models. The specific calculation formula is as follows:

[0123]

[0124] In the above formula, xi is the measured value; is the model output result; N is the number of samples.

[0125] The accuracy comparison between the BP neural network model and the BKA-BP neural network model is shown in the table below. The mean absolute error and root mean square error of the parameters solved by the BKA-BP neural network are close to 0 and less than the value of the BP neural network. Therefore, the prediction accuracy of the BP neural network optimized by the Black Kite algorithm is higher. From the perspective of the mean absolute percentage error, the maximum value of the BKA-BP neural network is 9.68, and the minimum value of the BP neural network is 10.50%, indicating that the BKA-BP neural network predicts that the parameters characterizing the law of surface subsidence are highly accurate and the model is stable.

[0126] Table 2 Comparison of the accuracy of BP neural network and BKA-BP neural network models

[0127]

[0128] After testing, it was found that the optimal number of hidden layer nodes for the sinking coefficient is 7, the optimal number of hidden layer nodes for the main influencing angle tangent value is 5, and the optimal number of hidden layer nodes for the inflection point offset is 10.

[0129] The model output results are organized as follows Figure 6 As shown in the figure, the solution results of both the BKA-BP neural network model and the BP neural network model are very close to the measured data, but the absolute error of the BKA-BP neural network model is smaller, indicating that its model prediction results are closer to the measured values ​​than the BP neural network model.

[0130] 1.4 Inversion of parameters characterizing surface subsidence in the western mining area

[0131] Using the established BKA-BP neural network model, the geological mining conditions of some western mining areas were collected (see Table 4) and the surface subsidence characterization parameters were inverted (see Table 3).

[0132] Table 3 Characterization parameters of surface subsidence in some western mining areas

[0133]

[0134] Table 4 Geological mining conditions in some western mining areas

[0135]

[0136] 2. Mathematical model of surface subsidence characterization parameters with integrated independent variables considering the control effect of triple arch

[0137] Based on the surface subsidence characterization parameters of some western mining areas obtained by the BKA-BP neural network model in the previous section, this section uses the principal component analysis method to explore the impact of special geological mining conditions in the western region on the surface subsidence characterization parameters.

[0138] Table 5 Principal component eigenvalues ​​and variance contribution rates

[0139]

[0140] In selecting the principal components, the selection is made based on the cumulative variance contribution rate of the principal components. It is generally believed that the principal components that reach 85% can cover the original information. This embodiment selects the first three principal components that contain 85%. It can be seen from the above table that the variance contribution rate of the first eigenvalue is the largest, reaching 35%; the second eigenvalue reaches 32.11%, and the cumulative is 67.11%; the third eigenvalue reaches 18.51%, and the cumulative is 85.62%; that is, after the principal component analysis, this article selects the first three eigenvalues ​​as the main influencing factors of geological mining conditions, which are used to analyze the influence of special geological mining conditions in the western region on the parameters characterizing the law of surface subsidence. The changes in the contribution rate and cumulative contribution of each principal component are shown in the following figure. Figure 7 shown.

[0141] 2.2 Analysis of parameter response laws based on fusion of independent variables to characterize surface subsidence law

[0142] On the basis of the principal component analysis results in the previous section, the first three indicators of geological mining conditions, namely, the thickness of extremely thick sandstone, mining depth, and mining thickness, are taken, and the multivariate linear regression method is used to fit the parameters characterizing the surface subsidence law.

[0143] (1) Sinking coefficient q

[0144] The above-mentioned western mines all have the Zhidan Group sandstone rock mass, which is characterized by relatively weak lithology but relatively large overall rigidity. Under the influence of a rock layer, the movement law of the rock layer in the deep mining area of ​​the western mining area is somewhat similar to that of the rock layer in the hard overburden mining area in the eastern region, but there are great differences in the amount of subsidence and the scope of influence (such as Figure 8 As shown). Therefore, this section focuses on analyzing the response of the thickness of super thick sandstone to the parameters characterizing the law of surface subsidence. After comparative analysis, the ratio of super thick sandstone thickness to mining depth (sand depth ratio) and mining thickness were used for multivariate function regression to fit the subsidence coefficient, and the final empirical formula for the subsidence coefficient q is as follows:

[0145]

[0146] The average mining depth of the Hongqing River 301 working face is 674m, the thickness of the super thick sandstone is 92.3m, and the ratio of the super thick sandstone thickness to the mining depth is 0.14. According to the above fitting formula, the subsidence coefficient of the Hongqing River 301 working face is 0.165, while the subsidence coefficient of the Hongqing River 301 working face inverted by the BKA-BP neural network algorithm is 0.16, and the difference between the two is 0.005. It can be considered that the relationship between the subsidence coefficient (q) and the sand depth ratio (J / H) and the mining thickness (m) fitted by this formula is reasonable. The average mining depth of the 2-2 coal seam currently mined in the Yingpanhao Mine is 725m, the thickness of the super thick sandstone is 120m, the ratio of the super thick sandstone thickness to the mining depth is 0.165, and the mining thickness is 6.4m. The subsidence coefficient of the Yingpanhao Mine is 0.2 obtained by the above fitting formula.

[0147] (2) Main influencing angle tangent tanβ

[0148] The main influencing angle tangent is related to the lithology of the overlying strata, mining depth, coal seam inclination, and impact layer thickness. Considering the similarity of the strata in this area, the strength of the overlying strata is not much different, and they are all nearly horizontal coal seams without impact layers (such as Fig. 9 As shown in Figure 2 ). Therefore, the relationship between the main influencing angle tangent and the mining depth is mainly considered. At the same time, the existence of loose layer also expands the influence range of mining subsidence. Therefore, this paper fits the main influencing angle tangent by mining depth (H) and loose layer thickness (h), and obtains the following results.

[0149] tanβ=-2.0277+0.0056H-0.0042h(R 2 =0.99) (22)

[0150] The average mining depth of Hongqing River 301 working face is 674m, and the thickness of loose layer is 3m. According to the above fitting formula, the subsidence coefficient of Hongqing River 301 working face is 1.75, while the subsidence coefficient of Hongqing River 301 working face inverted by BKA-BP neural network algorithm is 1.73, and the difference between the two is 0.02. It can be considered that the relationship between the main influence angle tangent (tanβ) and the mining depth (H) and the thickness of loose layer (h) fitted by this formula is reasonable. The average mining depth of the 2-2 coal seam currently mined in Yingpanhao Mine is 725m, and the thickness of loose layer is 86m. The main influence angle tangent (tanβ) of Yingpanhao Coal Mine is 1.67 obtained by the above fitting formula.

[0151] (3) Inflection point offset

[0152] like Fig.10 As shown in the figure, after comparative analysis, the best effect of fitting the subsidence coefficient is to use the ratio of the thickness of the thick sandstone to the mining depth (sand depth ratio) and the mining thickness for multivariate function regression. The final inflection point offset distance is The empirical formula is as follows:

[0153]

[0154] In the above formula, H is the mining depth, J is the thickness of the thick sandstone, J / H is the sand depth ratio, h is the thickness of the loose layer, and m is the mining thickness.

[0155] The average mining depth of the Muduchaideng 30201 working face is 743m, the thickness of the super thick sandstone is 162m, and the ratio of the super thick sandstone thickness to the mining depth is 0.22. According to the above fitting formula, the inflection point offset distance of the Muduchaideng 30201 working face is 0.34H, and the subsidence coefficient of the Hongqinghe 301 working face inverted by the BKA-BP neural network algorithm is 0.36H. The difference between the two is 0.02 times the mining depth. It can be considered that the relationship between the inflection point offset distance (S / H), sand depth ratio (J / H) and mining thickness (m) fitted by this formula is reasonable.

[0156] The average mining depth of the 2-2 coal seam currently mined in Yingpanhao Mine is 725m, the thickness of the thick sandstone is 120m, the ratio of the thick sandstone thickness to the mining depth is 0.165, and the mining thickness is 6.4m. Using the above fitting formula, the inflection point offset distance of Yingpanhao Coal Mine is 0.46H, or 333.5m.

[0157] 3. Introduction to the priori model for surface subsidence prediction based on random medium theory

[0158] 3.1 Introduction to Probability Integration Method

[0159] The probability integral method was developed by Chinese scholars Liu Baochen and Liao Guohua based on the random medium theory proposed by Polish scholar Litwiniszyn. This method believes that the mining law caused by underground mining is similar to the random motion law of granular media in macroscopic phenomena. By integrating the unit mining subsidence basin, the surface movement and deformation values ​​caused by underground mining are obtained.

[0160] The predicted formulas for the movement deformation of any point (x, y) on the ground surface within the range of any shape working surface of a horizontal or gently inclined coal seam are shown in equations (1) to (5):

[0161] The formula for calculating ground subsidence is as follows:

[0162]

[0163] Surface horizontal movement calculation formula:

[0164]

[0165] Calculation formula for surface tilt deformation:

[0166]

[0167] Calculation formula for surface curvature deformation:

[0168]

[0169] Calculation formula for horizontal surface deformation:

[0170]

[0171] For the above formula, η and ξ are integral variables; θ is the propagation angle of mining influence; r is the main influence radius, and r = H / tanβ, H is the mining depth, tanβ is the tangent of the main influence angle; W max is the maximum surface subsidence value, and W max =mqcosα, m is the mining thickness, q is the subsidence coefficient, α is the coal seam inclination; U max is the maximum horizontal movement of the surface, and U max =bW max , b is the horizontal shift coefficient.

[0172] 3.2 Introduction to Probability Density Function Method

[0173] Through theoretical analysis and computer simulation, Guo Zhangzhang found that under extremely insufficient mining conditions, the movement of fragments in discrete media approaches the probability density function of the normal distribution. At this time, when using the probability integral method to predict surface subsidence, it has shown strong inadaptability. Therefore, Guo Zhangzhang proposed a probability density function to predict surface subsidence under extremely insufficient mining conditions.

[0174] Taking strike full mining and inclination insufficiency mining as an example, the movement deformation prediction formulas of any point (x, y) on the ground surface within the range of any shape working face under the condition of extremely insufficiency mining of horizontal or gently inclined coal seams are shown in equations (6) to (10):

[0175]

[0176] Calculation formula for the surface tilt deformation along the φ direction:

[0177]

[0178] Calculation formula of curvature deformation of the ground surface along the φ direction:

[0179]

[0180] Calculation formula for horizontal movement of the ground surface along the φ direction:

[0181]

[0182] Calculation formula for horizontal surface deformation:

[0183]

[0184] For the above formula, L is the strike length; l is the calculated length, and l = L-2S; r is the main influence radius, and r = H / tanβ, H is the mining depth, tanβ is the main influence angle tangent; W max is the maximum surface subsidence value, and W max =mqcosα, m is the mining thickness, q is the subsidence coefficient, α is the inclination of the coal seam; b is the horizontal movement coefficient.

[0185] 4. A priori model of bidirectional random medium considering the control effect of triple arch

[0186] 4.1 Correction of ground subsidence characterization parameters based on mining intensity

[0187] The degree of surface mining is directly related to the mining size of the working face. According to the product of the mining width-to-depth ratio and the mining degree coefficient K, the mining degree can be divided into extremely insufficient mining, inadequate mining, sufficient mining and super-sufficient mining. Mining before the overlying controlling rock formation is usually called extremely insufficient mining. The relationship between the ratio of mining width to mining depth and the mining degree is shown in the following table.

[0188] Table 6 Correspondence table of mining degree

[0189]

[0190] The value of the mining degree coefficient is generally determined by the lithology of the rock formation. According to the measured data of a mine adjacent to the Yingpanhao Coal Mine, K = 0.55. Taking the 2201 working face of the Yingpanhao Coal Mine as an example, its strike mining width is 2500m and its dip mining width is 300m. Then its strike mining degree is:

[0191]

[0192] Degree of tendency to mining:

[0193]

[0194] It is obvious that the strike direction has been over-mined, but due to the limitation of the working face inclination direction, the dip direction has not yet been extremely under-mined.

[0195] The author collected the measured surface data of Yingpanhao Coal Mine and established the relationship between the subsidence rate and the subsidence coefficient based on the strike, dip and degree of mining.

[0196] Table 7 Partial leveling observation results of Yingpanhao Coal Mine

[0197]

[0198]

[0199] The relationships established are as follows:

[0200]

[0201] For the above formula, q′ is the subsidence rate, q is the subsidence coefficient calculated according to formula (21), n1 and n2 are the strike and dip mining degrees respectively. From the leveling data in Table 7 and Fig.11 According to the fitting results, it is believed that when the width-to-depth ratio is greater than 3, the maximum subsidence value no longer changes, and the working face reaches super-sufficient mining. According to formulas (24) and (25), when n1 and n2 are greater than 1.65, they are taken as 1.65.

[0202] 4.2 Example of surface subsidence prediction for deep wide-band mining considering the control of extremely thick weakly cemented sandstone

[0203] Combined with the mining degree of Yingpanhao Coal Mine, this paper intends to use the probability integral method in strike and the probability density function method in dip to predict the surface subsidence of 2201, 2101 working faces, 2201, 2202 working faces and 2101, 2102 working faces respectively. The prediction method of this paper is verified from the two perspectives of superposition of multiple working faces under strip mining and pseudo-deep wide band mining, and the prediction model of surface subsidence under pseudo-deep wide band mining is constructed.

[0204] For the 2101 and 2201 working faces, the strike lengths are 1634 m and 1654 m respectively, the dip lengths are both 300 m, and the mining depth is 725 m. Substituting the above data into formula (26) the ground subsidence rate is obtained. The main influencing angles tangent and take the values ​​in Section 3.2 respectively. The estimated parameter values ​​of the two working faces are shown in Table 8. Fig.12 and Fig.13 The following are the contour maps of the surface subsidence of the 2101 working face and the 2201 working face respectively. After superimposing the mining subsidence results of the two working faces, we can get the following: Fig.14 The surface subsidence contour map when the 2101 working face and the 2201 working face are mined simultaneously is shown in the figure. The predicted results of the trend line extracted by ArcGIS are compared with the leveling data, and the following is obtained: Fig.15 Comparison chart of the predicted surface subsidence and the measured subsidence. The estimated mean error of the survey line is 21.4mm and the relative error is 8.5%. Therefore, it can be considered that the two working faces are independent of each other during the mining process, satisfying the principle of linear superposition. At the same time, it also proves the feasibility of the method of predicting surface subsidence in this paper.

[0205] Table 8 Mining characteristics and estimated parameters of 2101 and 2202 working faces

[0206]

[0207] The 2102 working face is located 30 meters below the 2101 working face. The dimensions of the working face are consistent with those of the 2101 working face, with the strike length and dip length both being 2100m and 300m. After the 2202 working face is mined, the 2102 working face will be mined. After the mining of the two working faces is completed, the surface subsidence rate is 0.2727, and the maximum subsidence value is 1746mm.

[0208] The estimated results of the points south of the boundary of the 2101 working surface on the west trend line were extracted using ArcGIS and compared with the leveling data. Fig.17 Comparison chart of the predicted surface subsidence and the measured subsidence. The estimated mean error of the survey line is 88.2mm and the relative error is 8.7%.

[0209] The 2202 working face is located 30 meters above the 2201 working face. The working face dimensions are consistent with the 2201 working face, with the strike length and dip length both being 2500m and 300m. Fig.18 After the mining of the 2201 working face was completed, the 2202 working face was mined. After the mining of the two working faces was completed, the surface subsidence rate was 0.2937, and the maximum subsidence value was 1879mm.

[0210] After the mining of the four working faces 2101, 2101, 2201 and 2202 was completed, a pseudo-deep wide-band mining pattern was formed, with a protective coal pillar between two consecutive collapsed working faces. Therefore, similar to the previous 2101 working face and 2201 working face, the subsidence of the two collapsed goaf areas can be superimposed to obtain the surface subsidence after the mining of the four working faces was completed, as shown in Fig.19 At this point, the predicted model of surface subsidence due to deep wide-band mining has been completed.

[0211] 5. Conclusion

[0212] After research, the main conclusions are as follows:

[0213] (1) Based on the BKA algorithm, a BKA-BP neural network parameter prediction model for characterizing the surface subsidence law was established. Surface measured data from 41 mining areas were selected for model verification and analysis. It was found that the training accuracy of the BKA-BP model was significantly better than that of the BP neural network, with an average relative error of 6.6% at the minimum and 9.6% at the maximum. The BKA-BP model can more accurately predict the parameters of the probability integral method.

[0214] (2) Combined with the principal component analysis method, a function model of the surface subsidence coefficient and the thickness of the extremely thick sandstone, mining depth and mining thickness was established, which is q = 0.9214-0.2537 J / H-0.1053 m, a function model of the main influencing angle tangent and the thickness of the loose layer and mining depth was tanβ = -2.0277 + 0.0056H-0.0042h, and a function model of the inflection point offset and the thickness of the extremely thick sandstone, mining depth and mining thickness was S / H = -0.0081-0.198 J / H + 0.009m.

[0215] (3) The leveling data obtained from the working face mining of Yingpanhao Coal Mine was used to correct the surface subsidence rate under extremely insufficient mining conditions, and the functional model between the strike and dip mining degree and the surface subsidence rate was obtained as follows:

[0216]

[0217] (4) Based on the function model obtained above, the surface subsidence of the 2101, 2201 and 2102 working faces that have been mined was predicted using a combination of probability density function method and probability integral method. The errors compared with the leveling data were 21.4 mm and 88.2 mm, respectively, and the relative errors were 8.5% and 8.7%. The surface subsidence of the 2202 working face was also predicted. After the 2201 working face was mined, the maximum surface subsidence of the 2202 working face was 1879 mm.

[0218] Symbol Notes

[0219] b is the horizontal shift coefficient; BK ij is the position of the i-th black-winged kite in the j-dimensional space; BK lb and BK ub are the upper and lower bounds of each black-winged kite; C(0,1) represents the Cauchy mutation; D1 is the strike length, m; D2 is the dip length, m; F i represents the current position of any black-winged kite in the jth dimension in the tth iteration; F ri represents the fitness value of any black kite at the random position in the jth dimension in the tth iteration; h is the thickness of the loose layer, m; H is the mining depth, m; i(x,y) is the tilt deformation of the point (x,y), mm / m; J is the thickness of the thick sandstone, m; K is the mining degree coefficient; K(x,y) is the curvature deformation of the point (x,y), mm / m 2 ; l is the calculated length, m; L is the strike length, m; L i trepresents the “leader” of the black-winged kite of the jth dimension at the tth iteration; m is the mining thickness, m; n1 is the mining degree of strike; n2 is the mining degree of dip; N is the number of samples; p is a constant; q is the sinking coefficient; q′ is the sinking rate; r is the main influence radius, m; rand is a random number between 0 and 1; S is the inflection point offset, m; t is the number of iterations that have been completed; T is the total number of iterations; tanβ is the main influence angle tangent; U max is the maximum horizontal movement value, mm; U(x,y) is the horizontal movement of the (x,y) point, mm; W max is the maximum sinking value, mm; W(x,y) is the sinking of point (x,y), mm; x i is the measured value; Output result of the model; X i is the fitness of the i-th black-winged kite; X L For the position of the leader bird; is the j-th dimension position of the i-th black-winged kite at the t-th time; is the position of the i-th black-winged kite in the j-th dimension at the t+1-th iteration; α is the coal seam inclination, °; ​​δ is the depth ratio; ε(x, y) is the horizontal deformation of the point (x, y) in mm / m; η is the integral variable; ξ is the integral variable.

[0220] The embodiments described above are only preferred specific implementation modes of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by any technician familiar with the field within the technical scope disclosed in the present invention belong to the protection scope of the present invention.

Claims

1. A method for predicting surface subsidence taking into account the influence of extremely thick and weakly cemented sandstone, characterized in that: The prediction method includes the following steps: Step 1: Establish a BKA-BP neural network model and use it to invert the parameters characterizing the surface subsidence law; Step 2: Use the PCA method to analyze the main controlling factors of surface subsidence characterization parameters, and establish a mathematical model of surface subsidence characterization parameters and main controlling factors. This mathematical model is a fusion independent variable surface subsidence characterization parameter mathematical model that takes into account the control effect of the triple arch; Step 3: Modify the surface subsidence characterization parameters based on the mining degree, and combine the surface subsidence characterization parameters established in step 2 with the mathematical model of the main controlling factors to obtain a bidirectional random medium prior model that takes into account the control effect of the triple arch; Step 4: The surface subsidence prediction of the working face in Yingpanhao Coal Mine was completed using the bidirectional random medium prior model considering the control effect of the triple arch.

2. A surface subsidence prediction method taking into account the influence of extremely thick and weakly cemented sandstone according to claim 1, characterized in that: The establishment process of the BKA-BP neural network model in step 1 is as follows: (1) Obtaining measured surface data from an already mined mining area, normalizing the measured data, and randomly dividing the processed data into training samples and test samples; (2) Determine the activation function of the BP neural network; (3) Reset the network structure parameters of the BP neural network according to the training samples; (4) Use the following formula to calculate the fitness and optimal individual of the population; X i =f(X i )=BK lb +rand(BK ub -BK lb ) (12) Among them, X i is the fitness of the i-th black-winged kite, BK lb and BK ub are the upper and lower bounds of each black-winged kite, rand is a random number between [0,1]; X L It is the "leader" in the initial group; During the initialization process, BKA selects the individual with the best fitness value as the "leader" in the initial group. L , taking the position of the "leader bird" as the optimal position of the black-winged kite, the "leader bird" can be expressed as the following formula, taking the minimum value as an example: (5) Individuals are updated through foraging and migration behaviors, that is, the position of the updated individuals is calculated using the following formula; n=0.05×exp(-2×(tT) 2 ) (15) In the above formula, and denote the position of the jth dimension of the black-winged kite in the ith position at the tth and t+1th iterations, respectively; ra is a random number between 0 and 1; p is a constant, which is 0.9 here; T is the total number of iterations; t is the number of completed iterations; represents the "leader" of the black-winged kite in the jth dimension at the tth iteration, F i represents the current position of any black-winged kite in the jth dimension in the tth iteration, F ri represents the fitness value of any black-winged kite at a random position in the jth dimension in the tth iteration, and C(0,1) represents the Cauchy mutation; (6) Determine whether the number of iterations has been reached. If the iteration is completed, proceed to the next step. Otherwise, return to step (4) and continue to search for the optimal position. (7) The obtained optimal position is assigned to the weight value and bias term of the BP neural network to obtain the trained BKA-BP neural network model, and then the test sample is input to solve and verify the accuracy of the BKA-BP neural network model.

3. A surface subsidence prediction method taking into account the influence of extremely thick and weakly cemented sandstone according to claim 2, characterized in that: The surface subsidence characterization parameters in step 1 include the subsidence coefficient, the main influencing angle tangent and the inflection point offset, which are represented by q, tanβ and S respectively.

4. A surface subsidence prediction method taking into account the influence of extremely thick and weakly cemented sandstone according to claim 3, characterized in that: In step 2, the PCA method is used to analyze the main controlling factors of the surface subsidence characterization parameters. The main controlling factors include the thickness of extremely thick sandstone, mining depth and mining thickness. Then, the function model of the subsidence coefficient and the thickness of extremely thick sandstone, mining depth and mining thickness, the function model of the main influencing angle tangent and the thickness of loose layer and mining depth, and the function model of the inflection point offset and the thickness of extremely thick sandstone, mining depth and mining thickness are established respectively.

5. A surface subsidence prediction method taking into account the influence of extremely thick and weakly cemented sandstone according to claim 4, characterized in that: The functional model of subsidence coefficient, thickness of thick sandstone, mining depth and mining thickness is expressed as: The function model of the main influencing angle tangent, loose layer thickness and mining depth is expressed as: tanβ=-2.0277+0.0056H-0.0042h(R 2 =0.99) (22) The function model of the inflection point offset, the thickness of the thick sandstone, the mining depth and the mining thickness is expressed as: In the above formula, H is the mining depth, J is the thickness of the thick sandstone, h is the thickness of the loose layer, J / H is the sand depth ratio, and m is the mining thickness.

6. A method for predicting surface subsidence taking into account the influence of extremely thick and weakly cemented sandstone according to claim 1, characterized in that: In step 3, the leveling data obtained from the coal mine working face mining is used to correct the surface subsidence rate under the extremely insufficient mining conditions, and the function model between the strike and dip mining degree and the surface subsidence rate is expressed as: In the above formula, q' is the subsidence rate, n1 and n2 are the strike and dip mining degrees respectively, and the calculation formulas are D1 is the strike length, D2 is the dip length, and K is the mining intensity coefficient.

7. A surface subsidence prediction system taking into account the influence of extremely thick and weakly cemented sandstone, characterized in that: The prediction system is used to implement the prediction method described in any one of claims 1 to 6, including: Model building module, used to build the BKA-BP neural network model and deploy the BKA-BP neural network model to the background; The characterization parameter acquisition module calls the BKA-BP neural network model to invert the characterization parameters of the surface subsidence law; The mathematical model building module uses the PCA method to analyze the main controlling factors of the surface subsidence characterization parameters, and establishes a mathematical model of the surface subsidence characterization parameters and the main controlling factors. The mathematical model is a fusion independent variable surface subsidence characterization parameter mathematical model that takes into account the control effect of the triple arch, and the mathematical model is deployed to the background; The prior model building module modifies the surface subsidence characterization parameters based on the mining degree, and combines the surface subsidence characterization parameters with the mathematical model of the main controlling factors to obtain a bidirectional random medium prior model that takes into account the control effect of the triple arch, and deploys the prior model to the background; The surface subsidence prediction module calls the prior model to predict the surface subsidence of the coal mine working face after mining.

8. An electronic device comprising a processor and a memory, characterized in that: The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory to implement the steps of a surface subsidence prediction method taking into account the influence of extremely thick and weakly cemented sandstone as described in any one of claims 1-6.

9. A computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause a computer to execute the steps of a surface subsidence prediction method taking into account the influence of extremely thick and weakly cemented sandstone as described in any one of claims 1-6.

Images