Gas blasting parameter prediction and hole arrangement design method based on back propagation neural network

Abstract

The invention provides a gas blasting parameter prediction and hole arrangement design method based on a back propagation neural network. The method comprises the following steps: S1, obtaining physical and mechanical parameters of a rock mass in a blasting area; s2, carrying out a single-hole gas bursting test, and determining an optimal resistance line; s3, the step S1 and the step S2 are repeated in different sites, corresponding data are counted, and a database is established; s4, constructing and training a back propagation neural network, and predicting reasonable gas blasting parameters by inputting training data; and S5, optimizing hole arrangement design according to a network prediction result, and determining hole depth, hole pitch and hole pitch. According to the method, the blasting parameters are accurately predicted, the rock mass parameters and the site conditions are combined, the hole distribution parameters are reasonably designed, the rock breaking cost is remarkably reduced, the rock breaking efficiency is improved, and then the economic benefits of blasting operation are optimized.

Description

Technical Field

[0001] The present invention belongs to the field of rock mass engineering blasting construction, and particularly relates to a method for predicting gas blasting parameters and hole layout design based on a backpropagation neural network. Background Art

[0002] Explosive blasting has advantages such as high efficiency and economy, and is widely used in rock mass excavation of projects such as water conservancy and hydropower, transportation, energy, and municipal engineering. However, the disadvantages of explosive blasting are also very prominent. The use of explosives has high safety risks and strict management and control. The generated explosion shock waves will bring relatively large vibrations, noises, flying rocks, and damage to the rock mass in the reserved area. As a result, environmental protection problems and engineering disasters have emerged in an endless stream.

[0003] To achieve the purpose of controlling vibrations, flying rocks, and surrounding rock damage and overcome the disadvantages of explosive blasting, gas blasting technologies such as CO 2 phase change blasting are often used to replace explosives for rock breaking and excavation operations. CO 2 Blasting is to quickly heat the carbon dioxide in the fracturing pipe to rapidly increase its pressure, and then suddenly release it, generating an impact force on the surrounding rock mass, thereby achieving the effect of rock breaking. Due to the fact that the activator materials used in the early CO 2 phase change blasting technology belong to dangerous chemicals and there are still certain safety risks, some manufacturers have improved the activator formula, thus avoiding the use of dangerous chemicals. In addition, driven by market demand, the supercritical fluid impact rock breaking technology has emerged, as described in the Chinese invention patent "Method for open cut construction of rock mass based on supercritical fluid impact rock breaking equipment" CN118463736A. The supercritical fluid impact rock breaking technology impacts the rock by quickly releasing high-pressure gas, and the principle is similar to that of CO 2 phase change blasting. In the present invention, technologies with similar principles such as CO 2 blasting and supercritical fluid impact rock breaking technology are collectively referred to as gas blasting.

[0004] So far, the design of hole pattern parameters in actual gas blasting construction still mainly relies on experience and has a certain degree of blindness. The Chinese invention patent "A method for carbon dioxide blasting construction to control bedrock damage and vibration effect" CN112364489B adjusts the construction parameters according to the requirements of damage and vibration control, rather than designing the hole layout parameters with the goal of improving rock breaking efficiency and reducing rock breaking cost.

[0005] A low profit margin is a significant feature of the construction industry. In gas blasting construction, if reasonable hole layout parameters can be determined based on rock mass parameters and site conditions, the rock breaking cost will be significantly reduced, and the market competitiveness of gas blasting technology will be greatly improved. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a gas blasting parameter prediction and hole layout design method based on a backpropagation neural network to solve the problems raised in the above technical background.

[0007] To achieve the above object, the present invention is realized through the following technical solutions:

[0008] A gas blasting parameter prediction and hole layout design method based on a backpropagation neural network includes the following steps:

[0009] Step S1: Obtain the physical and mechanical parameters of the rock mass in the blasting area, including: the point load strength I of the rock block s(50) , the wave impedance ρV of the rock block pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of the main structural plane 5 ;

[0010] A rock mass is a geological body composed of structural planes and the rock blocks enclosed by them, with complex physical and mechanical properties. The selected physical and mechanical indexes should have a significant impact on the gas blasting effect and be easy to measure. The rock mass parameters selected in the present invention include: the point load strength I of the rock block s(50) , the wave impedance ρV of the rock block pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of the main structural plane 5 . Among them, ρ is the density of the rock block, and the rock mass integrity index K v is calculated according to formula (1):

[0011] K v =(V pm / V pr ) 2 (8)

[0012] In the formula, V pm is the longitudinal wave velocity of the rock mass, and V pr is the longitudinal wave velocity of the rock block. The influence coefficient K of the occurrence of the main structural plane 5 is calculated according to formula (2):

[0013] K 5 =F 1 ×F 2 ×F 3 (9)

[0014] In the formula, F 1 represents the influence of the angle between the dip direction of the structural plane and the dip direction of the fracture surface, F 2 represents the influence of the dip angle of the structural plane, and F 3 represents the influence of the difference between the dip angle of the structural plane and the dip angle of the slope surface. F 1 , F 2 , F3 Take the value according to Table 5.3.2-3 in the "Engineering Rock Mass Classification Standard" (GB / T50218-2014).

[0015] Step S2: Conduct multiple single-hole gas blasting tests in the same site, measure and record the rock-breaking volume of each test, and determine the optimal burden and other key blasting parameters under the corresponding site conditions, including blasting energy, specific consumption, and gas blasting energy per meter of hole length;

[0016] Further, Step S2 includes the following steps:

[0017] Step S201: Conduct multiple single-hole gas blasting rock-breaking tests in the same site. Keep the test conditions consistent each time, only adjust the burden W. By statistically analyzing the rock-breaking volume V of each test, determine the maximum rock-breaking volume V O The optimal burden W under the corresponding site conditions O

[0018] Step S202: Calculate the blasting energy E of the single-hole test according to the blasting energy calculation formula for compressed gas and water vapor containers g :

[0019]

[0020] In the formula, k is the adiabatic index of carbon dioxide, taking 1.295; p is the blasting pressure; p atm is the atmospheric pressure; V 0 is the initial gas volume of a single blasting, that is, the volume of the fracturer multiplied by the number of single-hole fracturers;

[0021] Step S203: Based on the blasting energy E g Calculate the optimal specific consumption e under the corresponding rock mass properties and site conditions O :

[0022]

[0023] And the gas blasting energy λ per meter along the borehole:

[0024]

[0025] In the formula, H is the hole depth and T is the stemming length.

[0026] Step S3: On the blasting sites under different geological conditions, repeat Step S1 and Step S2, accumulate data through multiple experiments, and statistically analyze the corresponding data and the key blasting parameters obtained from the experiments to establish a gas blasting rock-breaking operation database including different geological conditions; The established database is in tabular form as shown in Table 1. In the table, α represents the bench slope angle.

[0027] Table 1 Gas Blasting Database in List Form

[0028] Venue Number <![CDATA[I s(50) > <![CDATA[ρV pr > <![CDATA[K v > <![CDATA[K 5 > α λ <![CDATA[W O > <![CDATA[e O > 1 2 ... N

[0029] Step S4: Construct and train a gas blasting parameter prediction model based on a backpropagation neural network. The input training data includes physical and mechanical parameters and experimental data. The output is the predicted gas blasting parameters, and the neural network structure is optimized through the backpropagation algorithm to improve the prediction accuracy;

[0030] Step S401: Determine the topological structure of the backpropagation neural network according to the characteristics and correlations of gas blasting parameters, and construct a preliminary neural network model;

[0031] Establish a backpropagation neural network, including an input layer, an output layer, and a hidden layer. Preferably, a backpropagation neural network model established in the present invention is provided with one hidden layer. Taking the point load strength I of rock blocks s(50) , the wave impedance ρV of rock blocks pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of main structural planes 5 , the bench slope angle α, and the gas blasting energy per meter λ as the 6 input indexes, and taking the reasonable burden W O and the reasonable specific charge e O as the output indexes, that is, 6 neurons are set in the input layer and 2 neurons are set in the output layer. The role of the hidden layer neurons is to extract and store the internal laws from the samples. Each hidden layer neuron has several weight values, and each weight value is a parameter to enhance the network mapping ability. If the number of hidden layer neurons is too small, the network's ability to obtain information from the samples is poor and it is not enough to generalize and reflect the laws in the training samples; if the number of hidden layer neurons is too large, it may also learn and remember the non-regular content in the samples, resulting in the problem of "overfitting", which instead reduces the network's generalization ability. Therefore, the number of hidden layer neurons depends on the number of training samples, the size of sample noise, and the complexity of the laws contained in the samples. According to the Kolmogolov theorem, the number of hidden layer units in the present invention is taken as 2m + 1, where m is the number of training samples.

[0032] Step S402: Select an activation function suitable for gas blasting parameter prediction and set the initial weight values;

[0033] The activation function f in the backpropagation neural network model usually takes a differentiable monotonic increasing function. In the present invention, the activation function of the backpropagation neural network model adopts the sigmoid-type logarithmic function logsig. The advantage of choosing the sigmoid function is that the input of any data can be transformed into a number between (0, +1), and the activation function of the last layer of neurons adopts the purelin-type function.

[0034] For a backpropagation neural network, different initial weights of the network result in different training outcomes each time, which is caused by the large number of local minimum points on the error surface. In the present invention, the initial value is a random number between (-1, 1).

[0035] Step S403: Perform normalization preprocessing on the sample data, including normalizing the input samples and cleaning outliers;

[0036] In order to uniformly organize the data and make the prediction results more reasonable and credible, before using the actual engineering data as samples to train the neural network, since the numerical values of the data vary greatly, it is likely to increase the difficulty of network training and reduce the accuracy. To eliminate the influence of the dimension between variables, the variables should be subjected to a normalization transformation. The sample data can be used by the model only after being preprocessed, and the performance of the model can be exerted. In the present invention, the original data is normalized according to the following method:

[0037]

[0038] In the formula, X is the original data; X max 、X min are the maximum and minimum values of the original data; T is the target data; T max 、T min are the maximum and minimum values of the target data, taking 0.8 and 0.2 respectively.

[0039] Step S404: Use the normalized sample data to train the neural network using the error backpropagation algorithm, and optimize the weights through multiple iterations to form a neural network model for predicting gas blasting parameters;

[0040] The backpropagation learning algorithm for the forward multi-layer neural network error is used to train the neural network, and the training function uses the Bayesian regularization training function trainbr. In this algorithm, the parameters to be set include the network expected error, learning rate, and number of training times. When the error is less than or equal to the set value, the weights and thresholds of the network are saved and the training is ended.

[0041] The trained neural network model can be used to predict the reasonable optimal burden W O and the reasonable optimal specific charge e O .

[0042] Step S5: Based on the neural network model trained in step S4, use this model to optimize the design of the hole layout parameters for gas blasting, including the reasonable setting of the burden, hole spacing, row spacing, and hole depth.

[0043] Further, step S5 specifically includes the following steps:

[0044] Step S501: Obtain the rock mass parameters of the blasting area, including the point load strength I of the rock blocks on the bench slope s(50) , the wave impedance ρV of the rock blocks pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of the main structural plane 5 , the inclination angle α of the bench slope, and the bench height H B ;

[0045] Step S502: Determine the hole depth H and the suitable type of fracturer according to the bench height, and calculate the gas blasting energy per unit length λ according to Equation (5); in gas blasting, the fracturer has a fixed type, and the length and blasting energy of each type are fixed. The hole depth H is generally an integer multiple of the fracturer length plus the stemming length, and the hole depth H is less than or close to the bench height H B . After determining the type of fracturer and the hole depth, the gas blasting energy per unit length λ can be determined according to Equation (5);

[0046] Step S503: Input the parameters into the trained neural network model to predict the reasonable optimal burden W O and the reasonable optimal specific charge e O ; Input the six parameters of the point load strength I of the rock blocks s(50) , the wave impedance ρV of the rock blocks pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of the main structural plane 5 , the inclination angle α of the bench slope, and the gas blasting energy per unit length λ into the trained backpropagation neural network to obtain the reasonable burden W O and the reasonable specific charge e O ;

[0047] Step S504: Determine the hole layout parameters and adopt triangular hole layout. Among them, the distance between the first row of holes and the bench edge is W O , and the spacing between holes in the same row is:

[0048]

[0049] In the formula, b is the row spacing. Since gas blasting requires a relatively high free face for rock breaking, generally, in actual blasting, the method of successive row initiation or successive hole initiation is adopted. After the first row of holes is blasted, a free face is created for the second row of holes. Therefore, the row spacing is also set as b = W O ; At the same time, if conditions permit, triangular hole layout should be adopted as much as possible, that is, the drilling holes between adjacent rows are staggered.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention provides a method for predicting gas blasting parameters and designing blast hole layout based on a backpropagation neural network, which can accurately predict blasting parameters, automatically optimize the blast hole layout design, improve the blasting effect, resource utilization efficiency, and reduce human errors. This method has strong adaptability and can effectively improve engineering safety and economic benefits. Description of the Drawings

[0052] Figure 1 Structural diagram of the backpropagation neural network constructed for the embodiment of the present invention;

[0053] Figure 2 Neural network structure and operation process displayed during program operation in the embodiment;

[0054] Figure 3 Variation trend of the mean square error of the neural network with the number of training times in the embodiment. Detailed Implementation Manner

[0055] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0056] The purpose of the present invention is to overcome the deficiencies of the prior art, provide a reasonable parameter prediction model for gas blasting and a blast hole parameter design method based on a backpropagation neural network, determine reasonable blast hole parameters according to rock mass parameters and site conditions, reduce the rock breaking cost, and improve the market competitiveness of gas blasting technology.

[0057] The present invention provides a reasonable parameter prediction model for gas blasting based on a backpropagation neural network, including the following steps:

[0058] Step S1: Obtain the physical and mechanical parameters of the rock mass in the blasting area, including: the point load strength I of the rock block s(50) , the wave impedance ρV of the rock block pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of the main structural plane 5 ;

[0059] In a certain greyish limestone breccia site, the average value I of the point load strength of the rock block s(50) is 2.828 MPa, the average value ρV of the wave impedance of the rock block pr is 11.693×10 6 Pa·s / m, the rock mass integrity index K vis 0.38; the difference between the dip angle of the control structural plane in the site and the dip angle of the bench slope is greater than 10°. According to Table 5.3.2-3 in the "Standard for Engineering Rock Mass Classification" (GB / T 50218-2014), F 3 is taken as 0, then K 5 = F 1 × F 2 × F 3 = 0.

[0060] Step S2: Conduct multiple single-hole gas blasting tests in the same site, measure and record the rock-breaking volume of each test, and determine the optimal burden and other key blasting parameters under the corresponding site conditions, including blasting energy, specific consumption, and gas blasting energy per meter of hole length;

[0061] The bench height in the site is about 10 m, the dip angle is 85°, and a Φ89×4×1200 type disposable fracturing device is used, with a single gas blasting energy of 658.4 kJ. A single-hole blasting test is carried out with a drilling depth of 2.5 m, and 1 fracturing device is placed in each hole with a stemming length of 1.3 m. Then the gas blasting energy per meter of hole length is λ = 548.67 kJ / m. 5 single-hole gas blasting tests are carried out, and the burdens are 0.9 m, 1.1 m, 1.3 m, 1.5 m, and 1.7 m respectively. Among them, the rock-breaking volume corresponding to a burden of 1.5 m is the largest (6.5 m 3 ). Therefore, the reasonable burden in this area is W O = 1.5 m, and the reasonable specific consumption is e O = 101.3 kJ / m 3 .

[0062] Step S3: On the blasting sites under different geological conditions, repeat Step S1 and Step S2, accumulate data through multiple experiments, and statistically analyze the corresponding data and the key blasting parameters obtained from the experiments to establish a gas blasting rock-breaking operation database including different geological conditions;

[0063] Carry out gas blasting rock-breaking operations under different rock mass and site conditions according to the requirements described in Step S1 and Step S2, statistically analyze the corresponding data, and the established sample database is shown in Table 2.

[0064] Table 2 Gas Blasting Sample Database

[0065] Number <![CDATA[I s(50) (MPa)]]> <![CDATA[ρV pr (MPa·s / m)]]> <![CDATA[K v > <![CDATA[K 5 > α(°) λ (kJ / m) <![CDATA[W O (m)]]> <![CDATA[e O (kJ / m 3 )]]> 1 2.83 11.7 0.38 0 85 548.7 1.5 101.3 2 5.92 13.5 0.22 0.8 87 902.7 2.0 150.5 3 7.63 15.6 0.65 0.32 80 366.1 1.5 252.6 4 6.14 16.3 0.74 0 80 902.7 1.1 548.3 5 7.69 13.6 0.55 0.17 80 366.1 1.4 236.5 6 8.32 12.9 0.43 0.17 80 366.1 1.3 284.9 7 10.2 13.5 0.54 0 80 902.7 1.1 532.1 8 11.6 15.6 0.46 0.68 85 902.7 1.6 317.1 9 10.3 15.0 0.48 0.14 90 902.7 1.3 430.1 10 8.95 14.9 0.58 0 78 548.7 1.0 419.3 11 8.36 14.8 0.66 0 82 548.7 1.1 413.9 12 7.55 15.0 0.65 0.08 82 366.1 1.2 295.6 13 9.78 15.8 0.28 0.32 84 366.1 1.5 198.7 14 9.99 11.4 0.48 0 81 902.7 1.1 484.1 15 6.27 12.3 0.37 0.08 78 548.7 1.2 303.4 16 9.53 13.8 0.69 0.3 88 902.7 1.4 400.9

[0066] Step S4: Construct and train a gas blasting parameter prediction model based on the backpropagation neural network. The input training data includes physical and mechanical parameters and experimental data, and the output is the predicted gas blasting parameters. Optimize the neural network structure through the backpropagation algorithm to improve the prediction accuracy, which specifically includes the following steps:

[0067] Step S401: Determine the topological structure of the backpropagation neural network according to the characteristics and correlations of gas blasting parameters, and construct a preliminary neural network model;

[0068] The first 12 samples in Table 2 are used as training samples, and the last 4 samples are used as test samples to test the prediction accuracy of the neural network. As Figure 1 shown, a backpropagation neural network with 1 input layer, 1 hidden layer, and 1 output layer is established; the input layer contains 6 nodes, corresponding to the point load strength I of rock blocks s(50) , the wave impedance ρV of rock blocks pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of main structural planes 5 , the slope angle α of the bench slope, and the gas explosion energy λ per meter of extension, these 6 parameters; the output layer contains 2 nodes, corresponding to the reasonable burden W O and the reasonable specific charge e O ; the number of training samples is 12, so the number of hidden layer nodes is set to 12×2 + 1 = 25.

[0069] Step S402: Select an activation function suitable for gas blasting parameter prediction and set the initial weights;

[0070] The activation function of the middle layer neurons of the backpropagation neural network model adopts the sigmoid logarithmic function logsig, and the activation function of the last layer neurons adopts the purelin function. The initial weights of the network adopt random numbers between (-1, 1).

[0071] Step S403: Perform normalized preprocessing on the sample data, including normalizing the input samples and cleaning outliers;

[0072] In order to eliminate the influence of dimensions existing between variables, the variables should be subjected to a normalization transformation. The sample data can be used by the model and the performance of the model can be exerted only after preprocessing. The original data is normalized according to the method in Equation (6).

[0073] Step S404: Use the normalized sample data to train the neural network using the error backpropagation algorithm, and optimize the weights through multiple iterations to form a neural network model for gas blasting parameter prediction;

[0074] The backpropagation learning algorithm for the error of the forward multi-layer neural network is used to train the neural network, and the training function adopts the Bayesian regularization training function trainbr. In this algorithm, the parameters to be set include the network expected error, the learning rate, and the number of training times, which are set to 1.0×10 -10 , 0.01, and 10 11When the error is less than or equal to the set value, save the weights and thresholds of the network and end the training. The network structure and running process displayed during the program operation are as Figure 2 shown. During the training process, the mean square error gradually converges as the number of training times increases, as Figure 3 shown.

[0075] The prediction accuracy of the neural network model was not tested. The measured results and prediction results of the last 4 samples in Table 2 are listed in Table 3. It can be seen from Table 3 that the prediction accuracy of this model is relatively high.

[0076] Table 3 Comparison between prediction results and measured results

[0077]

[0078]

[0079] The trained neural network can be used to predict the reasonable optimal burden W O and reasonable optimal specific charge e O .

[0080] Step S5: Based on the neural network model trained in Step S4, use this model to optimize the design of the hole layout parameters for gas blasting, including the reasonable setting of the burden, hole spacing, row spacing, and hole depth.

[0081] Step S501: Obtain the rock mass parameters of the blasting area, including the point load strength I s(50) of the rock blocks on the bench slope surface, the wave impedance ρV pr of the rock blocks, the rock mass integrity index K v , the influence coefficient K 5 of the occurrence of the main structural planes, the bench slope angle α, and the bench height H B ;

[0082] Taking a gas blasting site in a certain mining area as an example, through experiments and measurements, it is obtained that: the point load strength I s(50) of the rock blocks in the blasting area is 6.0 MPa, the wave impedance ρV pr of the rock blocks is 12.8 MPa·s / m, the rock mass integrity index K v is 0.38, the influence coefficient K 5 of the occurrence of the main structural planes is 0.14, the bench slope angle α is about 85°, and the bench height H B is about 10 m.

[0083] Step S502: Determine the hole depth H and the suitable type of fracturer according to the bench height, and calculate the gas blasting energy per meter λ according to Equation (5);

[0084] In gas blasting, the fracturer has a fixed model, and the length and blasting energy of each model are fixed. The hole depth H is generally an integral multiple of the fracturer length plus the stemming length, and the hole depth H is less than or close to the bench height H B ., so the hole depth of this site is set to 5 m, and a Φ89×4×1200 type disposable fracturer is used. 3 fracturers are placed in each hole, and the stemming length is 1.4 m. Then the gas blasting energy λ per meter along the drilling direction is 548.7 kJ / m.

[0085] Step S503: Input the parameters into the trained neural network model to predict the reasonable optimal burden W O and the reasonable optimal specific charge e O ;

[0086] Input the point load strength I of the rock block s(50) , the wave impedance ρV of the rock block pr , the rock mass integrity index K v , the influence coefficient K of the occurrence of the main structural plane 5 , the bench slope angle α, and the gas blasting energy λ per meter, a total of 6 parameters, into the trained BP neural network to obtain the reasonable burden W O and the reasonable specific charge e O are 1.48 m and 200.8 kJ / m respectively 3 .

[0087] Step S504: Determine the hole layout parameters;

[0088] The distance from the first row of holes to the bench edge is the burden, which is set to W O , that is, 1.5 m. Since gas blasting for rock breaking has high requirements for the free face, generally in actual blasting, the method of row-by-row initiation or hole-by-hole initiation is adopted. After the first row of holes is blasted, a free face is created for the second row of holes. Therefore, the row spacing is also set to b = W O = 1.5 m, and the spacing between holes in the same row is calculated according to Equation (7), about 1.8 m.

Claims

1. A gas blasting parameter prediction and hole layout design method based on back propagation neural network, characterized in that: The following steps are involved: Step S1: Obtain the physical and mechanical parameters of the rock mass in the blasting area, including: the point load strength I of the rock mass s(50) , wave impedance of rock mass ρV pr , rock mass integrity index K v , the main structural surface attitude influence coefficient K5; Step S2: Conduct multiple single-hole gas blasting tests in the same site, measure and record the rock breaking volume of each test, and determine the optimal resistance line and other key blasting parameters under the corresponding site conditions, including blasting energy, unit consumption and gas blasting energy per hole meter; Step S3: Repeating steps S1 and S2 at blasting sites under different geological conditions, accumulating data through multiple experiments, and statistically analyzing the corresponding data and key blasting parameters obtained from the experiments, so as to establish a gas blasting rock breaking operation database containing different geological conditions; Step S4: construct and train a gas explosion parameter prediction model based on a back propagation neural network, input training data including physical and mechanical parameters and experimental data, output predicted gas explosion parameters, and optimize the neural network structure through a back propagation algorithm to improve prediction accuracy; Step S5: Based on the neural network model trained in step S4, the model is used to optimize the hole layout parameters of gas blasting, including reasonable settings of resistance line, hole layout spacing, row spacing, and hole depth.

2. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 1, characterized in that: In step S1, ρ is the density of the rock mass, and the rock mass integrity index K is v Calculate according to formula (1): K v =(V pm / V pr ) 2 (1) Where V pm is the elastic longitudinal wave velocity of the rock mass, V pr is the elastic longitudinal wave velocity of the rock block; the main structural surface attitude influence coefficient K5 is calculated according to formula (2): K5=F1×F2×F3 (2) Where F1 represents the influence of the angle between the structural surface inclination and the fracture surface inclination, F2 represents the influence of the structural surface inclination, and F3 represents the influence of the difference between the structural surface inclination and the slope surface inclination.

3. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 1, characterized in that: Step S2 includes the following steps: Step S201: Conduct multiple single-hole gas blasting rock breaking tests at the same site. The test conditions are kept the same each time, and only the resistance line W is adjusted. The rock breaking volume V of each test is counted to determine the maximum rock breaking volume V. O The corresponding optimal resistance line W under the site conditions O Step S202: Calculate the explosion energy E of the single hole test according to the explosion energy calculation formula of the compressed gas and water vapor container g : Where k is the adiabatic index of carbon dioxide; p is the burst pressure; p atm is the atmospheric pressure; V0 is the initial volume of gas in a single explosion; Step S203: Based on the blasting energy E g Calculate the optimal unit consumption e corresponding to the rock mass properties and site conditions O : And the gas explosion energy λ per linear meter along the borehole: Where H is the hole depth and T is the plugging length.

4. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 1, characterized in that: Step S4 includes the following steps: Step S401: Determine the topological structure of the back propagation neural network according to the characteristics of the gas explosion parameters and their correlation, and construct a preliminary neural network model; Step S402: Select an activation function suitable for gas explosion parameter prediction and set initial weights; Step S403: performing normalization preprocessing on the sample data, including normalization processing on the input samples and cleaning of outliers; Step S404: using the normalized sample data, adopting the error back propagation algorithm to train the neural network, and optimizing the weights through multiple iterations to form a neural network model for gas explosion parameter prediction.

5. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 5, characterized in that: In step S401, the preliminary neural network model includes an input layer, a hidden layer and an output layer, wherein the hidden layer network structure is designed as follows: The rock point load strength I s(50) 、Rock wave impedance ρV pr , rock mass integrity index K v , the main structural surface occurrence influence coefficient K5, the step slope inclination angle α, and the gas explosion energy per linear meter λ are the input indicators, which correspond to the neurons of the input layer respectively; with a reasonable optimal resistance line W O and reasonable optimal unit consumption O are output indicators, corresponding to the neurons in the output layer respectively; the number of neurons in the hidden layer is dynamically set according to the number of training samples.

6. The method for gas explosion parameter prediction and hole layout design based on back propagation neural network according to claim 5, characterized in that: In step S402, the activation function of the hidden layer neurons adopts the sigmoid logarithmic function logsig, and the activation function of the output layer neurons adopts the purelin linear function; the initial weights are generated by random numbers in the range of (-1, 1).

7. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 5, characterized in that: In step S403, the original data is normalized according to the following method: In the formula, X is the original data; X max , X min is the maximum and minimum value of the original data; T is the target data; T max , T min are the maximum and minimum values ​​of the target data.

8. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 5, characterized in that: In step S404, the forward multi-layer neural network error back propagation learning algorithm is used to train the neural network. The training function uses the Bayesian regularization training function trainbr, and the network expected error, learning rate and number of training times are set as training parameters; when the error is less than or equal to the set value, the network weights and thresholds are saved and the training ends.

9. The method for predicting gas explosion parameters and designing holes based on back propagation neural network according to claim 1, characterized in that: Step S5 specifically includes the following steps: Step S501: Obtain rock mass parameters in the blasting area, including the point load strength I of the step slope rock mass s(50) 、Rock wave impedance ρV pr , rock mass integrity index K v , main structural surface attitude influence coefficient K5, step slope inclination angle α, step height H B ; Step S502: Determine the hole depth H and the appropriate fracturing device model according to the step height, and calculate the gas explosion energy λ per linear meter according to formula (5); Step S503: Input the parameters into the trained neural network model to predict the reasonable optimal resistance line W O and reasonable optimal unit consumption O ; Step S504: Determine hole arrangement parameters and adopt a triangular hole arrangement, where the distance between the first row of holes and the step edge is W O , the spacing between holes in the same row is: Where b is the row spacing, b = W O According to the design requirements, the holes are arranged in a triangular pattern.

Images