Post-blasting rock average lumpiness prediction method considering rock in-situ block size

Abstract

The invention discloses a post-blasting rock average lumpiness prediction method considering the rock mass in-situ block size, which comprises the following steps: investigating the distribution characteristics of a rock mass structural surface in a to-be-blasted area, and quantitatively representing the development characteristics of the structural surface by adopting a distribution function; a structural surface is generated in the numerical model according to a Monte Carlo method, and a rock mass three-dimensional numerical model of the to-be-blasted area is constructed; analyzing the condition that an area to be blasted in the rock mass three-dimensional numerical model is segmented by a structural surface, taking independent rock blocks completely segmented by the structural surface as in-situ blocks, counting the sizes of the in-situ blocks, and obtaining in-situ block parameters based on the sizes of all the in-situ blocks; constructing a back-propagation neural network and a multiple linear regression equation, substituting all blasting parameters including in-situ block parameters into the back-propagation neural network, and determining parameters having significant influence on the rock lumpiness after blasting by adopting a parameter weight analysis method; and substituting the parameters which have obvious influence on the rock lumpiness after blasting into the existing blasting lumpiness prediction model, and constructing an average rock lumpiness prediction model after blasting.

Description

Technical Field

[0001] This invention belongs to the field of rock modeling technology, and in particular relates to a method for predicting the average block size of rock after blasting, taking into account the size of in-situ rock blocks. Background Technology

[0002] In the field of rock blasting engineering, the initial structural planes in the rock mass naturally divide it into rock blocks of varying sizes. These blocks are called in-situ blasting blocks. The rock blasting process is essentially a re-fragmentation process achieved by using explosive energy on the weak surfaces (such as structural planes) of the rock mass. The core is to transform the rock distribution of the in-situ blocks into the final distribution of the separated rock blocks. Currently, the industry widely uses blasting block size characteristic size prediction formulas (such as the Kuznetsov prediction equation and the Kuz-Ram model) to estimate the average block size of the rock after blasting, in order to guide the design of on-site blasting schemes. At the same time, considering that there are many parameters affecting the blasting block size distribution (such as aperture, step height, explosive consumption, rock mechanical properties, etc.), and that there are complex interactions between these factors, some studies have introduced deep learning methods such as neural networks (such as backpropagation neural networks) to attempt to quantify the weights of the parameters affecting blasting block size, in order to screen out the parameters that have a more significant impact on the blasting results and provide technical support for optimizing the prediction model.

[0003] Existing technologies suffer from two key shortcomings that limit the accuracy and practical applicability of predicting the average rock block size after blasting. Firstly, widely used blasting block size prediction formulas (including the Kuznetsov prediction equation and the Kuz-Ram model) do not include the "in-situ block size" as a core parameter. However, the in-situ block size, as a key indicator of the initial fragmentation state of the rock mass, directly determines the starting point and difficulty of blasting re-fragmentation. Its absence leads to a significant deviation between the predicted results and the actual block size on site, resulting in low accuracy. Secondly, while some studies have analyzed the weights of blasting parameters using methods such as neural networks, these studies have neither supplemented the core parameter of in-situ block size nor constructed specific prediction model formulas that can be directly applied to on-site engineering based on the parameter analysis results; they remain only at the parameter selection stage. Furthermore, the determination of some parameters in traditional prediction models (such as the rock factor A in the Kuznetsov formula) relies on empirical judgment, lacking clear theoretical and data support, further reducing the reliability and universality of the models and failing to meet the needs of precise on-site blasting design. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method for predicting the average block size of blasted rock that considers the size of in-situ rock blocks, thereby resolving the issues present in the prior art.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for predicting the average block size of blasted rock considering the in-situ block size of the rock mass, comprising:

[0006] The distribution characteristics of rock mass structural planes in the area to be blasted were investigated, and the development characteristics of the structural planes were quantitatively characterized by the distribution function.

[0007] Based on the distribution and development characteristics, the structural surfaces are generated in the numerical model using the Monte Carlo method to construct a three-dimensional numerical model of the rock mass in the area to be blasted.

[0008] The analysis shows that the area to be blasted in the three-dimensional numerical model of the rock mass is divided by the structural surface. The independent rock blocks that are completely divided by the structural surface are taken as in-situ blocks and their dimensions are counted. The in-situ block parameters are obtained based on the dimensions of all in-situ blocks.

[0009] A backpropagation neural network and a multiple linear regression equation were constructed. All blasting parameters, including the in-situ block parameters, were substituted into the backpropagation neural network. The parameter weight analysis method was used to determine the parameters that have a significant impact on the rock block size after blasting.

[0010] By substituting parameters that significantly affect the size of rock blocks after blasting into the existing blasting block size prediction model, a prediction model for the average size of rock blocks after blasting is constructed.

[0011] Preferably, the process of investigating the distribution characteristics of rock mass structural planes in the area to be blasted is as follows:

[0012] Multi-angle photographs were taken of the free face of the area to be blasted, and the dip, dip angle, trace length and location information of the structural surface were obtained in conjunction with the geological exploration report of the area to be blasted.

[0013] Preferably, the process of using a distribution function to quantify the development characteristics of the structural surface is as follows:

[0014] All structural surfaces with similar dip direction, dip angle, and trace length within the rock mass area to be blasted are divided into a set of structural surfaces. Statistical analysis is performed on the dip direction and dip angle of each set of structural surfaces, and a probability distribution function of the attitude of the set of structural surfaces is obtained by fitting. The development characteristics of the structural surfaces are quantitatively characterized by the probability distribution function.

[0015] Preferably, the process of generating the structural surface in the numerical model using the Monte Carlo method based on the distribution and development characteristics is as follows:

[0016] The dip, dip angle, spacing, and trace length of the structural surface are used as random variables. The distribution form of each random variable is determined based on the distribution function. A sequence of structural surfaces conforming to the distribution form is generated by random number generation. The structural surface is generated in the numerical model using the sequence of structural surfaces.

[0017] Preferably, the process of obtaining in-situ block parameters based on the dimensions of all in-situ blocks is as follows:

[0018] The dimensions of all the in-situ blocks are statistically analyzed and an in-situ block size distribution curve is generated. The average value of all in-situ block dimensions in the in-situ block size distribution curve is calculated, and the average value is used as the in-situ block parameter.

[0019] Preferably, all blasting parameters include borehole diameter, step height, borehole over-depth, spacing, row spacing, plugging length, specific parameters, number of boreholes, total explosive mass, explosive consumption per unit volume, uniaxial compressive strength of rock, elastic modulus, and the parameters of the in-situ block. The specific parameters include S / B, T / B, H / B, J / B, and B / D, where S is the borehole spacing, B is the borehole row spacing, T is the plugging length, H is the step height, J is the borehole over-depth, and D is the borehole diameter.

[0020] Preferably, the process of constructing the backpropagation neural network is as follows:

[0021] Using all the blasting parameters as input layer units and the average block size of the rock after blasting as output layer units, the number of hidden layer units is set and the backpropagation neural network is trained until the backpropagation neural network meets the preset accuracy requirements.

[0022] The process of constructing the multiple linear regression equation is as follows: using all the blasting parameters as independent variables and the average block size of the rock after blasting as the dependent variable, the multiple linear regression equation is obtained through linear fitting, and the fitting accuracy of the multiple linear regression equation is verified.

[0023] Preferably, the process of determining the parameters that significantly affect the size of the rock blocks after blasting using the parameter weighting analysis method is as follows:

[0024] Using the average input value of all blasting parameters as a baseline, calculate the gradient metric for each of the blasting parameters;

[0025] The importance index of each of the blasting parameters is obtained based on the gradient metric.

[0026] Blasting parameters whose importance index is greater than a preset threshold are selected as parameters that have a significant impact on the size of the rock blocks after blasting.

[0027] Preferably, the formula for calculating the gradient metric for each input parameter is:

[0028] ;

[0029] Among them, g a For each input parameter, a is the gradient metric, where 'a' is the parameter to be analyzed, 'L' represents the number of changes in that input parameter within a specific range, and 'm' represents the total number of input parameters. This represents the sensitivity response of this parameter.

[0030] Preferably, the existing blasting block size prediction model is the Kuznetsov prediction model, and the formula is:

[0031] ;

[0032] Where, x m Q is the average size of the blasted fragments, A is the rock factor, and Q is the average size of the blasted fragments. e The TNT explosive weight in a single borehole is given, and S, B, and H represent the borehole spacing, row spacing, and borehole depth, respectively.

[0033] Substituting the parameters that significantly affect the size of the rock fragments after blasting into the Kuznetsov prediction model, and correcting the coefficients in the Kuznetsov prediction model, we obtain the prediction model for the average size of the rock fragments after blasting, as shown in the formula:

[0034] ;

[0035] Where T is the blockage length, UCS is the uniaxial compressive strength, and X is the uniaxial compressive strength. B This refers to the size of the in-situ block.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] This invention fills the gap in existing blasting block size prediction models that ignore the size of in-situ rock blocks. By investigating the distribution characteristics of rock mass structural surfaces, constructing a three-dimensional numerical model, and statistically analyzing the size of in-situ blocks to obtain in-situ block parameters, the prediction model incorporates the initial rock mass fragmentation state parameters that have a decisive influence on the size of the rock blocks after blasting. This better reflects the physical nature of rock blasting, which involves "in-situ block re-fragmentation," and improves the adaptability of the prediction model to actual blasting scenarios.

[0038] This invention utilizes a backpropagation neural network and a multiple linear regression equation combined with parameter weighting analysis to accurately select parameters that significantly affect the size of rock blocks after blasting from a large number of blasting parameters. This avoids the subjectivity and bias caused by traditional prediction methods that rely on experience to select parameters, and provides core parameter support with statistical and engineering significance for the subsequent construction of prediction models, ensuring the scientific and rational nature of parameter selection.

[0039] This invention incorporates selected significant influencing parameters (including in-situ block parameters) into the existing blasting block size prediction model, thereby optimizing and upgrading the traditional model. It retains the existing model's application basis and operational convenience in engineering sites, while significantly improving the accuracy of the predicted average block size of rock after blasting by supplementing key parameters and optimizing model input. This provides a more reliable theoretical basis for on-site blasting scheme design and blasting effect evaluation, and reduces the waste of engineering resources or safety risks caused by prediction deviations. Attached Figure Description

[0040] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0041] Figure 1 This is a flowchart of the method for predicting the average block size of rock after blasting according to an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the orientation of the first set of structural planes in an embodiment of the present invention;

[0043] Figure 3 This is a realistic three-dimensional numerical model of the area to be blasted, as described in this embodiment of the invention.

[0044] Figure 4 This is a schematic diagram of the in-situ block size distribution curve according to an embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of the backpropagation neural network architecture according to an embodiment of the present invention;

[0046] Figure 6 This is a schematic diagram of the weight analysis results in an embodiment of the present invention;

[0047] Figure 7 This is a schematic diagram comparing the prediction results of the new model and the Kuznetsov model in an embodiment of the present invention. Detailed Implementation

[0048] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0049] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides a method for predicting the average block size of blasted rock considering the in-situ block size of the rock mass, including:

[0052] S1. Investigate the distribution characteristics of rock mass structural planes in the area to be blasted, and use distribution functions to quantitatively characterize the development characteristics of the structural planes;

[0053] Furthermore, the process of investigating the distribution characteristics of rock mass structural planes in the area to be blasted is as follows:

[0054] Multi-angle photographs were taken of the free face of the area to be blasted, and the dip, dip angle, trace length and location information of the structural surface were obtained in conjunction with the geological exploration report of the area to be blasted.

[0055] Furthermore, the process of using a distribution function to quantify the developmental characteristics of the structural surface is as follows:

[0056] All structural surfaces with similar dip direction, dip angle, and trace length within the rock mass area to be blasted are divided into a set of structural surfaces. Statistical analysis is performed on the dip direction and dip angle of each set of structural surfaces, and a probability distribution function of the attitude of the set of structural surfaces is obtained by fitting. The development characteristics of the structural surfaces are quantitatively characterized by the probability distribution function.

[0057] Specifically, all structural surfaces within the rock mass area to be blasted that have similar dip direction, dip angle, and trace length are referred to as a structural surface set. Based on this grouping, the geometric parameters and probabilistic models of this structural surface set can be studied. The collected primary joint structural surface information is statistically grouped to determine the dominant orientation of the structural surfaces and to classify the number of structural surface groups.

[0058] Fitting the probability model of structural surface attitude: After completing the statistics of the structural surface set, statistical analysis is performed on the dip and dip angle of each set of structural surfaces. Figure 2 List the histograms of the dip distribution and the probability density fitting curves for 12 sets of structural surfaces. Figure 2 It can be seen that the dip and tilt angle of structure surface set 1 follow a uniform distribution.

[0059] S2. Based on the distribution and development characteristics, the structural surface is generated in the numerical model using the Monte Carlo method to construct a three-dimensional numerical model of the rock mass in the area to be blasted.

[0060] Furthermore, based on the aforementioned distribution and developmental characteristics, the process of generating the structural surface in the numerical model using the Monte Carlo method is as follows:

[0061] The dip, dip angle, spacing, and trace length of the structural surface are used as random variables. The distribution form of each random variable is determined based on the distribution function. A sequence of structural surfaces conforming to the distribution form is generated by random number generation. The structural surface is generated in the numerical model using the sequence of structural surfaces.

[0062] Specifically, when the probability of a random variable occurring conforms to a specific distribution, a sequence of variables with a similar distribution can be generated using random number generation. This inverse process of sampling statistics is called the Monte Carlo method. By substituting the dip, dip angle, spacing, and trace length of the structural surface as inputs, after determining the distribution form of the structural surface attitude probability model, a series of structural surface models conforming to the statistical distribution form are obtained by randomly generating them within a range.

[0063]

[0064] In the formula, v, j, t and c represent the dip direction, dip angle, spacing and trace length of the generated structure surface, respectively.

[0065] By substituting the generated structural surface information into numerical simulation software, a realistic three-dimensional numerical model of the area to be blasted can be constructed, such as... Figure 3 As shown.

[0066] S3. Analyze the situation where the area to be blasted in the three-dimensional numerical model of the rock mass is divided by the structural surface. Take the independent rock blocks that are completely divided by the structural surface as in-situ blocks and count their size. Obtain the in-situ block parameters based on the size of all in-situ blocks.

[0067] Furthermore, the process of obtaining in-situ block parameters based on the dimensions of all in-situ blocks is as follows:

[0068] The dimensions of all the in-situ blocks are statistically analyzed and an in-situ block size distribution curve is generated. The average value of all in-situ block dimensions in the in-situ block size distribution curve is calculated, and the average value is used as the in-situ block parameter.

[0069] Furthermore, all blasting parameters include borehole diameter, step height, borehole over-depth, spacing, row spacing, plugging length, specific parameters, number of boreholes, total explosive mass, explosive consumption per unit volume, uniaxial compressive strength of rock, elastic modulus, and the in-situ block parameters. The specific parameters include S / B, T / B, H / B, J / B, and B / D, where S is the borehole spacing, B is the borehole row spacing, T is the plugging length, H is the step height, J is the borehole over-depth, and D is the borehole diameter.

[0070] Specifically, after the structural plane is segmented, each independent rock block completely segmented by the structural plane is considered a single in-situ block, and its size is statistically analyzed. All in-situ blocks are statistically analyzed to obtain the in-situ block size distribution curve, and the average size of the in-situ blocks is calculated as the in-situ block parameter for the area to be blasted. The in-situ block size distribution curve is shown below. Figure 4 As shown.

[0071] Numerous factors influence the size of blasted blocks, including not only borehole diameter (D), step height (H), borehole depth (J), spacing (Space), burden spacing (B), and plugging length (m), but also specific parameters (S / B), (T / B), (H / B), (J / B), and (B / D). These factors also include blasting design parameters such as the number of boreholes (NH), total explosive mass (Qe), and explosive consumption per unit volume (Pf), as well as uniaxial compressive strength of rock (UCS), elastic modulus (E), and in-situ block size (X). B There are a total of 17 parameters.

[0072] S4. Construct a backpropagation neural network and a multiple linear regression equation. Substitute all blasting parameters, including the in-situ block parameters, into the backpropagation neural network and use the parameter weight analysis method to determine the parameters that have a significant impact on the rock block size after blasting.

[0073] Furthermore, the process of constructing the backpropagation neural network is as follows:

[0074] Using all the blasting parameters as input layer units and the average block size of the rock after blasting as output layer units, the number of hidden layer units is set and the backpropagation neural network is trained until the backpropagation neural network meets the preset accuracy requirements.

[0075] The process of constructing the multiple linear regression equation is as follows: using all the blasting parameters as independent variables and the average block size of the rock after blasting as the dependent variable, the multiple linear regression equation is obtained through linear fitting, and the fitting accuracy of the multiple linear regression equation is verified.

[0076] Furthermore, the process of determining the parameters that significantly affect the size of the rock fragments after blasting using the parameter weighting analysis method is as follows:

[0077] Using the average input value of all blasting parameters as a baseline, calculate the gradient metric for each of the blasting parameters;

[0078] The importance index of each of the blasting parameters is obtained based on the gradient metric.

[0079] Blasting parameters whose importance index is greater than a preset threshold are selected as parameters that have a significant impact on the size of the rock blocks after blasting.

[0080] Specifically, in order to determine the blasting parameters that have a significant impact in order to construct a blasting block size prediction model, a backpropagation neural network and a multiple linear regression equation are constructed. All blasting parameters, including in-situ block parameters, are substituted into the backpropagation neural network, and the parameter weight analysis method is used to determine the four parameters that have the greatest impact on blasting.

[0081] Artificial Neural Networks (ANNs) model data by simulating how the brain processes it, and are widely used to handle complex causal relationships. A standard ANN consists of three parts: the input layer receives the independent variables, the hidden layers perform calculations on these independent variables, and the predicted target value is displayed in the output layer. The number of nodes in the input layer corresponds to the number of input parameters, the output parameters are the target prediction parameters, and the number of hidden layers and the number of nodes in each layer are parameters; therefore, the optimal hidden layer structure needs to be determined through tuning for different problems. Backpropagation Neural Networks (BPNNs) are chosen to represent neural network technology because previous research has shown that this method can better predict the average size of rock fragments after blasting. BPNNs use the backpropagation algorithm to repeatedly correct the weights and biases of each connection unit, exhibiting stronger optimization computational capabilities and greater stability compared to ordinary feedforward neural networks.

[0082] Seventeen input parameters obtained from the blasting site were used as input layer units, with five hidden layer units and the output layer unit being the average size of the blasted block. A backpropagation neural network was then constructed and trained. Figure 5 As shown.

[0083] Multiple linear regression expresses the relationship between the dependent and output variables in the form of a linear equation. Due to its simple structure and clear results, it is widely used in regression tasks.

[0084] Calculate the mean absolute error (MAE) and correlation coefficient (R). 2 The root mean square error (RMSE) is used to evaluate whether the backpropagation neural network and the multiple linear regression equation have been successfully trained, i.e., to evaluate whether the model can be used to analyze the relationship between blasting parameters and blasting block size.

[0085]

[0086]

[0087]

[0088] In the formula, x mea and x pre These are the actual block size and the predicted block size, respectively. and These represent the average actual block size and the average predicted block size, respectively. n is the number of data points. The final backpropagation neural network achieved a MAE of 0.04, an RMSE of 0.04, and an R... 2 The value was 0.881; while the MAE of the multiple linear regression equation was 0.026, the RMSE was 0.0346, and the R-squared value was 0.881. 2The value is 0.920. 2 The values ​​are all greater than 0.85, indicating that the backpropagation neural network and the multiple linear regression equation were successfully trained. Furthermore, the R-squared value of the multiple linear regression equation is... 2 A value higher than 0.9 indicates a linear relationship between blasting parameters and blasting block size, meaning that when constructing the model, the significant blasting parameters can be substituted into the model using linear regression.

[0089] The parameter importance index I for backpropagation neural network and multiple linear regression was calculated using parameter weighting analysis. a The higher the importance index, the more important it is. Parameter weighting analysis uniformly selects each input parameter within a range, while the remaining parameters are fixed to a given baseline value.

[0090] Using the average input blasting parameter value as a benchmark, calculate the gradient metric (g) for each input parameter. a ):

[0091]

[0092] The importance of each parameter was then analyzed to obtain the parameter importance index I. a :

[0093]

[0094] In the formula, 'a' is the parameter to be analyzed, 'L' represents the number of changes in the input parameter within a specific range (taken as 10), and 'm' represents the total number of input parameters. This is the sensitivity response of this parameter. The calculation results are as follows: Figure 6 As shown.

[0095] It is evident that the parameters with significant influence include uniaxial compressive strength (UCS) and in-situ block parameter X. B Burden spacing, Space spacing, and blockage length T.

[0096] S5. Substitute the parameters that have a significant impact on the size of the rock after blasting into the existing blasting block size prediction model to construct a prediction model for the average size of the rock after blasting.

[0097] Furthermore, the existing blasting block size prediction model is the Kuznetsov prediction model. Kuznetsov proposed a formula after statistically analyzing the post-blasting block size results, which is the initial Kuznetsov prediction model:

[0098]

[0099] In the formula: x mQ is the average size of the blasted fragments, and A is the rock factor, which is 7 for medium-hard rock, 10 for hard rock, and 13 for micro-fractured hard rock. e The value represents the weight (kg) of TNT explosive in a single borehole. S, B, and H represent the borehole spacing, row spacing, and borehole depth, respectively.

[0100] The method for determining rock factor A is ambiguous, leading to low model accuracy. Therefore, when constructing a new blasting block size prediction model, the coefficient A is modified, while the subsequent parts are retained. The new model form is:

[0101]

[0102] The formula does not consider the uniaxial compressive strength UCS and the in-situ block size X, which have significant effects. B Given the blockage length T, the coefficient k is therefore a coefficient related to these three factors.

[0103] Based on the results of the multiple linear regression analysis, the in-situ block parameter X can be determined. B The blockage length coefficient exhibits a linear regression relationship with the post-blast blockage size. The uniaxial compressive strength is then uniformly defined as parameter RT (rock mass and blockage length parameter). The formula then takes the form:

[0104]

[0105] In the formula, RT includes two parameters: the blockage length coefficient TF and the UCS. The equation for calculating the blockage length coefficient TF is defined as TF = aT + b, and the above equation is defined as follows:

[0106]

[0107] In the formula, the unknowns are k1, a, and b. Substituting the actual blasting parameters and post-blast block size obtained on-site into the above formula, a new blast block size prediction model can be obtained as follows:

[0108]

[0109] On the one hand, the new model retains the form of the Kuznetsov prediction model widely used in the field, meaning that it is applicable to the blasting block size prediction scenarios of the Kuznetsov prediction model and will certainly be applicable to the new model as well. On the other hand, the parameters in the new model are obtained through theoretical analysis using the parameter weighting method, ensuring the model's effectiveness and scientific rigor, thereby improving its accuracy. A comparison of the results of the new model and the Kuznetsov prediction model is shown below. Figure 7 As shown, the new model has a strong predictive ability.

[0110] The beneficial effects of this embodiment:

[0111] This embodiment combines field investigation and numerical modeling to obtain the average size of in-situ rock blocks; based on the results of backpropagation neural network and multiple linear regression equation, it clarifies that the size of in-situ blocks and the average size of rock blocks after blasting exhibit a linear relationship; based on existing blasting block size prediction models, this embodiment constructs a prediction model formula for the average size of blasted rock blocks that can be used in the field, taking into account parameters such as in-situ blocks and plugging length.

[0112] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the average block size of blasted rock considering the in-situ block size of the rock mass, characterized in that, Includes the following steps: The distribution characteristics of rock mass structural planes in the area to be blasted were investigated, and the development characteristics of the structural planes were quantitatively characterized by the distribution function. Based on the distribution and development characteristics, the structural surfaces are generated in the numerical model using the Monte Carlo method to construct a three-dimensional numerical model of the rock mass in the area to be blasted. The analysis shows that the area to be blasted in the three-dimensional numerical model of the rock mass is divided by the structural surface. The independent rock blocks that are completely divided by the structural surface are taken as in-situ blocks and their dimensions are counted. The in-situ block parameters are obtained based on the dimensions of all in-situ blocks. A backpropagation neural network and a multiple linear regression equation were constructed. All blasting parameters, including the in-situ block parameters, were substituted into the backpropagation neural network. The parameter weight analysis method was used to determine the parameters that have a significant impact on the rock block size after blasting. By substituting parameters that significantly affect the size of rock blocks after blasting into the existing blasting block size prediction model, a prediction model for the average size of rock blocks after blasting is constructed.

2. The method according to claim 1, characterized in that, The process of investigating the distribution characteristics of rock mass structural planes in the area to be blasted is as follows: Multi-angle photographs were taken of the free face of the area to be blasted, and the dip, dip angle, trace length and location information of the structural surface were obtained in conjunction with the geological exploration report of the area to be blasted.

3. The method according to claim 2, characterized in that, The process of using a distribution function to quantify the developmental characteristics of the structural surface is as follows: All structural surfaces with similar dip direction, dip angle, and trace length within the rock mass area to be blasted are divided into a set of structural surfaces. Statistical analysis is performed on the dip direction and dip angle of each set of structural surfaces, and a probability distribution function of the attitude of the set of structural surfaces is obtained by fitting. The development characteristics of the structural surfaces are quantitatively characterized by the probability distribution function.

4. The method according to claim 1, characterized in that, Based on the aforementioned distribution and developmental characteristics, the process of generating the structural surface in the numerical model using the Monte Carlo method is as follows: The dip, dip angle, spacing, and trace length of the structural surface are used as random variables. The distribution form of each random variable is determined based on the distribution function. A sequence of structural surfaces conforming to the distribution form is generated by random number generation. The structural surface is generated in the numerical model using the sequence of structural surfaces.

5. The method according to claim 1, characterized in that, The process of obtaining in-situ block parameters based on the dimensions of all in-situ blocks is as follows: The dimensions of all the in-situ blocks are statistically analyzed and an in-situ block size distribution curve is generated. The average value of all in-situ block dimensions in the in-situ block size distribution curve is calculated, and the average value is used as the in-situ block parameter.

6. The method according to claim 1, characterized in that, All blasting parameters include borehole diameter, step height, borehole over-depth, spacing, row spacing, plugging length, specific parameters, number of boreholes, total explosive mass, explosive consumption per unit volume, uniaxial compressive strength of rock, elastic modulus, and the parameters of the in-situ block. The specific parameters include S / B, T / B, H / B, J / B, and B / D, where S is the borehole spacing, B is the borehole row spacing, T is the plugging length, H is the step height, J is the borehole over-depth, and D is the borehole diameter.

7. The method according to claim 1, characterized in that, The process of constructing a backpropagation neural network is as follows: Using all blasting parameters as input layer units and the average block size of rock after blasting as output layer units, the number of hidden layer units is set and the backpropagation neural network is trained until the backpropagation neural network meets the preset accuracy requirements. The process of constructing the multiple linear regression equation is as follows: using all the blasting parameters as independent variables and the average block size of the rock after blasting as the dependent variable, the multiple linear regression equation is obtained through linear fitting, and the fitting accuracy of the multiple linear regression equation is verified.

8. The method according to claim 1, characterized in that, The process of determining the parameters that significantly affect the size of the rock fragments after blasting using the parametric weighting analysis method is as follows: Using the average input value of all blasting parameters as a baseline, calculate the gradient metric for each of the blasting parameters; The importance index of each of the blasting parameters is obtained based on the gradient metric. Blasting parameters whose importance index is greater than a preset threshold are selected as parameters that have a significant impact on the size of the rock blocks after blasting.

9. The method according to claim 1, characterized in that, The formula for calculating the gradient metric for each input parameter is: ； Among them, g a For each input parameter, a is the gradient metric, where 'a' is the parameter to be analyzed, 'L' represents the number of changes in that input parameter within a specific range, and 'm' represents the total number of input parameters. This represents the sensitivity response of the parameter.

10. The method according to claim 1, characterized in that, The existing blasting block size prediction model is the Kuznetsov prediction model, and the formula is: ； Where, x m Q is the average size of the blasted fragments, A is the rock factor, and Q is the average size of the blasted fragments. e The TNT explosive weight in a single borehole is given, and S, B, and H represent the borehole spacing, row spacing, and borehole depth, respectively. Substituting the parameters that significantly affect the size of the rock fragments after blasting into the Kuznetsov prediction model, and correcting the coefficients in the Kuznetsov prediction model, we obtain the prediction model for the average size of the rock fragments after blasting, as shown in the formula: ； Where T is the blockage length, UCS is the uniaxial compressive strength, and X is the uniaxial compressive strength. B This refers to the size of the in-situ block.

Images