Prediction method of blasting peak particle velocity based on WOA-BP model and numerical simulation

Abstract

The present application provides a blasting peak vibration velocity prediction method based on WOA-BP model and numerical simulation, and relates to the field of engineering blasting technology. The method comprises the following steps: through laboratory impact test, vibration monitoring data set is constructed and BP neural network is trained to obtain sensor angle distortion correction model; online monitoring of blasting vibration is carried out in the target blasting area, and the monitoring data is corrected by using the correction model to obtain the field vibration monitoring data set to train the WOA-BP prediction model; the blasting vibration under different working conditions is simulated by using the bench blasting numerical model, the expanded vibration monitoring data set is constructed, and the trained WOA-BP prediction model is corrected in combination with the field vibration monitoring data set to obtain the final WOA-BP prediction model for predicting the peak vibration velocity and the main frequency corresponding to any blasting working condition parameters. The present application can reliably predict the blasting peak vibration velocity and the main frequency in cold regions by integrating data correction and multi-condition sample generation.

Description

Technical Field

[0001] This invention relates to the field of engineering blasting technology, and in particular to a method for predicting peak blasting velocity based on the WOA-BP model and numerical simulation. Background Technology

[0002] In cold-region metal mining, rock drilling and blasting are the primary methods of rock breaking. With frequent blasting operations, the resulting vibrations can pose potential hazards to surrounding buildings and structures, slope stability, and personnel safety. Therefore, accurately predicting blasting vibration velocities in cold regions is a crucial measure to ensure project safety.

[0003] Currently, the prediction of blasting vibration velocity in cold regions mainly relies on monitoring data from the mine site, and is predicted using empirical formulas (such as the Sadovsky formula) and machine learning methods based on monitoring data. Existing blasting vibration testing instruments mainly consist of two core components: a blasting vibration recorder and a vibration velocity sensor. The vibration velocity sensor has three channels: X, Y, and Z. During blasting monitoring, the vibration velocity sensor needs to be installed in a fixed position, and the angle between the X-axis of the sensor and the blast source center must be zero (referred to as the "angle"). However, the blasting area in the mine changes continuously with the progress of mining, and the angle also changes accordingly. This means that online monitoring systems with fixed sensors cannot perform long-term online monitoring of mine blasting vibration because the angle does not meet the requirements of the "Blasting Safety Regulations." While empirical formulas have certain applicability within specific lithology and blasting parameter ranges, they are difficult to accurately reflect the nonlinear propagation law of vibration under complex geological conditions in cold regions. In addition, although machine learning methods such as BP neural networks have strong nonlinear fitting capabilities, they still have obvious limitations in practical applications. The model training is prone to getting trapped in local optima, the convergence speed is slow, the prediction stability is poor, and the prediction accuracy is insufficient in the special environment of cold regions.

[0004] Against this backdrop, existing research has focused on improving the accuracy of blasting vibration prediction by refining single intelligent algorithms. However, these methods do not take into account the angular influence of fixed sensors and the impact of complex freeze-thaw environments in cold regions on the physical and mechanical parameters of rocks. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention deeply integrates intelligent algorithm optimization with numerical simulation. It employs the Whale Optimization Algorithm (WOA) to globally optimize the BP neural network, proposing a BP neural network model based on the Whale Optimization Algorithm - BackPropagation (WOA-BP) and a numerical simulation-based method for predicting blasting peak velocity, aiming to predict the blasting peak velocity and dominant frequency in cold regions.

[0006] This invention proposes a method for predicting peak blast velocity based on the WOA-BP model and numerical simulation. The method includes the following steps:

[0007] Laboratory impact tests were conducted to obtain several sets of raw vibration monitoring data and a vibration monitoring dataset was constructed. ;

[0008] Using vibration monitoring datasets A BP neural network model is trained and used as a sensor angle distortion correction model; wherein the input of the BP neural network model is the original vibration monitoring data, and the output is the peak velocity and dominant frequency obtained after correcting the original vibration monitoring data;

[0009] On-site blasting was carried out in the target blasting area, and online monitoring of blasting vibration was conducted. The monitoring data was corrected using a sensor angle distortion correction model to obtain an on-site vibration monitoring dataset. ;

[0010] A WOA-BP prediction model was constructed and a field vibration monitoring dataset was used. Train the model to obtain a trained WOA-BP prediction model;

[0011] A numerical model of bench blasting in the target blasting area was constructed. This model was used to simulate blasting vibrations under different working conditions, and an expanded vibration monitoring dataset was built. ;

[0012] Using field vibration monitoring datasets and expanding vibration monitoring datasets The trained WOA-BP prediction model is then corrected to obtain the final WOA-BP prediction model.

[0013] Obtain any set of blasting condition parameters for the target blasting area and input them into the final WOA-BP prediction model to obtain the peak vibration velocity prediction value and the dominant frequency prediction value.

[0014] Furthermore, by conducting laboratory impact tests, several sets of raw vibration monitoring data are obtained, and a vibration monitoring dataset is constructed. The specific content is as follows:

[0015] On a horizontal surface, the vibration wave generated by the free fall of a heavy hammer impacting the ground is used as the vibration source. The landing point of the heavy hammer is used as the center to divide the area into several circles with different test radii.

[0016] Several vibration velocity sensors are arranged at a preset angle on the circumference of an arbitrary test radius, and each vibration velocity sensor is connected to a blasting vibration recorder.

[0017] Multiple impact tests were conducted by varying the drop height of the hammer, and the three-dimensional vibration waveform data collected by all blasting vibration recorders were obtained in each impact test. At the same time, the gravitational potential energy of the hammer drop, the horizontal angle between each vibration velocity sensor and the vibration center, and the horizontal distance between each vibration velocity sensor and the vibration source were recorded for each impact test.

[0018] For the three-dimensional vibration waveform data collected by each blasting vibration recorder, Fourier transform (FFT) was used for spectrum analysis to extract the dominant frequency and peak velocity in the X, Y, and Z directions, respectively; where X, Y, and Z represent three mutually perpendicular vibration directions.

[0019] In each impact test, the gravitational potential energy of the falling hammer corresponding to a single vibration velocity sensor, the horizontal angle between the vibration velocity sensor and the vibration center, the horizontal distance between the vibration velocity sensor and the vibration source, and the dominant frequency and peak vibration velocity in the X, Y, and Z directions are used as a set of raw vibration monitoring data.

[0020] A vibration monitoring dataset was constructed using all the original vibration monitoring data. .

[0021] Furthermore, the use of vibration monitoring datasets The specific content of training a BP neural network model and using the trained BP neural network model as a sensor angle distortion correction model is as follows:

[0022] The vibration monitoring dataset is divided according to a preset ratio. It is divided into a vibration monitoring training set and a vibration monitoring test set;

[0023] A first BP neural network is constructed; wherein the input layer of the first BP neural network is used to receive the gravity potential energy of the falling hammer, the horizontal angle between the vibration velocity sensor and the vibration center, the horizontal distance between the vibration velocity sensor and the vibration source, and the dominant frequency and peak vibration velocity in the X, Y, and Z directions contained in each set of original vibration monitoring data; the output layer of the first BP neural network is used to output the corrected dominant frequency and peak vibration velocity in the X, Y, and Z directions.

[0024] The vibration monitoring training set is input into the first BP neural network and trained using the backpropagation algorithm to obtain the trained first BP neural network.

[0025] The trained first BP neural network was validated using a vibration monitoring test set, and the validated first BP neural network was used as a sensor angle distortion correction model.

[0026] Furthermore, the process involves conducting on-site blasting in the target blasting area and performing online monitoring of blasting vibrations. The monitoring data is then corrected using a sensor angle distortion correction model to obtain an on-site vibration monitoring dataset. The specific content is as follows:

[0027] Several monitoring points were set up in the target blast area, and the coordinates and azimuth of each monitoring point were recorded;

[0028] Blasting operations are carried out in the target blasting area according to the preset working condition variables. For each blasting operation, the center coordinates and parameters of the blasting area are obtained, and the three-dimensional vibration waveform data of each monitoring point are obtained simultaneously.

[0029] For any monitoring point, Fourier transform (FFT) is used to perform spectral analysis on the three-dimensional vibration waveform data of the monitoring point to extract the dominant frequency and peak vibration velocity in the X, Y, and Z directions.

[0030] Based on the azimuth of the monitoring point, the dominant frequency and peak vibration velocity in the X, Y, and Z directions are corrected using the sensor angle distortion correction model, and the corrected dominant frequency and peak vibration velocity in the X, Y, and Z directions are used as the actual peak vibration velocity and actual dominant frequency of the monitoring point.

[0031] Using the coordinates of the monitoring point and the coordinates of the blasting zone center, calculate the distance from the monitoring point to the blasting zone center and the elevation difference between the monitoring point and the blasting zone;

[0032] The parameters of the blasting zone for each blasting operation, the distance from any monitoring point to the center of the blasting zone, the elevation difference between the monitoring point and the blasting zone, and the actual peak vibration velocity and actual dominant frequency of the monitoring point are used as a set of training samples. Then, using all the training samples, a field vibration monitoring dataset is constructed. .

[0033] Furthermore, the parameters of the blasting zone include: maximum charge per section, borehole spacing, Poisson's ratio of the rock mass in the blasting zone, and compressive strength.

[0034] Furthermore, the WOA-BP prediction model is constructed and the field vibration monitoring dataset is used. The trained WOA-BP prediction model contains the following details:

[0035] A second BP neural network is constructed, consisting of an input layer, a hidden layer, and an output layer. The input data of the input layer includes: the distance from the monitoring point to the center of the blasting zone, the elevation difference between the monitoring point and the blasting zone, the maximum charge per segment, the borehole spacing, the Poisson's ratio of the rock mass in the blasting zone, and the compressive strength. The output parameters of the output layer are: the actual peak velocity and the actual dominant frequency.

[0036] The on-site vibration monitoring dataset is divided according to a preset ratio. It is divided into a field vibration monitoring training set, a field vibration monitoring verification set, and a field vibration monitoring test set;

[0037] Define the parameter search space, including: number of restarts, number of hidden layer neurons, activation function, learning algorithm, and maximum number of training iterations;

[0038] Traverse the parameter search space. For each parameter combination, based on the number of restarts in the current parameter combination, repeat the following operations:

[0039] Based on the number of hidden layer neurons and activation function in the current parameter combination, determine the structure of the second BP neural network, and randomly initialize the weights and thresholds of the current second BP neural network;

[0040] Calculate the length of decision variables for the Whale Optimization Algorithm (WOA) and use the mean squared error as the optimization objective function;

[0041] The on-site vibration monitoring training set is input into the current second BP neural network, and the Whale Optimization Algorithm (WOA) is used for iterative optimization. When the preset stopping criterion is met, the optimal weights and thresholds of the current second BP neural network are obtained.

[0042] The obtained optimal weights and thresholds are assigned to the current second BP neural network. Based on the learning algorithm and maximum number of training iterations in the current parameter combination, the current second BP neural network is trained using the field vibration monitoring training set. The trained current second BP neural network is evaluated using the field vibration monitoring validation set, and the mean square error, AIC value, BIC value and determination coefficient are calculated as evaluation indicators of the current parameter combination.

[0043] Based on the evaluation index of all parameter combinations on the field vibration monitoring validation set, the optimal parameter combination is selected and assigned to the second BP neural network to obtain the trained WOA-BP prediction model.

[0044] Input the field vibration monitoring test set into the trained WOA-BP prediction model, and calculate the prediction accuracy of the model based on the output peak velocity prediction value and dominant frequency prediction value.

[0045] If the calculated prediction accuracy reaches the preset threshold, the model is used as the trained WOA-BP prediction model; if the calculated prediction accuracy does not reach the preset threshold, the parameter combination is modified and the second BP neural network is retrained.

[0046] Furthermore, the step blasting numerical model of the target blasting area is constructed, and the step blasting numerical model is used to simulate blasting vibrations under different working conditions, and an expanded vibration monitoring dataset is constructed. The specific content is as follows:

[0047] Construct a numerical model for bench blasting in the target blasting area;

[0048] Several different sets of working condition variables are set; each set of working condition variables includes: maximum charge per section, borehole spacing, row spacing, hole depth, minimum resistance line length, Poisson's ratio of rock mass, and compressive strength.

[0049] The numerical model of bench blasting was used to simulate each set of working condition variables, and the simulated peak vibration velocity and main frequency of each monitoring point in the numerical model of bench blasting were obtained.

[0050] Using the blast zone parameters corresponding to each set of working condition variables, the distance from any monitoring point to the center of the blast zone, the elevation difference between the monitoring point and the blast zone, and the simulated peak vibration velocity and dominant frequency of the monitoring point as an extended sample, and then using all the extended samples, an extended vibration monitoring dataset is constructed. .

[0051] Furthermore, the use of on-site vibration monitoring datasets and expanding vibration monitoring datasets The trained WOA-BP prediction model is then corrected to obtain the final WOA-BP prediction model, which contains the following:

[0052] Merging field vibration monitoring datasets and expanding vibration monitoring datasets The fused vibration monitoring dataset was obtained. ;

[0053] Using fused vibration monitoring datasets The trained WOA-BP prediction model is retrained to obtain the final WOA-BP prediction model.

[0054] Furthermore, the blasting condition parameters include: blasting zone parameters, the distance from any monitoring point to the center of the blasting zone, and the elevation difference between the monitoring point and the blasting zone.

[0055] The beneficial effects of adopting the above technical solution are as follows:

[0056] Considering the impact of freeze-thaw cycles in cold regions on rock mass parameters in mining areas, this invention deeply integrates intelligent algorithm optimization with numerical simulation. The Whale Algorithm (WOA) is used to globally optimize the BP neural network, overcoming its inherent tendency to get trapped in local optima and improving the model's nonlinear fitting ability and generalization performance. Furthermore, the optimized BP neural network is used to correct distorted monitoring data caused by changes in sensor angle.

[0057] Building upon this foundation, and considering the impact of complex geological conditions such as freeze-thaw cycles in cold regions, this invention incorporates ANSYS / LS-DYNA numerical simulation. By constructing a blasting response model for freeze-thawed rock masses in cold regions, it simulates and generates blasting vibration data under different working conditions. This provides multi-condition training samples for the WOA-BP model, overcoming the limitation of insufficient training data in complex cold environments by single intelligent optimization algorithms. This method not only improves the prediction accuracy of peak blasting velocity and dominant frequency in cold regions but also enhances the model's universality, further improving the reliability of the prediction results and making them more aligned with the needs of mine blasting vibration monitoring. Attached Figure Description

[0058] Figure 1 This is a flowchart of the peak blasting velocity prediction method based on the WOA-BP model and numerical simulation in this embodiment;

[0059] Figure 2 This is a schematic diagram of the peak blasting velocity prediction method based on the WOA-BP model and numerical simulation in this embodiment.

[0060] Figure 3 This is a schematic diagram of the arrangement of the blasting vibration similarity test instrument in this embodiment;

[0061] Figure 4 This is a schematic diagram of the blasting vibration testing system in this embodiment;

[0062] Figure 5 This is a diagram showing the instrument layout of the blasting vibration testing system in this embodiment;

[0063] Figure 6 This is a flowchart of the BP training process in this embodiment;

[0064] Figure 7 This is a flowchart of the WOA algorithm optimization BP neural network in this embodiment;

[0065] Figure 8 This is a performance evaluation comparison chart of the WOA-BP model and the BP neural network in this embodiment;

[0066] Figure 9 This is the optimal prediction result diagram after restarting in the X direction in this embodiment; where (a) is the prediction curve diagram of the test sample; and (b) is the error curve diagram.

[0067] Figure 10 This is the optimal prediction result diagram after restarting in the Y direction in this embodiment; where (a) is the prediction curve diagram of the test sample; and (b) is the error curve diagram.

[0068] Figure 11 This is the optimal prediction result diagram after restarting in the Z direction in this embodiment; where (a) is the prediction curve diagram of the test sample; and (b) is the error curve diagram.

[0069] Figure 12 This is a numerical simulation diagram of bench blasting in a cold region in this embodiment;

[0070] Figure 13 This is a distribution map of the monitoring points of the verification model in this embodiment. Detailed Implementation

[0071] To facilitate understanding of this application, specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The following embodiments are illustrative of the invention but are not intended to limit its scope. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0072] Example 1:

[0073] This embodiment presents a method for predicting peak vibration velocity in cold-region blasting based on a combination of the WOA-BP model and numerical simulation, such as... Figure 1 and Figure 2 As shown, the method includes the following steps:

[0074] Laboratory impact tests were conducted to obtain several sets of raw vibration monitoring data and a vibration monitoring dataset was constructed. .

[0075] The process involves conducting laboratory impact tests to obtain several sets of raw vibration monitoring data and constructing a vibration monitoring dataset. The specific content is as follows:

[0076] On a horizontal surface, the vibration wave generated by the free fall of a heavy hammer impacting the ground is used as the vibration source. Several circles with different test radii are drawn with the landing point of the heavy hammer as the center.

[0077] Several vibration velocity sensors are arranged at preset angles on the circumference of an arbitrary test radius, and each vibration velocity sensor is connected to a blasting vibration recorder.

[0078] Multiple impact tests were conducted by varying the drop height of the hammer, and triaxial vibration waveform data were acquired from all blasting vibration recorders during each impact test. Simultaneously, the gravitational potential energy of the falling hammer, the horizontal angle between each vibration velocity sensor and the vibration center, and the horizontal distance between each vibration velocity sensor and the vibration source were recorded for each impact test. The horizontal distance between each vibration velocity sensor and the vibration source is the test radius.

[0079] In this embodiment, eight identical L20-N blasting vibration meters, a measuring tape, an angle measuring device, and a weight are prepared. Vibration data are collected on a horizontal surface. The velocity sensors of the eight L20-N blasting vibration meters are arranged as follows: Figure 3 The setup is as shown. A weight is dropped freely from different heights above the ground, and the resulting vibration waves are used as the vibration source. Eleven test radii (radii of 3m, 2.8m, 2.6m, 2.4m, 2.2m, 2.0m, 1.8m, 1.6m, 1.4m, 1.2m, and 1.0m) are defined by the point of impact of the weight. Eight blasting vibration meters are positioned on the circumference of each circle at included angles of 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°, and numbered 1-8 respectively. Then, according to the set test radii, the drop height of the weight is varied at each radius, from 2m to 1m, with intervals of 0.1m. Each drop height is repeated three times, and the vibration data are recorded.

[0080] For the three-dimensional vibration waveform data collected by each blasting vibration recorder, Fourier transform (FFT) was used for spectral analysis to extract the dominant frequency and peak velocity in the X, Y, and Z directions, respectively. Here, X, Y, and Z represent three mutually perpendicular vibration directions.

[0081] In this embodiment, the monitoring data of the blasting vibration meter is exported, and the waveform is analyzed by Fourier transform (FFT) to extract the dominant frequency and peak velocity of the vibration waveform at each monitoring point, thereby obtaining the relevant waveform dominant frequency and the amplitude corresponding to the dominant frequency of the test data.

[0082] In each impact test, the gravitational potential energy of the falling hammer corresponding to a single vibration velocity sensor, the horizontal angle between the vibration velocity sensor and the vibration center, the horizontal distance between the vibration velocity sensor and the vibration source, and the dominant frequency and peak vibration velocity in the X, Y, and Z directions are used as a set of raw vibration monitoring data.

[0083] A vibration monitoring dataset was constructed using all the original vibration monitoring data. .

[0084] Using vibration monitoring datasets A BP neural network model is trained and used as a sensor angle distortion correction model; wherein the input of the BP neural network model is the original vibration monitoring data, and the output is the peak velocity and dominant frequency obtained after correcting the original vibration monitoring data.

[0085] The use of vibration monitoring datasets The specific content of training a BP neural network model and using the trained BP neural network model as a sensor angle distortion correction model is as follows:

[0086] The vibration monitoring dataset is divided according to a preset ratio. It is divided into a vibration monitoring training set and a vibration monitoring test set.

[0087] In this embodiment, the vibration monitoring dataset will be... The datasets are divided into training and testing sets, with 80% and 20% of the datasets used respectively. These datasets are then used to combine the algorithm and to train and test the model.

[0088] A first BP neural network is constructed; wherein the input layer of the first BP neural network is used to receive the gravity potential energy of the falling hammer, the horizontal angle between the vibration velocity sensor and the vibration center, the horizontal distance between the vibration velocity sensor and the vibration source, and the dominant frequency and peak vibration velocity in the X, Y, and Z directions included in each set of original vibration monitoring data; the output layer of the first BP neural network is used to output the corrected dominant frequency and peak vibration velocity in the X, Y, and Z directions.

[0089] In this embodiment, the training is based on the original vibration monitoring data corresponding to 0°.

[0090] The vibration monitoring training set is input into the first BP neural network and trained using the backpropagation algorithm to obtain the trained first BP neural network.

[0091] The trained first BP neural network was validated using a vibration monitoring test set, and the validated first BP neural network was used as a sensor angle distortion correction model.

[0092] In this embodiment, a large amount of vibration data generated by laboratory impact tests is used as a dataset for training the sensor angle distortion correction model. The model parameters are corrected, and the artificial neural network BP model is used to train and correct the blasting vibration waveform data to obtain a correction model for the distortion data caused by the sensor orientation change. This model is used to correct the measurement distortion data caused by the change of the included angle.

[0093] On-site blasting was carried out in the target blasting area, and online monitoring of blasting vibration was conducted. The monitoring data was corrected using a sensor angle distortion correction model to obtain an on-site vibration monitoring dataset. .

[0094] The process involves conducting on-site blasting in the target blasting area and performing online monitoring of blasting vibrations. A sensor angle distortion correction model is used to correct the monitoring data, resulting in an on-site vibration monitoring dataset. The specific content is as follows:

[0095] Several monitoring points were set up in the target blast area, and the coordinates and azimuth of each monitoring point were recorded.

[0096] Blasting operations are carried out in the target blasting area according to the preset working condition variables. For each blasting operation, the center coordinates and parameters of the blasting area are obtained, and the three-dimensional vibration waveform data of each monitoring point are obtained simultaneously.

[0097] The parameters of the blasting zone include: maximum charge per section, borehole spacing, Poisson's ratio of the rock mass in the blasting zone, and compressive strength.

[0098] For any monitoring point, Fourier transform (FFT) is used to perform spectral analysis on the three-dimensional vibration waveform data of the monitoring point, and the dominant frequency and peak velocity in the X, Y and Z directions are extracted.

[0099] Based on the azimuth of the monitoring point, the dominant frequency and peak vibration velocity in the extracted X, Y, and Z directions are corrected using a sensor angle distortion correction model. The corrected dominant frequency and peak vibration velocity in the X, Y, and Z directions are then used as the actual peak vibration velocity and actual dominant frequency of the monitoring point.

[0100] In this embodiment, as Figure 4 As shown, a typical online monitoring system for blasting vibration includes: a vibration recorder, sensors, a GPS antenna, a 4G antenna, a charger, and a monitoring cloud platform. The vibration recorder has a built-in communication module, enabling real-time uploading of test data to the cloud server. Vibration recorders are deployed at each monitoring point to collect real-time location information and blasting vibration data. Taking the Wulagen lead-zinc mine as an example... Figure 5 As shown, six L20-N type blasting vibration monitoring instruments were set up at the mine site to record the coordinates of the deployment points and their azimuth angles, and to collect the coordinates and parameters of the blasting area.

[0101] Using the coordinates of the monitoring point and the center of the blasting area, calculate the distance from the monitoring point to the center of the blasting area and the elevation difference between the monitoring point and the blasting area.

[0102] The parameters of the blasting zone for each blasting operation, the distance from any monitoring point to the center of the blasting zone, the elevation difference between the monitoring point and the blasting zone, and the actual peak vibration velocity and actual dominant frequency of the monitoring point are used as a set of training samples. Then, using all the training samples, a field vibration monitoring dataset is constructed. .

[0103] A WOA-BP prediction model was constructed and a field vibration monitoring dataset was used. The model is trained to obtain a well-trained WOA-BP prediction model.

[0104] In this embodiment, after data correction, a BP neural network improved with the whale optimization algorithm is used to process the on-site vibration monitoring dataset. A predictive comparative analysis was conducted. Descriptive statistical analysis was performed on blasting design parameters, rock mass physical and mechanical parameters, and peak particle velocity (PPV). Combined with Pearson and Spearman correlation analysis, blasting design parameters (blast center distance, elevation difference, maximum charge per segment, borehole spacing) and lithological parameters (Poisson's ratio, compressive strength) were used as input variables. The corrected dominant frequencies and peak velocities in the horizontal radial, horizontal tangential, and vertical directions were used as output variables. The weights and thresholds of the backpropagation (BP) neural network were optimized using the WOA optimization algorithm to find two optimal initial weights and threshold parameters. The optimized parameters were then used in the model for regression fitting and prediction output, forming the WOA-BP prediction model.

[0105] The WOA-BP prediction model was constructed and the field vibration monitoring dataset was used. The trained WOA-BP prediction model contains the following details:

[0106] A second BP neural network is constructed, consisting of an input layer, a hidden layer, and an output layer. The input data of the input layer includes: the distance from the monitoring point to the center of the blasting area, the elevation difference between the monitoring point and the blasting area, the maximum charge amount per segment, the borehole spacing, the Poisson's ratio of the rock mass in the blasting area, and the compressive strength. The output parameters of the output layer are: the actual peak vibration velocity and the actual dominant frequency.

[0107] In this embodiment, a BP neural network model based on the whale optimization algorithm, namely the WOA-BP prediction model, is established. This model takes the distance from the monitoring point to the center of the blast zone, the elevation difference between the monitoring point and the blast zone, the maximum charge per segment, the borehole spacing, the Poisson's ratio of the rock mass in the blast zone, and the compressive strength as inputs, and the actual peak velocity and actual dominant frequency as outputs. The model has 6 nodes in the input layer, 2 nodes in the output layer, and 2 hidden neurons. The training process of the second BP neural network model is as follows: Figure 6 As shown.

[0108] The on-site vibration monitoring dataset is divided according to a preset ratio. It is divided into a field vibration monitoring training set, a field vibration monitoring verification set, and a field vibration monitoring test set;

[0109] Define the parameter search space, including: number of restarts, number of hidden layer neurons, activation function, learning algorithm, and maximum number of training iterations;

[0110] Traverse the parameter search space. For each parameter combination, based on the number of restarts in the current parameter combination, repeat the following operations:

[0111] Based on the number of hidden layer neurons and activation function in the current parameter combination, determine the structure of the second BP neural network, and randomly initialize the weights and thresholds of the current second BP neural network.

[0112] Calculate the length of the decision variables in the Whale Optimization Algorithm (WOA) and use the mean squared error as the objective function.

[0113] The on-site vibration monitoring training set is input into the current second BP neural network, and the Whale Optimization Algorithm (WOA) is used for iterative optimization. When the preset stopping criterion is met, the optimal weights and thresholds of the current second BP neural network are obtained.

[0114] In this embodiment, the iterative optimization process of the Whale Optimization Algorithm (WOA) is as follows: Figure 7 As shown, the stopping criteria for the WOA algorithm are set to one or a combination of the following conditions: (1) reaching the preset maximum number of iterations (200 in this embodiment); (2) the training error (e.g., mean squared error MSE) reaches the preset target accuracy (0.001 in this embodiment); (3) training stops when the validation set error no longer decreases within several consecutive iterations (e.g., 10 times). The weights and thresholds of the neural network are optimized using the whale optimization algorithm. The optimized weights and thresholds are assigned to the second BP neural network. The optimized BP neural network is trained and tested, and error analysis and accuracy are compared with the second BP neural network before optimization. Figure 8 As shown.

[0115] The obtained optimal weights and thresholds are assigned to the current second BP neural network. Based on the learning algorithm and maximum number of training iterations in the current parameter combination, the current second BP neural network is trained using the field vibration monitoring training set. The trained current second BP neural network is evaluated using the field vibration monitoring validation set, and the mean square error, AIC value, BIC value, and coefficient of determination are calculated as evaluation indicators for the current parameter combination.

[0116] In this embodiment, prediction accuracy is improved by adjusting model parameters. Based on the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), and the MSE function, a two-layer optimization structure is constructed to improve the generalization ability of the network model. The impact of model restart count, number of hidden layer neurons, activation function, learning algorithm, and maximum training iterations on the model's prediction performance is considered. The initial weights and threshold parameters of the BP neural network model are optimized using the Whale Algorithm (WOA). The influence of the maximum number of iterations and population size of the Whale Algorithm on the prediction performance of the WOA-BP model is analyzed, and the optimal model parameters are obtained. The optimal model parameters obtained in this embodiment are: Logsig activation function, Levenberg-Maquardt training algorithm, training accuracy of 0.001, 2 hidden layer neurons, maximum training iterations of 200, initial maximum iterations of 60, and population size of 60 for the WOA-BP neural network model.

[0117] Based on the evaluation index of all parameter combinations on the field vibration monitoring validation set, the optimal parameter combination is selected and assigned to the second BP neural network to obtain the trained WOA-BP prediction model.

[0118] The on-site vibration monitoring test set is input into the trained WOA-BP prediction model, and the prediction accuracy of the model is calculated based on the output peak velocity prediction value and dominant frequency prediction value.

[0119] If the calculated prediction accuracy reaches the preset threshold, the model is used as the trained WOA-BP prediction model; if the calculated prediction accuracy does not reach the preset threshold, the parameter combination is modified and the second BP neural network is retrained.

[0120] In this embodiment, the field vibration monitoring test set is input into the trained WOA-BP model, and the accuracy of the WOA-BP prediction values ​​is calculated. If the calculated accuracy does not meet the requirements, the parameters of the WOA-BP model are adjusted until the accuracy reaches a set threshold, and the final WOA-BP model is established. Specifically, the field vibration monitoring test set is input into the trained WOA-BP model, and the WOA-BP prediction values ​​are output. The blasting vibration prediction results in three directions are as follows: Figure 9 , Figure 10 and Figure 11As shown, the determination coefficients of the model's predictions in the X, Y, and Z directions are 0.9372, 0.9113, and 0.9051, respectively. The accuracy of WOA-BP predictions must meet the following requirements: a determination coefficient R² consistently above 0.90, minimum mean square error (MSE) and root mean square error (RMSE), and mean absolute percentage error (MAPE) < 10%.

[0121] A numerical model of bench blasting in the target blasting area was constructed. This model was used to simulate blasting vibrations under different working conditions, and an expanded vibration monitoring dataset was built. .

[0122] The step blasting numerical model of the target blasting area is constructed, and the blasting vibration under different working conditions is simulated using the step blasting numerical model, and an expanded vibration monitoring dataset is constructed. The specific content is as follows:

[0123] Construct a numerical model for bench blasting in the target blasting zone.

[0124] In this embodiment, a numerical model of bench blasting in the target blasting area is constructed based on the collected field data. For example... Figure 12 As shown, a two-tiered stepped model, 200m long and 30m high, was constructed using a continuous charge structure. There were four boreholes, with a charge (maximum charge per section) of 125kg and a loading length of 9m, including a 7.2m length for rock debris filling and intervals. The inter-bore and inter-row delay times were set to 17ms and 40ms, respectively. The borehole depth was 16.2m, including a 1.2m over-depth, with a minimum resist line length of 4.5m. The rock mass was modeled using the RHT dynamic constitutive model, with a Poisson's ratio of 0.378 and a compressive strength of 24.42MPa. Monitoring points were set on the stepped surfaces and slopes of the model, such as... Figure 13 As shown.

[0125] Several different sets of working condition variables are set; each set of working condition variables includes: maximum charge per section, borehole spacing, row spacing, hole depth, minimum resistance line length, Poisson's ratio of rock mass, and compressive strength.

[0126] The numerical model of bench blasting was used to simulate each set of working condition variables, and the simulated peak vibration velocity and main frequency of each monitoring point in the numerical model of bench blasting were obtained.

[0127] Using the blast zone parameters corresponding to each set of working condition variables, the distance from any monitoring point to the center of the blast zone, the elevation difference between the monitoring point and the blast zone, and the simulated peak vibration velocity and dominant frequency of the monitoring point as an extended sample, and then using all the extended samples, an extended vibration monitoring dataset is constructed. .

[0128] In this embodiment, ANSYS / LS-DYNA numerical simulation software was used in conjunction with the mechanical parameters of freeze-thaw rock masses in cold regions to simulate the bench blasting process in open-pit mines. Multiple different working conditions were set according to field conditions, with variables including maximum charge per section, hole spacing, row spacing, hole depth, and minimum resistance line length. The peak blasting velocity and dominant frequency were simulated at different monitoring points for each working condition. By comparing with field measured data, the correctness of the bench blasting numerical model can be verified, and the peak blasting velocity and dominant frequency at each corresponding monitoring point can be obtained to construct an expanded sample.

[0129] Using field vibration monitoring datasets and expanding vibration monitoring datasets The trained WOA-BP prediction model is then corrected to obtain the final WOA-BP prediction model.

[0130] The use of field vibration monitoring datasets and expanding vibration monitoring datasets The trained WOA-BP prediction model is then corrected to obtain the final WOA-BP prediction model, which contains the following:

[0131] Merging field vibration monitoring datasets and expanding vibration monitoring datasets The fused vibration monitoring dataset was obtained. .

[0132] Using fused vibration monitoring datasets The trained WOA-BP prediction model is retrained to obtain the final WOA-BP prediction model.

[0133] In this embodiment, numerical simulation software is used to simulate bench blasting under different cold-region conditions, taking into account lithology under different freeze-thaw conditions, to obtain peak velocities of blasting vibrations, providing sufficient training data for the WOA-BP model. A WOA-BP neural network prediction model under cold-region geological conditions is constructed by combining the WOA-BP model with numerical simulation, thereby predicting the peak particle velocity and dominant frequency under different freeze-thaw conditions and regions.

[0134] Obtain any set of blasting condition parameters for the target blasting area and input them into the final WOA-BP prediction model to obtain the peak vibration velocity prediction value and the dominant frequency prediction value.

[0135] The blasting condition parameters include: blasting zone parameters, the distance from any monitoring point to the center of the blasting zone, and the elevation difference between the monitoring point and the blasting zone.

[0136] Based on the predicted results, the impact of blasting vibrations on surrounding buildings, slope stability, and personnel safety is assessed and compared with current safety standards such as the "Blasting Safety Regulations." If the predicted value exceeds the allowable threshold, the blasting design parameters (such as maximum charge per section, borehole spacing, row spacing, hole depth, minimum resistance line length, etc.) are adjusted and re-entered into the model for prediction until the safety control requirements are met, thereby guiding the optimized design and safe implementation of on-site blasting operations.

[0137] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this disclosure and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A method for predicting peak blast velocity based on the WOA-BP model and numerical simulation, characterized in that, The method includes the following steps: Laboratory impact tests were conducted to obtain several sets of raw vibration monitoring data and a vibration monitoring dataset was constructed. ; Using vibration monitoring datasets A BP neural network model is trained and used as a sensor angle distortion correction model; wherein the input of the BP neural network model is the original vibration monitoring data, and the output is the peak velocity and dominant frequency obtained after correcting the original vibration monitoring data; On-site blasting was carried out in the target blasting area, and online monitoring of blasting vibration was conducted. The monitoring data was corrected using a sensor angle distortion correction model to obtain an on-site vibration monitoring dataset. ; A WOA-BP prediction model was constructed and a field vibration monitoring dataset was used. Train the model to obtain a trained WOA-BP prediction model; A numerical model of bench blasting in the target blasting area was constructed. This model was used to simulate blasting vibrations under different working conditions, and an expanded vibration monitoring dataset was built. ; Using field vibration monitoring datasets and expanding vibration monitoring datasets The trained WOA-BP prediction model is then corrected to obtain the final WOA-BP prediction model. Obtain any set of blasting condition parameters for the target blasting area and input them into the final WOA-BP prediction model to obtain the peak vibration velocity prediction value and the dominant frequency prediction value.

2. The method for predicting peak blast velocity based on the WOA-BP model and numerical simulation according to claim 1, characterized in that, The process involves conducting laboratory impact tests to obtain several sets of raw vibration monitoring data and constructing a vibration monitoring dataset. The specific content is as follows: On a horizontal surface, the vibration wave generated by the free fall of a heavy hammer impacting the ground is used as the vibration source. The landing point of the heavy hammer is used as the center to divide the area into several circles with different test radii. Several vibration velocity sensors are arranged at a preset angle on the circumference of an arbitrary test radius, and each vibration velocity sensor is connected to a blasting vibration recorder. Multiple impact tests were conducted by varying the drop height of the hammer, and the three-dimensional vibration waveform data collected by all blasting vibration recorders were obtained in each impact test. At the same time, the gravitational potential energy of the hammer drop, the horizontal angle between each vibration velocity sensor and the vibration center, and the horizontal distance between each vibration velocity sensor and the vibration source were recorded for each impact test. For the three-dimensional vibration waveform data collected by each blasting vibration recorder, Fourier transform (FFT) was used for spectrum analysis to extract the dominant frequency and peak velocity in the X, Y, and Z directions, respectively; where X, Y, and Z represent three mutually perpendicular vibration directions. In each impact test, the gravitational potential energy of the falling hammer corresponding to a single vibration velocity sensor, the horizontal angle between the vibration velocity sensor and the vibration center, the horizontal distance between the vibration velocity sensor and the vibration source, and the dominant frequency and peak vibration velocity in the X, Y, and Z directions are used as a set of raw vibration monitoring data. A vibration monitoring dataset was constructed using all the original vibration monitoring data. .

3. The method for predicting peak blasting velocity based on the WOA-BP model and numerical simulation according to claim 1, characterized in that, The use of vibration monitoring datasets The specific content of training a BP neural network model and using the trained BP neural network model as a sensor angle distortion correction model is as follows: The vibration monitoring dataset is divided according to a preset ratio. It is divided into a vibration monitoring training set and a vibration monitoring test set; A first BP neural network is constructed; wherein the input layer of the first BP neural network is used to receive the gravity potential energy of the falling hammer, the horizontal angle between the vibration velocity sensor and the vibration center, the horizontal distance between the vibration velocity sensor and the vibration source, and the dominant frequency and peak vibration velocity in the X, Y, and Z directions contained in each set of original vibration monitoring data; the output layer of the first BP neural network is used to output the corrected dominant frequency and peak vibration velocity in the X, Y, and Z directions. The vibration monitoring training set is input into the first BP neural network and trained using the backpropagation algorithm to obtain the trained first BP neural network. The trained first BP neural network was validated using a vibration monitoring test set, and the validated first BP neural network was used as a sensor angle distortion correction model.

4. The method for predicting peak blasting velocity based on the WOA-BP model and numerical simulation according to claim 1, characterized in that, The process involves conducting on-site blasting in the target blasting area and performing online monitoring of blasting vibrations. A sensor angle distortion correction model is used to correct the monitoring data, resulting in an on-site vibration monitoring dataset. The specific content is as follows: Several monitoring points were set up in the target blast area, and the coordinates and azimuth of each monitoring point were recorded; Blasting operations are carried out in the target blasting area according to the preset working condition variables. For each blasting operation, the center coordinates and parameters of the blasting area are obtained, and the three-dimensional vibration waveform data of each monitoring point are obtained simultaneously. For any monitoring point, Fourier transform (FFT) is used to perform spectral analysis on the three-dimensional vibration waveform data of the monitoring point to extract the dominant frequency and peak vibration velocity in the X, Y, and Z directions. Based on the azimuth of the monitoring point, the dominant frequency and peak vibration velocity in the X, Y, and Z directions are corrected using the sensor angle distortion correction model, and the corrected dominant frequency and peak vibration velocity in the X, Y, and Z directions are used as the actual peak vibration velocity and actual dominant frequency of the monitoring point. Using the coordinates of the monitoring point and the coordinates of the blasting zone center, calculate the distance from the monitoring point to the blasting zone center and the elevation difference between the monitoring point and the blasting zone; The parameters of the blasting zone for each blasting operation, the distance from any monitoring point to the center of the blasting zone, the elevation difference between the monitoring point and the blasting zone, and the actual peak vibration velocity and actual dominant frequency of the monitoring point are used as a set of training samples. Then, using all the training samples, a field vibration monitoring dataset is constructed. .

5. The method for predicting peak blast velocity based on the WOA-BP model and numerical simulation according to claim 4, characterized in that, The parameters of the blasting zone include: maximum charge per section, borehole spacing, Poisson's ratio of the rock mass in the blasting zone, and compressive strength.

6. The method for predicting peak blast velocity based on the WOA-BP model and numerical simulation according to claim 5, characterized in that, The WOA-BP prediction model was constructed and the field vibration monitoring dataset was used. The trained WOA-BP prediction model contains the following details: A second BP neural network is constructed, consisting of an input layer, a hidden layer, and an output layer. The input data of the input layer includes: the distance from the monitoring point to the center of the blasting zone, the elevation difference between the monitoring point and the blasting zone, the maximum charge per segment, the borehole spacing, the Poisson's ratio of the rock mass in the blasting zone, and the compressive strength. The output parameters of the output layer are: the actual peak velocity and the actual dominant frequency. The on-site vibration monitoring dataset is divided according to a preset ratio. It is divided into a field vibration monitoring training set, a field vibration monitoring verification set, and a field vibration monitoring test set; Define the parameter search space, including: number of restarts, number of hidden layer neurons, activation function, learning algorithm, and maximum number of training iterations; Traverse the parameter search space. For each parameter combination, based on the number of restarts in the current parameter combination, repeat the following operations: Based on the number of hidden layer neurons and activation function in the current parameter combination, determine the structure of the second BP neural network, and randomly initialize the weights and thresholds of the current second BP neural network; Calculate the length of decision variables for the Whale Optimization Algorithm (WOA) and use the mean squared error as the optimization objective function; The on-site vibration monitoring training set is input into the current second BP neural network, and the Whale Optimization Algorithm (WOA) is used for iterative optimization. When the preset stopping criterion is met, the optimal weights and thresholds of the current second BP neural network are obtained. The obtained optimal weights and thresholds are assigned to the current second BP neural network. Based on the learning algorithm and maximum number of training iterations in the current parameter combination, the current second BP neural network is trained using the field vibration monitoring training set. The trained current second BP neural network is evaluated using the field vibration monitoring validation set, and the mean square error, AIC value, BIC value and determination coefficient are calculated as evaluation indicators of the current parameter combination. Based on the evaluation index of all parameter combinations on the field vibration monitoring validation set, the optimal parameter combination is selected and assigned to the second BP neural network to obtain the trained WOA-BP prediction model. Input the field vibration monitoring test set into the trained WOA-BP prediction model, and calculate the prediction accuracy of the model based on the output peak velocity prediction value and dominant frequency prediction value. If the calculated prediction accuracy reaches the preset threshold, the model is used as the trained WOA-BP prediction model; if the calculated prediction accuracy does not reach the preset threshold, the parameter combination is modified and the second BP neural network is retrained.

7. The method for predicting peak blast velocity based on the WOA-BP model and numerical simulation according to claim 6, characterized in that, The step blasting numerical model of the target blasting area is constructed, and the blasting vibration under different working conditions is simulated using the step blasting numerical model, and an expanded vibration monitoring dataset is constructed. The specific content is as follows: Construct a numerical model for bench blasting in the target blasting area; Several different sets of working condition variables are set; each set of working condition variables includes: maximum charge per section, borehole spacing, row spacing, hole depth, minimum resistance line length, Poisson's ratio of rock mass, and compressive strength. The numerical model of bench blasting was used to simulate each set of working condition variables, and the simulated peak vibration velocity and main frequency of each monitoring point in the numerical model of bench blasting were obtained. Using the blast zone parameters corresponding to each set of working condition variables, the distance from any monitoring point to the center of the blast zone, the elevation difference between the monitoring point and the blast zone, and the simulated peak vibration velocity and dominant frequency of the monitoring point as an extended sample, and then using all the extended samples, an extended vibration monitoring dataset is constructed. .

8. The method for predicting peak blast velocity based on the WOA-BP model and numerical simulation according to claim 7, characterized in that, The use of field vibration monitoring datasets and expanding vibration monitoring datasets The trained WOA-BP prediction model is then corrected to obtain the final WOA-BP prediction model, which contains the following: Merging field vibration monitoring datasets and expanding vibration monitoring datasets The fused vibration monitoring dataset was obtained. ; Using fused vibration monitoring datasets The trained WOA-BP prediction model is retrained to obtain the final WOA-BP prediction model.

9. The method for predicting peak blasting velocity based on the WOA-BP model and numerical simulation according to claim 1, characterized in that, The blasting condition parameters include: blasting zone parameters, the distance from any monitoring point to the center of the blasting zone, and the elevation difference between the monitoring point and the blasting zone.

Images