A deep learning-based open-pit mine cyclic blasting seismic wave propagation model construction method
By integrating spatial and temporal features through a deep learning model, the problem of insufficient accuracy of traditional models in predicting the propagation patterns of seismic waves during cyclic blasting in open-pit mines has been solved. This has enabled efficient and accurate prediction and optimization, adapting to complex geological conditions and meeting the needs of safe production in mines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BAOLI BLASTING LTD IN HAMI
- Filing Date
- 2026-03-18
- Publication Date
- 2026-06-16
AI Technical Summary
In existing technologies, traditional models are difficult to accurately depict the propagation law of seismic waves in cyclic blasting in open-pit mines. In particular, they lack accuracy under complex geological and topographical conditions and cannot effectively capture the temporal cumulative effect of cyclic blasting, thus failing to meet the needs of safe production in open-pit mines.
A deep learning-based approach was adopted, which integrates convolutional neural networks and long short-term memory networks to construct a seismic wave propagation model for cyclic blasting in open-pit mines. Through multi-dimensional monitoring data preprocessing and model optimization, spatial and temporal features were integrated into the model, capturing the dynamic changes in seismic wave propagation.
It improves the accuracy of predicting seismic wave propagation in cyclic blasting, adapts to complex geological conditions, enhances work efficiency, supports the long-term effectiveness and real-time optimization of the model, and meets the needs of mine production.
Smart Images
Figure CN122221160A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mining blasting technology, and in particular to a method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning. Background Technology
[0002] Blasting vibration is the most harmful effect among the four major adverse effects of blasting operations in open-pit mines, and its impact on slope stability is directly related to mine safety. In actual open-pit mining operations, slope failure is not caused by a single blast, but rather by the cumulative result of crack initiation, propagation, and connection under cyclic blasting vibration, ultimately leading to instability. Therefore, accurately characterizing the propagation law of cyclic blasting seismic waves is crucial for analyzing the cumulative slope failure mechanism and optimizing blasting parameters.
[0003] Currently, research on blasting seismic wave propagation models mainly focuses on single blasts, employing wave theory analytical models and numerical simulation models, such as empirical formulas based on elastic wave theory and finite element discrete element numerical models. However, these traditional models have significant drawbacks: Firstly, wave theory analytical models are mostly based on the assumption of homogeneous rock masses, making it difficult to adapt to the complex geological and topographical conditions of open-pit mines, such as interbedded soft and hard rock masses and rock masses with well-developed joints and fractures, resulting in low accuracy in characterizing the spatial characteristics of seismic wave propagation. Secondly, numerical simulation models are computationally intensive and inefficient, and cannot effectively capture the temporal cumulative effects of cyclic blasting, making it difficult to reflect the dynamic changes in seismic wave propagation patterns after multiple blasts. Furthermore, traditional models have low utilization of field monitoring data, failing to uncover the complex nonlinear relationships between blasting parameters, geological parameters, and seismic wave propagation parameters hidden within the data.
[0004] With the development of artificial intelligence technology, deep learning methods have shown significant advantages in nonlinear data modeling and temporal feature capture, and have been gradually applied in the field of mining engineering. However, in the current technology, the research on applying deep learning to the propagation model of blasting seismic waves is still in its early stages. No dedicated model for cyclic blasting has yet been formed. Most studies only consider the static characteristics of a single blast, without combining the temporal cumulative characteristics of cyclic blasting, and have not achieved the fusion modeling of spatial and temporal features. This results in insufficient accuracy in predicting the propagation law of cyclic blasting seismic waves, which cannot meet the needs of actual production in open-pit mines.
[0005] To address the aforementioned problems, this invention proposes a deep learning-based method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines. Summary of the Invention
[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning.
[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning includes the following steps: S1. Conduct on-site monitoring of cyclic blasting in open-pit mines and collect multi-dimensional monitoring data, including borehole blasting parameters, geological and topographical parameters and blasting seismic wave propagation parameters; S2. Preprocess the multi-dimensional monitoring data collected in step S1 to obtain a standardized dataset. The preprocessing includes outlier removal, missing value completion, data normalization, and feature encoding. S3. Based on wave theory, determine the influence characteristics and prediction targets of cyclic blasting seismic wave propagation. Use the influence characteristics as model input features and the prediction targets as model output features. Divide the standardized dataset to obtain training set, validation set and test set. S4. Build a deep learning model that integrates convolutional neural networks and long short-term memory networks. The convolutional neural network is used to extract the spatial features of the input features, and the long short-term memory network is used to capture the temporal cumulative features of cyclic bursts. S5. Input the training set into the deep learning model, set the model training hyperparameters, use the backpropagation algorithm to train the model, adjust the model parameters and hyperparameters in real time through the validation set, and use the test set to verify the prediction accuracy of the model to obtain a preliminary cyclic blasting seismic wave propagation model. S6. The preliminary cyclic blasting seismic wave propagation model is optimized by optimizing the model hyperparameters through grid search and cross-validation, and the model is iteratively fine-tuned by combining field measurement data to obtain the final high-precision deep learning-based cyclic blasting seismic wave propagation model for open-pit mines.
[0008] Preferably, in step S1, the drilling and blasting parameters include borehole spacing, borehole depth, charge structure, explosive consumption per unit, detonation network, delay time, and number of blasting cycles; the geological and topographical parameters include soil and rock type, topographical elevation, blasting direction, joint dip angle, and rock mass integrity coefficient; and the blasting seismic wave propagation parameters include blasting vibration velocity, frequency, duration, and seismic wave attenuation coefficient.
[0009] Preferably, in step S1, a blasting vibration monitoring instrument, a three-dimensional laser scanner, a drone, and a ground-penetrating radar are used for on-site monitoring. The blasting vibration monitoring instrument is set up at monitoring points at different elevations and different blasting directions on the slope to simultaneously collect seismic wave propagation parameters after each blast. The drone, combined with the three-dimensional laser scanner, acquires slope elevation point cloud data and three-dimensional geological and topographic features. The ground-penetrating radar detects rock joints and the distribution characteristics of soil and rock media.
[0010] Preferably, in step S2, outliers are removed using the three-standard-deviation criterion, missing values are filled using the nearest neighbor algorithm, the data is mapped to the zero-to-one interval using the minimum-maximum normalization method, and one-hot encoding is used to encode the categorical geological and topographic parameters.
[0011] Preferably, in step S3, the influencing features include borehole blasting parameters, geological and topographical parameters, and seismic wave propagation parameters of the previous several blasts, wherein the previous several blasts are positive integers greater than or equal to one; the prediction targets include blasting vibration velocity, frequency, duration, and seismic wave attenuation values at different propagation distances; the standardized dataset is divided into a training set, a validation set, and a test set in a ratio of seven to two to one.
[0012] Preferably, in step S4, the deep learning model structure includes an input layer, a convolutional neural network feature extraction layer, a long short-term memory network temporal feature layer, a fully connected layer, and an output layer. The convolutional neural network feature extraction layer includes convolutional layers, pooling layers, and activation layers, using a linear rectified function as the activation function to extract spatial correlation features of the input features. The long short-term memory network temporal feature layer has multiple hidden layers to capture the temporal cumulative change law of seismic wave propagation during cyclic blasting. The fully connected layer fuses the spatial features extracted by the convolutional neural network and the temporal features extracted by the long short-term memory network, and the output layer uses a linear activation function to output the prediction result.
[0013] Preferably, in step S5, the model training hyperparameters include learning rate, batch size, number of training epochs, number of hidden layer neurons, and regularization coefficient; mean squared error is used as the model loss function, and an adaptive moment estimation optimizer is used for backpropagation training. When the loss function of the validation set no longer decreases for ten consecutive epochs, the model training is stopped to prevent overfitting.
[0014] Preferably, in step S6, the hyperparameter search range of the grid search includes a learning rate of 0.0001 to 0.01, a batch size of 16 to 128, and a number of neurons in the hidden layer of the long short-term memory network of 64 to 512; the cross-validation adopts five-fold cross-validation, and selects the hyperparameter combination with the highest validation accuracy as the optimal hyperparameter of the model; the iterative fine-tuning involves supplementing the dataset with new field-tested cyclic blasting data, incrementally training the model, and updating the model parameters.
[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention integrates the advantages of convolutional neural networks and long short-term memory networks, and for the first time realizes the fusion modeling of spatial characteristics and temporal cumulative characteristics of seismic wave propagation in open-pit mine cyclic blasting. It effectively captures the dynamic changes in the propagation law of seismic waves after multiple blasts, solves the technical problem that traditional models can only depict single blasts and cannot reflect cumulative effects, and greatly improves the accuracy of seismic wave propagation prediction in cyclic blasting scenarios.
[0016] 2. This invention is based on a large amount of field monitoring data for modeling, covering multi-dimensional parameters such as borehole blasting, geological topography, and seismic wave propagation. It breaks through the limitations of traditional analytical models based on the assumption of homogeneous rock masses, and has strong adaptability to complex geological conditions such as interbedded soft and hard rock masses and rock masses with well-developed joints and fissures. It can be applied to the geological and topographical characteristics of different open-pit mines.
[0017] 3. The deep learning model of this invention has high prediction efficiency. Compared with traditional numerical simulation models, it does not require complex grid division and mechanical calculations. After inputting the parameters to be predicted, it can quickly output the predicted values of seismic wave propagation parameters, which meets the needs of real-time optimization of blasting parameters in mines and greatly improves work efficiency.
[0018] 4. This invention fully mines and utilizes on-site monitoring data. Through data preprocessing, feature engineering, and model optimization, it uncovers the complex nonlinear correlation between blasting parameters, geological parameters, and seismic wave propagation parameters, providing accurate seismic wave propagation data support for the analysis of the cumulative failure mechanism of slopes under cyclic blasting vibration.
[0019] 5. The model constructed by this invention supports incremental training and iterative optimization. New field measurement data from the mine can be continuously added to the dataset to update the model, enabling the model to adapt to the dynamic changes in geological topography and blasting parameters during the mining process, thus ensuring the long-term effectiveness of the model. Attached Figure Description
[0020] Figure 1 This is a logical block diagram of a method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as proposed in this invention. Detailed Implementation
[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0022] Reference Figure 1 A method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning includes the following steps: S1. Conduct on-site monitoring of cyclic blasting in open-pit mines and collect multi-dimensional monitoring data, including borehole blasting parameters, geological and topographical parameters, and blasting seismic wave propagation parameters. S2. Preprocess the multi-dimensional monitoring data collected in step S1 to obtain a standardized dataset. The preprocessing includes outlier removal, missing value completion, data normalization, and feature encoding. S3. Based on wave theory, determine the influence characteristics and prediction targets of cyclic blasting seismic wave propagation. Use the influence characteristics as model input features and the prediction targets as model output features. Divide the standardized dataset to obtain training set, validation set and test set. S4. Build a deep learning model. The deep learning model integrates convolutional neural networks and long short-term memory networks. Convolutional neural networks are used to extract spatial features of input features, and long short-term memory networks are used to capture temporal cumulative features of cyclic bursts. S5. Input the training set into the deep learning model, set the model training hyperparameters, use the backpropagation algorithm to train the model, adjust the model parameters and hyperparameters in real time through the validation set, and use the test set to verify the prediction accuracy of the model to obtain a preliminary cyclic blasting seismic wave propagation model. S6. The preliminary cyclic blasting seismic wave propagation model is optimized by optimizing the model hyperparameters through grid search and cross-validation, and the model is iteratively fine-tuned by combining field measurement data to obtain the final high-precision deep learning-based cyclic blasting seismic wave propagation model for open-pit mines.
[0023] In step S1, the drilling and blasting parameters include the spacing between borehole rows, borehole depth, charge structure, explosive consumption per unit, detonation network, delay time, and number of blast cycles; the geological and topographical parameters include the type of soil and rock medium, topographical elevation, blasting direction, joint dip angle, and rock mass integrity coefficient; and the blasting seismic wave propagation parameters include the blasting vibration velocity, frequency, duration, and seismic wave attenuation coefficient.
[0024] In step S1, blasting vibration monitoring instruments, 3D laser scanners, drones, and ground-penetrating radars are used for on-site monitoring. The blasting vibration monitoring instruments are set up at monitoring points at different elevations and different blasting directions on the slope to simultaneously collect seismic wave propagation parameters after each blast. The drones, combined with the 3D laser scanners, acquire slope elevation point cloud data and 3D geological and topographic features. The ground-penetrating radars detect rock joints and the distribution characteristics of soil and rock media.
[0025] In step S2, outliers are removed using the three-standard-deviation criterion, missing values are filled using the nearest neighbor algorithm, the data is mapped to the zero-to-one interval using the minimum-maximum normalization method, and one-hot encoding is used to encode the categorical geological and topographic parameters.
[0026] In step S3, the influencing features include borehole blasting parameters, geological and topographical parameters, and seismic wave propagation parameters of the previous few blasts, where "several" refers to a positive integer greater than or equal to one; the prediction targets include blasting vibration velocity, frequency, duration, and seismic wave attenuation values at different propagation distances; the standardized dataset is divided into training set, validation set, and test set in a ratio of seven to two to one.
[0027] In step S4, the deep learning model structure includes an input layer, a convolutional neural network feature extraction layer, a long short-term memory network temporal feature layer, a fully connected layer, and an output layer. The convolutional neural network feature extraction layer includes convolutional layers, pooling layers, and activation layers, using a linear rectified function as the activation function to extract spatial correlation features of the input features. The long short-term memory network temporal feature layer has multiple hidden layers to capture the temporal cumulative variation law of seismic wave propagation during cyclic blasting. The fully connected layer fuses the spatial features extracted by the convolutional neural network and the temporal features extracted by the long short-term memory network. The output layer uses a linear activation function to output the prediction results.
[0028] In step S5, the model training hyperparameters include learning rate, batch size, number of training epochs, number of hidden layer neurons, and regularization coefficient. The mean squared error is used as the model loss function, and an adaptive moment estimation optimizer is used for backpropagation training. When the loss function of the validation set no longer decreases for ten consecutive epochs, the model training is stopped to prevent overfitting.
[0029] In step S6, the hyperparameter search range for grid search includes a learning rate of 0.0001 to 0.01, a batch size of 16 to 128, and the number of neurons in the hidden layer of the long short-term memory network of 64 to 512. Cross-validation adopts five-fold cross-validation, and the hyperparameter combination with the highest validation accuracy is selected as the optimal hyperparameter of the model. Iterative fine-tuning involves supplementing the dataset with new field-tested cyclic blasting data, incrementally training the model, and updating the model parameters.
[0030] Working principle: The core of this invention is to integrate the spatial feature extraction capability of convolutional neural networks and the temporal feature capture capability of long short-term memory networks, and combine them with multi-dimensional monitoring data from open-pit mine cyclic blasting sites to construct a deep learning model capable of characterizing the spatial and temporal coupling features of seismic wave propagation. Its core working principle consists of three parts: Multi-dimensional data fusion principle: Full-dimensional data such as borehole blasting parameters, geological and topographical parameters, and blasting seismic wave propagation parameters are obtained through on-site monitoring, covering internal and external influencing factors of cyclic blasting seismic wave propagation. After standardization preprocessing, the differences in data dimensions and distribution are eliminated, providing high-quality training data for the model and mining nonlinear correlations between data. The principle of spatial-temporal feature fusion modeling is as follows: Convolutional neural networks are used to perform convolution and pooling operations on high-dimensional geological and topographic parameters and borehole blasting parameters to extract spatial features of seismic wave propagation, such as the spatial influence of different soil and rock media, topographic elevation, and joint dip angle on seismic wave propagation. Long short-term memory networks are used to model the temporal data of cyclic blasting, capturing the cumulative effect of previous blasts on the current blast's seismic wave propagation, thus solving the problem that traditional models cannot characterize the temporal features of cyclic blasting. Spatial and temporal features are fused through fully connected layers to achieve accurate modeling of the propagation law of cyclic blasting seismic waves. Model optimization and iterative learning principles: The model's hyperparameters are optimized through grid search and five-fold cross-validation to improve the model's generalization ability; the model is incrementally trained by combining new data from field measurements to achieve iterative updates, enabling the model to continuously adapt to the dynamic changes in geological and topographical conditions and blasting parameters during mining, thus ensuring the model's long-term prediction accuracy.
[0031] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, characterized in that, Includes the following steps: S1. Conduct on-site monitoring of cyclic blasting in open-pit mines and collect multi-dimensional monitoring data, including borehole blasting parameters, geological and topographical parameters and blasting seismic wave propagation parameters; S2. Preprocess the multi-dimensional monitoring data collected in step S1 to obtain a standardized dataset. The preprocessing includes outlier removal, missing value completion, data normalization, and feature encoding. S3. Based on wave theory, determine the influence characteristics and prediction targets of cyclic blasting seismic wave propagation. Use the influence characteristics as model input features and the prediction targets as model output features. Divide the standardized dataset to obtain training set, validation set and test set. S4. Build a deep learning model that integrates convolutional neural networks and long short-term memory networks. The convolutional neural network is used to extract the spatial features of the input features, and the long short-term memory network is used to capture the temporal cumulative features of cyclic bursts. S5. Input the training set into the deep learning model, set the model training hyperparameters, use the backpropagation algorithm to train the model, adjust the model parameters and hyperparameters in real time through the validation set, and use the test set to verify the prediction accuracy of the model to obtain a preliminary cyclic blasting seismic wave propagation model. S6. The preliminary cyclic blasting seismic wave propagation model is optimized by optimizing the model hyperparameters through grid search and cross-validation, and the model is iteratively fine-tuned by combining field measurement data to obtain the final high-precision deep learning-based cyclic blasting seismic wave propagation model for open-pit mines.
2. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S1, the drilling and blasting parameters include the borehole spacing, borehole depth, charge structure, explosive consumption per unit, detonation network, delay time, and number of blasting cycles; the geological and topographical parameters include the type of soil and rock medium, topographical elevation, blasting direction, joint dip angle, and rock mass integrity coefficient; and the blasting seismic wave propagation parameters include blasting vibration velocity, frequency, duration, and seismic wave attenuation coefficient.
3. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S1, blasting vibration monitoring instruments, 3D laser scanners, drones, and ground-penetrating radars are used for on-site monitoring. The blasting vibration monitoring instruments are set up at monitoring points at different elevations and different blasting directions on the slope to simultaneously collect seismic wave propagation parameters after each blast. The drones, combined with the 3D laser scanners, acquire slope elevation point cloud data and 3D geological and topographic features. The ground-penetrating radars detect rock joints and the distribution characteristics of soil and rock media.
4. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S2, outliers are removed using the three-standard-deviation criterion, missing values are filled using the nearest neighbor algorithm, the data is mapped to the zero-to-one interval using the minimum-maximum normalization method, and one-hot encoding is used to encode the categorical geological and topographic parameters.
5. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S3, the influencing features include borehole blasting parameters, geological and topographical parameters, and seismic wave propagation parameters of the previous several blasts, where the previous several blasts are positive integers greater than or equal to one; the prediction targets include blasting vibration velocity, frequency, duration, and seismic wave attenuation values at different propagation distances; the standardized dataset is divided into training set, validation set, and test set in a ratio of seven to two to one.
6. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S4, the deep learning model structure includes an input layer, a convolutional neural network feature extraction layer, a long short-term memory network temporal feature layer, a fully connected layer, and an output layer. The convolutional neural network feature extraction layer includes convolutional layers, pooling layers, and activation layers, using a linear rectified function as the activation function to extract spatial correlation features of the input features. The long short-term memory network temporal feature layer has multiple hidden layers to capture the temporal cumulative variation law of seismic wave propagation during cyclic blasting. The fully connected layer fuses the spatial features extracted by the convolutional neural network and the temporal features extracted by the long short-term memory network. The output layer uses a linear activation function to output the prediction result.
7. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S5, the model training hyperparameters include learning rate, batch size, number of training epochs, number of hidden layer neurons, and regularization coefficient; mean squared error is used as the model loss function, and an adaptive moment estimation optimizer is used for backpropagation training. When the loss function of the validation set no longer decreases for ten consecutive epochs, the model training is stopped to prevent overfitting.
8. The method for constructing a seismic wave propagation model for cyclic blasting in open-pit mines based on deep learning, as described in claim 1, is characterized in that... In step S6, the hyperparameter search range of the grid search includes a learning rate of 0.0001 to 0.01, a batch size of 16 to 128, and a number of neurons in the hidden layer of the long short-term memory network of 64 to 512. The cross-validation adopts five-fold cross-validation, and selects the hyperparameter combination with the highest validation accuracy as the optimal hyperparameter of the model. The iterative fine-tuning involves supplementing the dataset with new field-tested cyclic blasting data, incrementally training the model, and updating the model parameters.