Tunnel blasting parameter prediction method, system and device for asymmetric section

Abstract

This application relates to the field of tunnel engineering and blasting technology, and particularly to a method, system, and device for predicting tunnel blasting parameters for asymmetric cross-sections. The method includes acquiring an asymmetric three-dimensional geological grid and engineering context sequence for the target tunnel cross-section; inputting a tunnel blasting parameter prediction model; using a convolutional neural network layer to extract spatial features from the asymmetric three-dimensional geological grid; using a long short-term memory network layer to extract temporal features from the engineering context sequence; using a feature fusion layer to concatenate and jointly map the spatial and temporal features; using a fully connected regression layer to perform regression calculations on the fused features; and using a parameter output layer to output the predicted initial blasting parameters, which include blasting parameters for the weak side and blasting parameters for the hard side. The initial blasting parameters are then used for blasting simulation and multi-objective optimization to obtain the final blasting parameters for the target tunnel cross-section. This application can improve the adaptability of the blasting parameter prediction results to different working conditions on both sides of the cross-section.

Description

Technical Field

[0001] This application relates to the fields of tunnel engineering and blasting technology, and in particular to a method, system and equipment for predicting tunnel blasting parameters for asymmetric cross sections. Background Technology

[0002] Tunnel blasting excavation is a crucial operational step in tunnel construction, and the selection of blasting parameters directly impacts the excavation profile, control of surrounding rock disturbance, and construction cycle efficiency. As tunnel engineering increasingly extends to complex geological conditions, adjacent existing chambers, and localized weak interlayers, the construction target no longer exhibits a nearly homogeneous surrounding rock environment along the cross-section, but rather increasingly displays asymmetric characteristics with significant differences in rock conditions on both sides of the cross-section. Under these conditions, the surrounding rock on both sides of the tunnel often differs considerably in lithological composition, structural integrity, mechanical strength, and dynamic response characteristics. The propagation and attenuation patterns of blasting energy also change in different areas. Therefore, how to rationally determine blasting parameters based on asymmetric geological conditions has gradually become a key issue to consider in the field of tunnel blasting design.

[0003] In existing technologies, tunnel blasting parameters are typically determined based on the surrounding rock classification, cross-sectional dimensions, borehole type, and construction specifications, combined with the designer's experience, to set parameters such as charge quantity, borehole spacing, resistance line, detonation sequence, and time delay. In other words, in the actual design process, a set of blasting design criteria is usually first determined based on the overall geological conditions and construction conditions of the tunnel cross-section. Then, based on this set of blasting design criteria, the corresponding blasting parameters are designed, thus forming a blasting scheme for tunnel excavation. To improve the scientific rigor and efficiency of blasting design, existing technologies are beginning to incorporate big data analysis, machine learning models, numerical simulations, and optimization algorithms into the blasting parameter design process. For example, borehole parameters are first obtained and rock characteristics are inverted, then blasting parameters are determined based on these rock characteristics using numerical simulations and optimization algorithms, or blasting parameters are improved through parametric processing and intelligent optimization methods.

[0004] However, under asymmetric cross-section conditions, the surrounding rock on both sides of the tunnel typically exhibits significant differences in wave impedance characteristics, mechanical properties, and anti-interference capabilities. Current technologies primarily rely on overall cross-section analysis and parameter determination. Consequently, the different geological differences on both sides of the cross-section are difficult to stably and accurately represent during parameter design, and the parameter results obtained based on overall conditions are insufficient to accurately match the different working conditions on both sides. Furthermore, even with the introduction of intelligent algorithms and optimization methods, the underlying geological models are still largely constructed based on homogeneous or layered assumptions, and optimization objectives are mostly focused on the overall blasting effect, lacking dedicated constraints and adaptive adjustment mechanisms for asymmetric conditions. Therefore, it is difficult to proactively establish an accurate correspondence between asymmetric geological differences and blasting parameters, ultimately leading to a generalized parameter scheme oriented towards the overall cross-section, rather than accurately predicting blasting parameters that match the different working conditions on both sides of the cross-section. Summary of the Invention

[0005] This application provides a method, system, and equipment for predicting tunnel blasting parameters for asymmetric cross-sections, which can improve the adaptability of the predicted blasting parameters to different working conditions on both sides of the cross-section. The technical solution provided in this application is as follows: In a first aspect, this application provides a method for predicting tunnel blasting parameters for asymmetric cross-sections, the method comprising: Obtain an asymmetric three-dimensional geological grid and an engineering context sequence for the target tunnel cross-section, wherein the asymmetric three-dimensional geological grid is used to characterize the spatial geological distribution characteristics of the target tunnel cross-section, and the engineering context sequence is used to characterize the cross-sectional correlation characteristics of the target tunnel along the advancement direction; The asymmetric three-dimensional geological grid and the engineering context sequence are input into a pre-constructed tunnel blasting parameter prediction model. The tunnel blasting parameter prediction model includes a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer, and a parameter output layer. The convolutional neural network layer is used to extract spatial features from the asymmetric three-dimensional geological grid. The long short-term memory network layer is used to extract temporal features from the engineering context sequence. The feature fusion layer is used to concatenate and jointly map the spatial features and the temporal features. The fully connected regression layer is used to perform regression calculations on the fused features. The parameter output layer is used to output the initial blasting parameters of the target tunnel section. The initial blasting parameters include at least the blasting parameters of the weak side and the blasting parameters of the hard side. The initial blasting parameters are subjected to blasting simulation and multi-objective optimization to obtain the final blasting parameters of the target tunnel cross section.

[0006] In one specific implementation, obtaining the asymmetric three-dimensional geological mesh and engineering context sequence of the target tunnel cross-section includes: Acquire three-dimensional geometric morphology data, geological exploration data, and rock mechanics parameter data of the target tunnel cross section; A three-dimensional spatial geometric model of the target tunnel cross-section is established based on the aforementioned three-dimensional geometric morphology data; Based on the geological survey data, the spatial location range of joints, fissures, weak interlayers and various structural planes in the rock mass surrounding the target tunnel section is determined, and the rock mass surrounding the target tunnel section is spatially located and divided into regions according to the lithological distribution and structural plane distribution, resulting in multiple rock mass regions; Based on the rock mechanics parameter data, corresponding mechanical parameters are assigned to each of the rock mass regions to form a three-dimensional digital geological model of the target tunnel cross section; The three-dimensional digital geological model is discretized using voxelization to obtain an asymmetric three-dimensional geological mesh of the target tunnel cross section.

[0007] In one specific implementation, obtaining the asymmetric three-dimensional geological mesh and engineering context sequence of the target tunnel cross-section includes: The target tunnel is continuously sliced ​​along its advancement axis to form multiple sequentially arranged cross-sectional slices; For each cross-sectional slice, extract the contour geometry information of the cross-section and the constraint information of the surrounding environment; The contour geometry information and adjacent environmental constraint information corresponding to each cross-section slice are arranged in chronological order according to the direction of advancement to obtain the engineering context sequence of the target tunnel cross-section.

[0008] In one specific implementation scheme, the tunnel blasting parameter prediction model is a hybrid network structure oriented towards dual-source input of asymmetric cross-section, including two input ends corresponding to the asymmetric three-dimensional geological grid and the engineering context sequence, respectively, and a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer and a parameter output layer connected in sequence. The input end of the asymmetric three-dimensional geological grid is connected to a convolutional neural network layer to encode the spatial geological distribution information of the target tunnel cross-section. The input end of the engineering context sequence is connected to a long short-term memory network layer to encode the cross-sectional correlation information of the target tunnel along the advancement direction. The output ends of the convolutional neural network layer and the long short-term memory network layer are jointly connected to a feature fusion layer. The feature fusion layer is used to splice and jointly map spatial features and temporal features. The feature fusion layer is connected to a fully connected regression layer. The fully connected regression layer is connected to a parameter output layer. The parameter output layer includes two parallel output sub-layers corresponding to the weak side and the hard side, respectively, to output the initial blasting parameters of the weak side and the initial blasting parameters of the hard side.

[0009] In one specific feasible implementation, the process of constructing the tunnel blasting parameter prediction model includes: Acquire historical training samples, which include asymmetric three-dimensional geological grids, engineering context sequences, blasting parameter labels, and engineering effect labels; Historical training samples are input into a deep learning model with a pre-defined structure for training to obtain predicted values ​​of blasting parameters on the weak side and blasting parameters on the hard side. The parameter prediction loss is calculated based on the blasting parameter label, and the engineering effect prediction loss is calculated based on the engineering effect label. The parameter prediction loss, engineering effect prediction loss and regularization term are combined to form the total loss function, and the trainable parameters in the tunnel blasting parameter prediction model are updated according to the total loss function. When the preset training termination condition is met, a pre-constructed tunnel blasting parameter prediction model is obtained.

[0010] In a specific feasible implementation, the total loss function is composed of the weak-side parameter prediction loss, the hard-side parameter prediction loss, the engineering effect prediction loss, and a regularization term, and its expression is: ; in, Indicates the total loss; The weighting coefficients represent the predicted loss of the weak-side parameters; This represents the loss predicted by the weak-side parameters; The weighting coefficients represent the predicted loss for the hard-side parameters; This indicates the loss predicted by the hard-side parameters; Indicates the first Weighting coefficients for predicting the loss of project outcomes; Indicates the first Predicted losses corresponding to each project performance indicator; Represents the weight coefficient of the regularization term; Represents the regularization term; This represents the total number of engineering performance indicators included in the training. Among them, the weak-side parameter prediction loss This indicates the deviation between the weak-side explosion parameters output by the model and the true weak-side parameter labels. Represented as: ; in, Indicates the number of output parameters on the weak side; Indicating the weak side One real parameter label value; Indicating the weak side One predicted parameter value; Among them, the hard side parameter predicts the loss. This indicates the deviation between the hard-side blasting parameters output by the model and the true hard-side parameter labels. Represented as: ; in, Indicates the number of output parameters on the hard side; Indicates the hard side One real parameter label value; Indicates the hard side One predicted parameter value; Among them, the Project effect prediction loss Indicates the first The deviation between the predicted and actual values ​​of each project performance indicator Represented as: ; in, This indicates the number of samples in the current training batch; Indicates the first The corresponding sample of the th sample The true values ​​of each project performance indicator; Indicates the first The corresponding sample of the th sample Predicted values ​​of each project performance indicator; Among them, the regularization term Used to constrain the magnitude of model parameters. Represented as: ; in, This represents the set of all trainable parameters in the tunnel blasting parameter prediction model; Represents the parameter set Any trainable parameter in the dataset.

[0011] In a specific feasible implementation, the step of performing blasting simulation and multi-objective optimization on the initial blasting parameters to obtain the final blasting parameters of the target tunnel cross-section includes: A numerical calculation model for blasting is constructed based on an asymmetric three-dimensional geological grid of the target tunnel section and initial blasting parameters. Based on the aforementioned blasting numerical calculation model, blasting dynamic response simulation was performed to obtain stress wave propagation results, vibration velocity field results, and plastic zone distribution results. Based on the stress wave propagation results, vibration velocity field results, and plastic zone distribution results, the hard side fracture effect index, the weak side surrounding rock damage index, and the vibration index of the tunnel side are determined respectively. Using the charge amount, hole spacing, and detonation time difference as optimization variables, and taking the maximization of the hard side fracture effect index, the minimization of the soft side surrounding rock damage index, and the satisfaction of the safety threshold of the vibration index at the adjacent existing engineering object as optimization objectives, the initial blasting parameters are optimized in a multi-objective manner to obtain the final blasting parameters of the target tunnel section.

[0012] Secondly, this application provides a tunnel blasting parameter prediction system for asymmetric cross-sections, employing the following technical solution: A tunnel blasting parameter prediction system for asymmetric cross-sections includes: The data acquisition module is used to acquire the asymmetric three-dimensional geological grid and engineering context sequence of the target tunnel cross section. The asymmetric three-dimensional geological grid is used to characterize the spatial geological distribution characteristics of the target tunnel cross section, and the engineering context sequence is used to characterize the cross section correlation characteristics of the target tunnel along the advancement direction. The model prediction module is used to input the asymmetric three-dimensional geological grid and the engineering context sequence into a pre-constructed tunnel blasting parameter prediction model. The tunnel blasting parameter prediction model includes a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer, and a parameter output layer. The convolutional neural network layer is used to extract spatial features from the asymmetric three-dimensional geological grid. The long short-term memory network layer is used to extract temporal features from the engineering context sequence. The feature fusion layer is used to concatenate and jointly map the spatial features and the temporal features. The fully connected regression layer is used to perform regression calculations on the fused features. The parameter output layer is used to output the initial blasting parameters of the target tunnel section. The initial blasting parameters include at least the weak side blasting parameters and the hard side blasting parameters. The parameter determination module is used to perform blasting simulation and multi-objective optimization on the initial blasting parameters to obtain the final blasting parameters of the target tunnel cross section.

[0013] Thirdly, this application provides an electronic device, the device including a processor and a memory; the memory stores a program, the program being loaded and executed by the processor to implement a method for predicting tunnel blasting parameters for asymmetric cross sections as described in the first aspect.

[0014] Fourthly, this application provides a computer-readable storage medium storing a program that, when executed by a processor, is used to implement a method for predicting tunnel blasting parameters for asymmetric cross sections as described in the first aspect.

[0015] First, an asymmetric three-dimensional geological mesh and engineering context sequence of the target tunnel cross-section are obtained. The asymmetric three-dimensional geological mesh is used to characterize the spatial geological distribution characteristics of the target tunnel cross-section, and the engineering context sequence is used to characterize the cross-sectional correlation characteristics of the target tunnel along the advancement direction. Then, the asymmetric three-dimensional geological mesh and engineering context sequence are input into a pre-constructed tunnel blasting parameter prediction model. Spatial features are extracted through a convolutional neural network layer, and temporal features are extracted through a long short-term memory network layer. The spatial and temporal features are then spliced, jointly mapped, and regressed through a feature fusion layer and a fully connected regression layer. Finally, the parameter output layer outputs initial blasting parameters, which include at least the blasting parameters of the weak side and the blasting parameters of the hard side. Subsequently, blasting simulation and multi-objective optimization are performed on the initial blasting parameters to obtain the final blasting parameters of the target tunnel cross-section. Analysis based on the above technical solutions reveals that the asymmetric three-dimensional geological grid incorporates the geological distribution of both sides of the target tunnel cross-section within a spatial range into the parameter prediction process. The engineering context sequence further incorporates the cross-sectional correlation status of the target tunnel along the advancement direction into the parameter prediction process. Therefore, the input information is no longer limited to a single description of the overall cross-section, but simultaneously includes spatial difference information and longitudinal correlation information. On this basis, the convolutional neural network layer and the long short-term memory network layer respectively extract information from two different sources. Then, the feature fusion layer and the fully connected regression layer establish the correspondence between spatial features, temporal features, and blasting parameters. The parameter output layer outputs the blasting parameters for the weak side and the hard side respectively, so that the parameter results can fit the different working conditions on both sides of the cross-section. Furthermore, by performing blasting simulation and multi-objective optimization on the initial blasting parameters, the parameter results are verified and adjusted under the constraints of the actual blasting response. The final blasting parameters can not only more accurately reflect the spatial geological differences and advancement direction correlation characteristics of the asymmetric cross-section, but also improve the matching degree between the parameter results and the actual working conditions on both sides of the cross-section.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for predicting tunnel blasting parameters for asymmetric cross sections in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the overall process of the tunnel blasting parameter prediction method for asymmetric cross sections in the embodiments of this application.

[0019] Figure 3 This is a structural block diagram of a tunnel blasting parameter prediction system for asymmetric cross sections, as described in this application.

[0020] Figure 4 This is a block diagram of an electronic device for predicting tunnel blasting parameters for asymmetric cross sections, as described in this application. Detailed Implementation

[0021] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0022] Optionally, this application uses the tunnel blasting parameter prediction method for asymmetric cross sections provided in various embodiments as an example for application in an electronic device. The electronic device is a terminal or server. The terminal can be a computer, tablet computer, etc. This embodiment does not limit the type of electronic device.

[0023] Reference Figure 1 This is a flowchart illustrating a method for predicting tunnel blasting parameters for asymmetric cross-sections according to an embodiment of this application. The method includes at least the following steps: Step S101: Obtain the asymmetric three-dimensional geological grid and engineering context sequence of the target tunnel cross section. The asymmetric three-dimensional geological grid is used to characterize the spatial geological distribution characteristics of the target tunnel cross section, and the engineering context sequence is used to characterize the cross section correlation characteristics of the target tunnel along the advancement direction.

[0024] In step S101, an asymmetric three-dimensional geological mesh and engineering context sequence for the target tunnel cross-section are acquired. This step is used to generate input data characterizing the state of the target tunnel cross-section. The asymmetric three-dimensional geological mesh describes the rock mass distribution, structural features, and mechanical properties of the target tunnel cross-section within a spatial range, while the engineering context sequence describes the cross-sectional changes of the target tunnel along the advancement direction and the constraints of the adjacent environment. For asymmetric cross-sections, the rock masses on both sides of the tunnel differ in strength, integrity, and structural distribution. Describing the target tunnel cross-section solely with overall cross-sectional parameters is insufficient to accurately reflect the rock mass conditions on both sides. Therefore, input data needs to be constructed from both spatial geological distribution and longitudinal cross-sectional correlation perspectives.

[0025] Specifically, the process begins by acquiring three-dimensional geometric morphology data, geological survey data, and rock mechanics parameter data of the target tunnel cross-section. Specifically, a 3D laser scanner is used to scan the free face of the target tunnel, obtaining point cloud coordinate information of the free face. Based on this point cloud coordinate information, the contour line information and cross-sectional dimension information of the target cross-section are extracted, thus forming three-dimensional geometric morphology data characterizing the actual spatial form of the free face and the boundary contour of the target cross-section. Ground-penetrating radar is used to probe the surrounding rock on both sides of the target tunnel, and combined with borehole photography, the internal structure of the rock mass within the probed area is observed, identifying the spatial distribution of joints, fissures, weak interlayers, and various structural planes in the rock mass on both sides of the target tunnel. This process generates geological survey data to characterize the internal structure distribution of the rock mass on both sides of the target tunnel. Boreholes are drilled on both sides of the target tunnel to collect samples, and indoor rock mechanics tests are conducted on the obtained rock samples to obtain mechanical parameters such as uniaxial compressive strength, elastic modulus, Poisson's ratio, internal friction angle, and cohesion of the rock mass on the weak and hard sides. Combined with test results of wave velocity, density, or other parameters reflecting stress wave propagation characteristics, the wave impedance parameters corresponding to each rock mass region are determined, thus generating rock mechanics parameter data to characterize the material properties of different rock mass regions on both sides of the target tunnel.

[0026] After obtaining three-dimensional geometric morphology data, geological survey data, and rock mechanics parameter data, a three-dimensional spatial geometric model of the target tunnel cross-section is established based on the three-dimensional geometric morphology data, thus defining the spatial boundaries, free face morphology, and cross-sectional contour of the target tunnel cross-section in digital space. Based on this, the spatial location range of joints, fissures, weak interlayers, and various structural planes within the surrounding rock mass is determined according to the geological survey data. The surrounding rock mass is then spatially located and regionalized according to lithological distribution and structural plane distribution, resulting in multiple rock mass regions corresponding to different rock mass states. Then, based on the rock mechanics parameter data, each rock mass region is assigned corresponding mechanical parameters, giving each region both spatial location attributes and rock mass material properties. For rock mass regions with significant differences on both sides of the target tunnel cross-section, corresponding measured parameters are assigned, allowing the weak side and the hard side to form differentiated attribute expressions in the digital model. After the above processing, a three-dimensional digital geological model reflecting the geometric boundaries, structural distribution, and mechanical properties of the target tunnel cross-section is formed.

[0027] After the 3D digital geological model is formed, it undergoes voxel discretization, dividing the continuous 3D space into multiple voxel units to obtain an asymmetric 3D geological mesh. Each voxel unit corresponds to a specific spatial location around the target tunnel section, and each voxel unit records the geological attribute information at that location. The geological attribute information includes at least the uniaxial compressive strength, elastic modulus, wave impedance, and joint density of the rock. The uniaxial compressive strength characterizes the compressive bearing capacity of the rock mass at that spatial location; the elastic modulus characterizes the deformation stiffness of the rock mass at that spatial location; the wave impedance characterizes the propagation impedance characteristics of blasting stress waves in the rock mass at that spatial location; and the joint density characterizes the degree of development of internal structural planes within the rock mass at that spatial location. The joint density is statistically obtained based on the number, frequency, or density of joints, fissures, and various structural planes within the corresponding spatial range. After voxelization discretization, the geological properties at different spatial locations around the target tunnel section are recorded in the corresponding voxel units in a unified data format, thereby forming an asymmetric three-dimensional geological grid that can characterize the spatial geological distribution features of the target tunnel section.

[0028] While forming the asymmetric three-dimensional geological mesh, an engineering context sequence must also be constructed. The engineering context sequence is constructed as follows: the target tunnel is continuously sliced ​​along its advancement axis, forming multiple sequentially arranged cross-sectional slices. For each cross-sectional slice, its contour geometry and adjacent environmental constraint information are extracted, and the data of the corresponding cross-sectional slices are arranged in chronological order according to the advancement direction to form the engineering context sequence. The contour geometry information includes the design contour shape, cross-sectional dimensions, and free face boundary information of the corresponding cross-section. The adjacent environmental constraint information includes the location of adjacent existing engineering objects, the distance between adjacent existing engineering objects and the corresponding cross-section, and the spatial orientation information of adjacent existing engineering objects relative to the corresponding cross-section. Adjacent existing engineering objects include adjacent existing caverns and adjacent existing structures. Thus, each frame in the engineering context sequence corresponds to a specific cross-sectional location along the advancement direction, recording the cross-sectional geometry and adjacent existing engineering object constraint states at that location, allowing for a continuous expression of the cross-sectional changes of the target tunnel along the advancement direction and the constraint relationships of adjacent existing engineering objects.

[0029] Through the above processing, the asymmetric three-dimensional geological mesh characterizes the rock mass type, structural distribution, and mechanical properties of the target tunnel section from the perspective of spatial geological distribution, while the engineering context sequence characterizes the geometric changes of the target tunnel section and the constraints of adjacent existing engineering objects from the perspective of longitudinal advancement correlation. The two types of input data formed in this way correspond to the two dimensions of spatial geological differences and longitudinal section correlation, respectively, and can express the actual state of the asymmetric section relatively completely.

[0030] Step S102: Input the asymmetric three-dimensional geological grid and engineering context sequence into the pre-constructed tunnel blasting parameter prediction model. The tunnel blasting parameter prediction model includes a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer, and a parameter output layer. The convolutional neural network layer is used to extract spatial features from the asymmetric three-dimensional geological grid. The long short-term memory network layer is used to extract temporal features from the engineering context sequence. The feature fusion layer is used to splice and jointly map the spatial and temporal features. The fully connected regression layer is used to perform regression calculations on the fused features. The parameter output layer is used to output the initial blasting parameters of the target tunnel section. The initial blasting parameters include at least the blasting parameters of the weak side and the blasting parameters of the hard side.

[0031] In step S102, the tunnel blasting parameter prediction model is a dual-input hybrid network structure, including a geological grid input end, a context sequence input end, a convolutional neural network branch, a long short-term memory network branch, a feature fusion layer, a fully connected regression layer, and a parameter output layer. Among them, the geological grid input end is connected to the convolutional neural network branch, the context sequence input end is connected to the long short-term memory network branch, the output end of the convolutional neural network branch and the output end of the long short-term memory network branch are jointly connected to the feature fusion layer, the output end of the feature fusion layer is connected to the fully connected regression layer, and the output end of the fully connected regression layer is connected to the parameter output layer. Specifically, the asymmetric 3D geological mesh is input into the convolutional neural network branch via the geological mesh input, while the engineering context sequence is input into the long short-term memory network branch via the context sequence input. The convolutional neural network branch employs a structure of sequentially stacked multi-layer 3D convolutional layers and pooling layers, while the long short-term memory network branch employs a structure of sequentially connected long short-term memory network units at least one layer. The spatial feature vector output by the convolutional neural network branch and the temporal feature vector output by the long short-term memory network branch are concatenated in the feature fusion layer to form a joint feature vector. After the joint feature vector is input into the fully connected regression layer, the initial blasting parameters of the target tunnel cross-section are output by the parameter output layer. The parameter output layer includes two parallel output sub-layers, one corresponding to the blasting parameters on the weak side and the other corresponding to the blasting parameters on the hard side. The above model structure consists of a spatial information processing pathway, a sequence information processing pathway, and a fusion regression pathway, thus forming a parameter prediction network structure for dual-source input of asymmetric cross-sections.

[0032] In one embodiment, before being used to predict parameters for the current target tunnel section, the tunnel blasting parameter prediction model is first constructed and trained based on historical blasting case data to obtain a pre-constructed tunnel blasting parameter prediction model. The historical blasting case data consists of multiple historical training samples. Each historical training sample includes input data and label data. The historical training samples are derived from geological modeling data, engineering environment data, actual blasting parameter data, and corresponding construction monitoring data collected in previous asymmetric section tunnel blasting projects. The input data includes an asymmetric three-dimensional geological grid and engineering context sequence with the same data structure as in step S101. The label data includes blasting parameter labels matching the corresponding historical samples. The blasting parameter labels include at least weak-side blasting parameter labels and hard-side blasting parameter labels, which respectively include at least charge quantity, hole spacing, and time delay. For historical samples with engineering monitoring records, engineering effect labels may further be included, which include at least one of vibration velocity, over-excavation, and under-excavation. By using historical training samples, a correspondence can be established between the input information of asymmetric cross sections and the actual blasting parameters and engineering effects.

[0033] Before model training begins, the historical training samples are preprocessed. Specifically, the asymmetric 3D geological mesh is converted into multi-channel tensor data that can be read by the convolutional neural network branch, and the engineering context sequence is converted into sequence tensor data that can be read by the long short-term memory network branch. The blasting parameter labels and engineering effect labels are normalized to ensure that different parameters and effect indicators participate in loss calculation on a unified numerical scale. Then, the preprocessed historical training samples are divided into training and validation sets, and the trainable parameters in the convolutional neural network branch, long short-term memory network branch, feature fusion layer, fully connected regression layer, and parameter output layer are initialized.

[0034] After data preprocessing and parameter initialization, a multi-task loss function is used to supervise the training of the tunnel blasting parameter prediction model. The multi-task loss function consists of the weak-side parameter prediction loss, the hard-side parameter prediction loss, the engineering effect prediction loss, and a regularization term, and its expression is: ; in, Indicates the total loss; The weighting coefficients represent the predicted loss of the weak-side parameters; This represents the loss predicted by the weak-side parameters; The weighting coefficients represent the predicted loss for the hard-side parameters; This indicates the loss predicted by the hard-side parameters; Indicates the first Weighting coefficients for predicting the loss of project outcomes; Indicates the first Predicted losses corresponding to each project performance indicator; Represents the weight coefficient of the regularization term; Represents the regularization term; This represents the total number of engineering performance indicators included in the training.

[0035] Among them, the weak-side parameter prediction loss This indicates the deviation between the weak-side explosion parameters output by the model and the true weak-side parameter labels. Represented as: ; in, Indicates the number of output parameters on the weak side; Indicating the weak side One real parameter label value; Indicating the weak side Each predicted parameter value.

[0036] Among them, the hard side parameter predicts the loss. This indicates the deviation between the hard-side blasting parameters output by the model and the true hard-side parameter labels. Represented as: ; in, Indicates the number of output parameters on the hard side; Indicates the hard side One real parameter label value; Indicates the hard side Each predicted parameter value.

[0037] Among them, the Project effect prediction loss Indicates the first The deviation between the predicted and actual values ​​of each project performance indicator Represented as: ; in, This indicates the number of samples in the current training batch; Indicates the first The corresponding sample of the th sample The true values ​​of each project performance indicator; Indicates the first The corresponding sample of the th sample Predicted values ​​of various project performance indicators.

[0038] Among them, the regularization term Used to constrain the magnitude of model parameters. Represented as: ; in, This represents the set of all trainable parameters in the tunnel blasting parameter prediction model; Represents the parameter set Any trainable parameter in the model. The advantages of the above loss function are as follows: First, it can simultaneously constrain the model's prediction accuracy for blasting parameters on both the weak and hard sides, so that the model does not only output averaged parameter results based on the overall cross-sectional conditions, but learns the parameter mapping relationship under different rock mass conditions on both sides. Second, by introducing engineering effect prediction loss, the model can learn the influence of parameter changes on engineering results such as vibration, over-excavation, and under-excavation while learning the correspondence between blasting parameters and input features. This makes the obtained parameter prediction results not only fit the historical parameter labels, but also better conform to the actual engineering constraints. Third, by setting a regularization term, it can suppress the excessive increase of model parameters, reduce the risk of overfitting during training, and improve the model's generalization ability when facing new asymmetric cross-sectional samples. Therefore, this loss function can improve the training effect of the tunnel blasting parameter prediction model from three aspects: parameter accuracy, engineering adaptability, and model stability.

[0039] After the loss function is determined, historical training samples are input into a deep learning model with a pre-defined structure for forward propagation calculation to obtain predicted values ​​of blasting parameters on the weak side, blasting parameters on the hard side, and engineering effects. In this embodiment, the deep learning model refers to the network structure to be trained constructed according to the aforementioned convolutional neural network layer, long short-term memory network layer, feature fusion layer, fully connected regression layer, and parameter output layer. After parameter training is completed, a tunnel blasting parameter prediction model is formed. Then, the total loss is calculated based on the blasting parameter labels and engineering effect labels. Then based on the total loss Backpropagation is performed to calculate the gradient of each trainable parameter with respect to the total loss, and the model parameters are updated using an optimization algorithm. As the training epochs increase, the parameters in the convolutional neural network branch, long short-term memory network branch, feature fusion layer, fully connected regression layer, and parameter output layer are continuously adjusted to gradually reduce the total loss. Training stops when the loss value on the validation set converges, or when the preset number of training epochs or the preset error threshold is reached, resulting in a pre-constructed tunnel blasting parameter prediction model. It should be noted that the engineering effect prediction loss is used to guide parameter updates during model training; after training, the model outputs initial blasting parameters during the prediction phase.

[0040] After completing the construction and training of the tunnel blasting parameter prediction model, the asymmetric three-dimensional geological grid and engineering context sequence obtained in step S101 are input into the trained tunnel blasting parameter prediction model to obtain the initial blasting parameters of the target tunnel section.

[0041] Specifically, the asymmetric 3D geological mesh is input into a convolutional neural network branch via the geological mesh input. The convolutional neural network branch performs layer-by-layer convolution and pooling on the multi-channel voxel data in the asymmetric 3D geological mesh, extracting the property distribution characteristics of the rock mass surrounding the target tunnel section within a spatial range, forming a spatial feature vector characterizing the spatial geological state of the target tunnel section. This spatial feature vector reflects the spatial differences in intensity distribution, structural distribution, and wave impedance distribution of the rock mass surrounding the target tunnel section. The engineering context sequence is input into a long short-term memory (LSTM) network branch via the context sequence input. The LSM network branch processes each section slice in the engineering context sequence sequentially according to the advance direction, encoding the contour geometry information and adjacent environmental constraint information corresponding to each section slice into temporal features, forming a temporal feature vector characterizing the cross-sectional correlation state of the target tunnel along the advance direction. This temporal feature vector reflects the cross-sectional change relationship of the target tunnel along the advance direction and the longitudinal change relationship of the constraints from adjacent structures.

[0042] After obtaining the spatial and temporal feature vectors, they are jointly input into the feature fusion layer for concatenation and joint mapping within a unified feature space, forming a joint feature vector. This joint feature vector is then input into a fully connected regression layer, which performs regression calculations to obtain parameter predictions corresponding to the current state of the target tunnel cross-section. After the fully connected regression layer completes its calculations, the parameter output layer outputs the initial blasting parameters for the target tunnel cross-section. The parameter output layer comprises two parallel output sub-layers: one outputs the initial blasting parameters for the weak side, and the other outputs the initial blasting parameters for the hard side. Both the initial blasting parameters for the weak and hard sides include at least the charge amount, hole spacing, and time delay.

[0043] Through the above processing, the spatial geological distribution characteristics in the asymmetric three-dimensional geological grid and the cross-sectional correlation characteristics in the engineering context sequence are jointly processed in the tunnel blasting parameter prediction model and converted into initial blasting parameters corresponding to the weak side and the hard side, respectively, so that the initial blasting parameters of the target tunnel cross section can correspond to the rock mass conditions on both sides and the longitudinal cross-sectional constraints.

[0044] Step S103: Perform blasting simulation and multi-objective optimization on the initial blasting parameters to obtain the final blasting parameters of the target tunnel cross section.

[0045] In step S103, blasting simulation and multi-objective optimization are performed on the initial blasting parameters to obtain the final blasting parameters for the target tunnel cross-section. This step is used to numerically verify and fine-tune the initial blasting parameters output in step S102. Since the initial blasting parameters are predictions based on an asymmetric three-dimensional geological grid, engineering context sequence, and a trained prediction model, although they can reflect the rock mass conditions on both sides of the target tunnel cross-section and the cross-sectional correlation state, it is still necessary to combine the actual geological distribution and blasting response patterns of the target tunnel cross-section to further determine whether these parameters meet the engineering requirements in terms of safety control, fragmentation effect, and surrounding rock stability. Therefore, it is necessary to simulate the blasting dynamic response of the initial blasting parameters and perform multi-objective optimization based on the simulation results to determine the final blasting parameters for the target tunnel cross-section.

[0046] Specifically, a numerical calculation model for blasting is first constructed based on the asymmetric three-dimensional geological grid obtained in step S101 and the initial blasting parameters output in step S102. Before constructing the numerical calculation model, the geometric, geological, and parameter data involved in the calculation are preprocessed. For geometric data, the tunnel design outline and borehole layout are cleaned and checked to ensure that the geometric boundaries are closed and meet the grid division requirements. For geological data, the discrete geological attributes in the asymmetric three-dimensional geological grid are extended into a continuous attribute field covering the entire three-dimensional computational space through spatial interpolation, and dynamic parameters such as wave impedance and damping ratio of the corresponding area are determined by combining lithology and dynamic empirical parameters. For all input data, the coordinate system, unit system, and data format are unified so that all types of data can be used together for numerical solution. After preprocessing, according to the spatial distribution range of weak and hard areas in the asymmetric three-dimensional geological grid, the corresponding mechanical properties and dynamic parameters are assigned to the corresponding unit sets in the numerical calculation model, so that the two sides of the target tunnel cross-section form material partitions with different mechanical properties in the calculation model. After completing the preprocessing and material property assignment, a three-dimensional computational domain is established based on the tunnel design outline, borehole layout, and surrounding rock range of the target tunnel section. This domain is then meshed to form an initial computational grid composed of multiple computational units. A relatively dense mesh is used in the borehole area, the area adjacent to the tunnel outline, and the boundary between weak and hard zones to improve the accuracy of the blasting response calculation. Next, the spatial location of each computational unit is determined based on the asymmetric three-dimensional geological grid. According to the geological property information at each unit's corresponding spatial location, corresponding rock mechanics and dynamic parameters are assigned to each unit, making each unit correspond to the material properties of either the weak or hard zone. Subsequently, the charge quantity, borehole spacing, and detonation time difference in the initial blasting parameters are mapped to the borehole charge state, borehole spacing layout, and detonation sequence settings, respectively, thus defining the blasting conditions within the initial computational grid. After these processes, a blasting numerical calculation model is obtained that simultaneously includes three-dimensional geometric boundaries, material zoning properties, and blasting condition parameters.

[0047] After constructing the numerical calculation model for blasting, the initial blasting parameters are substituted into the model to simulate the blasting dynamic response. During the simulation, blasting conditions are set according to the charge quantity, hole spacing, and initiation time difference corresponding to the initial blasting parameters, and the dynamic response of the rock mass surrounding the target tunnel section is calculated under explicit dynamic solution conditions. The calculation results include at least stress wave propagation results, vibration velocity field results, and plastic zone distribution results. Specifically, the stress wave propagation results reflect the propagation, reflection, and transmission of blasting stress waves in the rock mass surrounding the target tunnel section; the vibration velocity field results reflect the vibration response at key locations around the target tunnel section, and peak vibration velocities are extracted from the vibration velocity time history curves; the plastic zone distribution results reflect the extent of areas in the weak and hard surrounding rock that have entered a plastic damage state, and the volume or degree of damage in the plastic zone is statistically analyzed accordingly.

[0048] After obtaining the simulation results of the blasting dynamic response, the evaluation indicators required for multi-objective optimization are further extracted. For the hard side, the rock fragmentation size, fragmentation volume, or volumetric fracturing energy are used to characterize the fracturing effect on the hard side; for the weak side, the plastic zone volume, damage factor, or damage range of the key area are used to characterize the degree of damage to the surrounding rock on the weak side; for the side constrained by existing engineering objects, the peak vibration velocity at a preset monitoring point is used to characterize the vibration response level at the adjacent existing engineering objects, and the peak vibration velocity is compared with a safety threshold to determine whether the vibration response at the adjacent existing engineering objects meets the safety control requirements. In one embodiment, the safety threshold for the peak vibration velocity on the side adjacent to the tunnel is set to 2.5 cm / s.

[0049] After the evaluation indicators are determined, the initial blasting parameters are optimized using a multi-objective approach. The multi-objective optimization uses the charge amount, hole spacing, and detonation time difference as optimization variables, and aims to maximize the hard-side fracturing effect, minimize the damage to the weak-side surrounding rock, and ensure the vibration response at adjacent existing engineering objects meets a safety threshold. Since there are mutual constraints among the multiple optimization objectives, a multi-objective optimization algorithm is used to automatically iterate and optimize the parameter combinations. In one embodiment, the multi-objective optimization algorithm employs a second-generation non-dominated sorting genetic algorithm. After optimization begins, the multi-objective optimization algorithm first generates multiple sets of parameter combinations within a preset range of variation. Then, for each set of parameter combinations, it calls a blasting dynamic response simulation to obtain the corresponding hard-side fracturing effect index, weak-side surrounding rock damage index, and adjacent-tunnel vibration index. Next, based on the target values ​​corresponding to each parameter combination, it performs non-dominated sorting and optimal selection, retaining the better parameter combinations and generating a new round of parameter combinations. After multiple iterations, a Pareto optimal parameter set is output.

[0050] After obtaining the Pareto optimal parameter set, the final blasting parameters for the target tunnel section are determined by screening from the Pareto optimal parameter set based on the safety control requirements, fracturing effect requirements, and construction economic requirements of the target tunnel section. Through the above processing, the initial blasting parameters are transformed into final blasting parameters after blasting simulation verification and multi-objective optimization, ensuring that the final blasting parameters match the spatial geological distribution of the target tunnel section, while simultaneously considering the fracturing requirements on the hard side, the control of the surrounding rock on the weak side, and the vibration safety requirements near existing engineering objects.

[0051] Furthermore, preferably, after the actual blasting construction of the current target tunnel section is completed based on the final blasting parameters obtained in step S103, the model parameters in the tunnel blasting parameter prediction model can be updated based on the actual construction results of the current target tunnel section. It should be noted that the loss function used in the aforementioned model construction stage is used to complete the initial training of the model based on historical training samples. Its role is to establish the mapping relationship between the asymmetric three-dimensional geological grid, the engineering context sequence, and the initial blasting parameters. The feedback update in this step occurs after the actual blasting construction of the current target tunnel section has been completed. Its role is to correct the aforementioned mapping relationship based on the actual construction results of the current section, making the model closer to the actual engineering requirements in the parameter prediction process of subsequent sections. Since the initial blasting parameters output in step S102 are the prediction results directly given by the tunnel blasting parameter prediction model under the current section conditions, and the final blasting parameters obtained in step S103 are the parameter results that are more suitable for the current section construction requirements after combining the prediction results with blasting simulation and multi-objective optimization, the deviation between the initial blasting parameters and the final blasting parameters can be used as the basis for model parameter correction. Meanwhile, considering that the model's adaptability to the current cross-sectional conditions is reflected not only in whether the parameters themselves have been adjusted, but also in whether the simulated excavation results obtained based on this set of parameters are consistent with the actual excavation results, the intensity of this feedback update can be adaptively adjusted by using the difference between the actual cross-sectional profile deviation and the predicted cross-sectional profile deviation.

[0052] In one embodiment, the actual cross-sectional profile deviation of the current target tunnel cross-section is first determined based on the cross-sectional measurement results after actual blasting. Then, the corresponding predicted cross-sectional profile deviation is determined based on the blasting simulation results in step S103. Finally, the feedback update step size is determined based on the difference between the two, as expressed as: ; in, This indicates the feedback update step size for the current target tunnel section; The pre-defined base update step size refers to the default parameter update magnitude determined in advance based on the validation results in historical training samples after the initial training of the tunnel blasting parameter prediction model is completed and before the actual construction feedback data of the current target tunnel section is introduced. Specifically, after the initial training of the model is completed, the feedback update process can be simulated using the validation set in the historical training samples. The parameter correction effects corresponding to multiple candidate step sizes can be tested respectively, and the candidate step size that makes the model maintain stable convergence on the validation set and the parameter prediction error reduction meets the preset requirements can be selected as the base update step size. .therefore, These are not arbitrarily set constants, but rather preset parameters used to characterize the default correction strength of the model under normal feedback update conditions; This indicates the actual deviation of the cross-sectional profile of the current target tunnel section; This represents the predicted cross-sectional profile deviation obtained under the blasting simulation conditions in step S103 for the current target tunnel cross-section; This indicates a small positive number to prevent the denominator from being zero. During the feedback update process of the current target tunnel cross-section, the actual cross-section profile deviation and the predicted cross-section profile deviation are calculated using the same deviation representation method. To ensure the stability of the feedback update process, when the feedback update step size exceeds the preset upper limit, the feedback update step size is set to this preset upper limit.

[0053] Among them, the actual cross-sectional profile deviation The deviation of the actual excavated cross-section profile from the designed cross-section profile can be characterized by the following method: after the blasting construction of the current target tunnel cross-section is completed, the contour coordinate data of the actual excavated cross-section is collected using a 3D laser scanner, total station or cross-section profile measuring device. Then, the actual excavated cross-section profile is compared with the corresponding designed cross-section profile, and the average deviation value, maximum deviation value or area deviation value between the two is calculated. Any one of these deviation results is taken as the actual cross-section profile deviation. The reason for selecting the cross-sectional profile deviation is that the cross-sectional profile is the most direct result representation after blasting construction. It can reflect the formation of the excavation boundary after the release of blasting energy, as well as the overall matching degree between the current parameter configuration and the design target. Moreover, this quantity can be directly obtained through on-site measurement, the data source is clear, and the engineering operability is strong.

[0054] Predicted cross-sectional profile deviation The simulated excavation profile used to characterize the deviation of the simulated excavation profile from the design cross-sectional profile in step S103 can be obtained by: reconstructing the corresponding simulated excavation profile based on the plastic zone distribution results, fracture range results, or simulated excavation boundary results obtained in step S103; then comparing the simulated excavation profile with the design cross-sectional profile; calculating the average deviation value, maximum deviation value, or area deviation value between the two; and using any one of the deviation results as the predicted cross-sectional profile deviation amount. The reason for acting as with The corresponding comparison quantity is because step S103 itself has already performed blasting dynamic response simulation and multi-objective optimization on the initial blasting parameters. Therefore, the simulated excavation results obtained from this step can characterize the excavation profile state corresponding to the current parameter configuration under theoretical calculation conditions. and By comparing the simulation results with the actual results, we can determine the degree of consistency between the simulation results and the actual results, and thus further determine whether the current model accurately represents the blasting response under the cross-sectional conditions.

[0055] After determining the feedback update step size, the parameter deviation between the initial blasting parameter set and the final blasting parameter set is constructed as a feedback loss term. The feedback update step size is then used as the parameter update intensity corresponding to this feedback loss term to update the model parameters in the tunnel blasting parameter prediction model. The update of the model parameters in the tunnel blasting parameter prediction model after obtaining the feedback update step size is expressed as follows: ; in, Indicates the first The next step is to update the model parameters before the update. Indicates the first The model parameters are updated in the feedback; these model parameters include the trainable parameters in the convolutional neural network layer, long short-term memory network layer, feature fusion layer, fully connected regression layer, and parameter output layer. This represents the initial blasting parameter set output by the tunnel blasting parameter prediction model in step S102. This refers to the final blasting parameter set determined in step S103 after blasting simulation and multi-objective optimization, and actually used in the blasting construction of the current target tunnel section. Both the initial and final blasting parameter sets include at least the charge amount, hole spacing, and initiation time difference corresponding to the weak and hard sides. To ensure that different parameter items can characterize the overall differences between parameter sets within the same computational framework, during the calculation... First, the corresponding parameter items in the initial blasting parameter set and the final blasting parameter set are aligned one by one. Then, normalization is performed based on the value range, mean, variance or design allowable range of each parameter item in the historical training samples, so as to convert the parameter items with different dimensions and different numerical ranges into parameter components under a unified scale. Then, the normalized parameter components are arranged in a fixed order to form parameter set expression vectors corresponding to the initial blasting parameter set and the final blasting parameter set respectively, and the squared Euclidean distance between the two is calculated as the parameter deviation term.

[0056] Here we adopt The reason for the parameter deviation in the feedback update is as follows: the initial blasting parameters output in step S102 are the prediction results directly given by the model under the current cross-section conditions, which can represent the model's current judgment on the cross-section conditions; the final blasting parameters obtained in step S103 are the parameters that are more suitable for the current cross-section construction requirements, further combined with the asymmetric three-dimensional geological grid, blasting dynamic response simulation results, and multi-objective optimization results based on the initial blasting parameters. In other words, the deviation between the initial blasting parameters and the final blasting parameters essentially represents the difference between the model's direct prediction results and the feasible results of the project. Updating the model parameters based on this deviation allows the model to gradually learn the change law from direct prediction to engineering correction, thereby outputting initial blasting parameters that are closer to the actual engineering requirements under subsequent adjacent or similar cross-section conditions.

[0057] At the same time, based on the above parameter deviation terms, further utilizing The function of adaptively adjusting the feedback update intensity is to: when the actual cross-sectional profile deviation... Deviation from predicted cross-sectional profile When the difference between the values ​​is small, it indicates that the simulation results in step S103 are close to the actual construction results, and the current model's representation of the blasting response of the cross section is relatively accurate. At this time, the feedback update step size is close to the basic update step size, and the model parameters maintain a relatively stable correction state. and When the difference between the simulation and actual construction results increases, it indicates a decrease in the consistency between the simulation results and the actual construction results. The current model's representation of the blasting response under these conditions is insufficiently accurate. In this case, the feedback update step size is increased accordingly, thereby enhancing the correction of model parameters. Through this method, the model feedback update process not only considers the parameter deviation between the model's direct predictions and the final engineering results at the current cross-section, but also further considers the consistency between the simulated excavation results and the actual excavation results. This allows the model to be gradually corrected based on real engineering feedback during continuous construction, improving the adaptability of subsequent cross-section blasting parameter predictions to asymmetric geological conditions and on-site construction requirements.

[0058] In summary, combining Figure 2 This application provides a method for predicting tunnel blasting parameters for asymmetric cross-sections. First, an asymmetric three-dimensional geological mesh and engineering context sequence for the target tunnel cross-section are obtained. The asymmetric three-dimensional geological mesh characterizes the rock mass distribution, structural features, and mechanical properties of the target tunnel cross-section within a spatial range, while the engineering context sequence characterizes the cross-sectional changes of the target tunnel along the advancement direction and the constraints of the adjacent environment. Then, the asymmetric three-dimensional geological mesh and engineering context sequence are input into a pre-constructed tunnel blasting parameter prediction model. Through the collaborative processing of convolutional neural network layers, long short-term memory network layers, feature fusion layers, fully connected regression layers, and parameter output layers, initial blasting parameters corresponding to the weak and hard sides are obtained, respectively. Next, blasting dynamic response simulation is performed on the initial blasting parameters, extracting simulation results such as stress wave propagation, vibration velocity field, and plastic zone distribution. Based on multiple objectives such as the hard side fracture effect, the weak side surrounding rock damage, and vibration control on the adjacent side, the initial blasting parameters are optimized using a multi-objective approach to obtain the final blasting parameters for the target tunnel cross-section.

[0059] Because this application first uses an asymmetric three-dimensional geological grid to digitally represent the spatial distribution, structural differences, and mechanical property differences of the rock masses on both sides of the target tunnel cross-section, and then uses an engineering context sequence to characterize the cross-sectional change relationship and the adjacent environmental constraints along the advancement direction, the spatial geological differences and longitudinal correlation state of the target tunnel cross-section can be simultaneously incorporated into the parameter prediction process. Furthermore, the tunnel blasting parameter prediction model does not directly provide uniform parameters based on the overall cross-sectional average conditions, but rather uses a dual-input hybrid network structure to jointly model spatial and temporal information, and outputs the initial blasting parameters for the weak side and the hard side respectively. This allows the predicted parameters to correspond to the different rock mass conditions on both sides of the cross section. Based on this, by combining blasting dynamic response simulation and multi-objective optimization, the stress wave propagation, vibration response, and surrounding rock damage of the initial blasting parameters under real asymmetric geological conditions are verified and adjusted. This ensures that the final blasting parameters can not only more accurately reflect the parameter requirements under different working conditions on both sides of the cross section, but also simultaneously take into account the requirements for fracture on the hard side, the control of the surrounding rock on the weak side, and the vibration safety requirements on the side adjacent to the tunnel. This improves the matching degree between the predicted blasting parameters and the actual geological conditions and engineering constraints of the asymmetric cross section.

[0060] Figure 3 This is a structural block diagram of a tunnel blasting parameter prediction system for asymmetric cross sections provided in one embodiment of this application. The system includes at least the following modules: The data acquisition module is used to acquire the asymmetric three-dimensional geological grid and engineering context sequence of the target tunnel cross section. The asymmetric three-dimensional geological grid is used to characterize the spatial geological distribution characteristics of the target tunnel cross section, and the engineering context sequence is used to characterize the cross section correlation characteristics of the target tunnel along the advancement direction. The model prediction module is used to input the asymmetric 3D geological grid and engineering context sequence into a pre-constructed tunnel blasting parameter prediction model. The tunnel blasting parameter prediction model includes a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer, and a parameter output layer. The convolutional neural network layer is used to extract spatial features from the asymmetric 3D geological grid, the long short-term memory network layer is used to extract temporal features from the engineering context sequence, the feature fusion layer is used to splice and jointly map the spatial and temporal features, the fully connected regression layer is used to perform regression calculations on the fused features, and the parameter output layer is used to output the initial blasting parameters of the target tunnel section. The initial blasting parameters include at least the blasting parameters of the weak side and the blasting parameters of the hard side. The parameter determination module is used to perform blasting simulation and multi-objective optimization on the initial blasting parameters to obtain the final blasting parameters of the target tunnel cross section.

[0061] For relevant details, please refer to the above method implementation examples.

[0062] Figure 4 This is a block diagram of an electronic device provided in one embodiment of this application. The electronic device includes at least a processor 401 and a memory 402.

[0063] The processor 401 executes computer program instructions stored in the memory 402 to implement the tunnel blasting parameter prediction method for asymmetric cross sections provided in this embodiment. The processor 401 may include one or more processing cores, such as a 4-core processor or an 8-core processor. The processor 401 may be implemented using at least one hardware form selected from DSP, FPGA, and PLA. In some embodiments, the processor 401 may further include a dedicated processing unit for performing neural network calculations, feature extraction, parameter regression calculations, and optimization calculations.

[0064] The memory 402 may include one or more computer-readable storage media, which may be non-transitory. The memory 402 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. The memory 402 is used to store at least one computer program instruction, which, when executed by the processor 401, causes the processor 401 to implement the tunnel blasting parameter prediction method for asymmetric cross-sections provided in this embodiment. Exemplarily, the memory 402 may store program instructions for obtaining asymmetric three-dimensional geological grids and engineering context sequences, program instructions for constructing and calling tunnel blasting parameter prediction models, and program instructions for performing blasting simulation and multi-objective optimization.

[0065] In some embodiments, the electronic device may further include other functional components connected to the processor 401 and the memory 402 to support functions such as data input, data storage, result output, or network communication. This embodiment does not limit the specific composition of the electronic device.

[0066] Optionally, this application also provides a computer-readable storage medium storing a program that is loaded and executed by a processor to implement the tunnel blasting parameter prediction method for asymmetric cross sections described in the above method embodiments.

[0067] Optionally, this application also provides a computer product including a computer-readable storage medium storing a program, which is loaded and executed by a processor to implement the tunnel blasting parameter prediction method for asymmetric cross sections described in the above method embodiments.

[0068] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0069] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting tunnel blasting parameters for asymmetric cross-sections, characterized in that, The method includes: Obtain an asymmetric three-dimensional geological grid and an engineering context sequence for the target tunnel cross-section, wherein the asymmetric three-dimensional geological grid is used to characterize the spatial geological distribution characteristics of the target tunnel cross-section, and the engineering context sequence is used to characterize the cross-sectional correlation characteristics of the target tunnel along the advancement direction; The asymmetric three-dimensional geological grid and the engineering context sequence are input into a pre-constructed tunnel blasting parameter prediction model. The tunnel blasting parameter prediction model includes a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer, and a parameter output layer. The convolutional neural network layer is used to extract spatial features from the asymmetric three-dimensional geological grid. The long short-term memory network layer is used to extract temporal features from the engineering context sequence. The feature fusion layer is used to concatenate and jointly map the spatial features and the temporal features. The fully connected regression layer is used to perform regression calculations on the fused features. The parameter output layer is used to output the initial blasting parameters of the target tunnel section. The initial blasting parameters include at least the blasting parameters of the weak side and the blasting parameters of the hard side. The initial blasting parameters are subjected to blasting simulation and multi-objective optimization to obtain the final blasting parameters of the target tunnel cross section.

2. The method for predicting tunnel blasting parameters for asymmetric cross-sections according to claim 1, characterized in that, The acquisition of the asymmetric three-dimensional geological mesh and engineering context sequence of the target tunnel cross-section includes: Acquire three-dimensional geometric morphology data, geological exploration data, and rock mechanics parameter data of the target tunnel cross section; A three-dimensional spatial geometric model of the target tunnel cross-section is established based on the aforementioned three-dimensional geometric morphology data; Based on the geological survey data, the spatial location range of joints, fissures, weak interlayers and various structural planes in the rock mass surrounding the target tunnel section is determined, and the rock mass surrounding the target tunnel section is spatially located and divided into regions according to the lithological distribution and structural plane distribution, resulting in multiple rock mass regions; Based on the rock mechanics parameter data, corresponding mechanical parameters are assigned to each of the rock mass regions to form a three-dimensional digital geological model of the target tunnel cross section; The three-dimensional digital geological model is discretized using voxelization to obtain an asymmetric three-dimensional geological mesh of the target tunnel cross section.

3. The method for predicting tunnel blasting parameters for asymmetric cross-sections according to claim 1, characterized in that, The acquisition of the asymmetric three-dimensional geological mesh and engineering context sequence of the target tunnel cross-section includes: The target tunnel is continuously sliced ​​along its advancement axis to form multiple sequentially arranged cross-sectional slices; For each cross-sectional slice, extract the contour geometry information of the cross-section and the constraint information of the surrounding environment; The contour geometry information and adjacent environmental constraint information corresponding to each cross-section slice are arranged in chronological order according to the direction of advancement to obtain the engineering context sequence of the target tunnel cross-section.

4. The method for predicting tunnel blasting parameters for asymmetric cross-sections according to claim 1, characterized in that, The tunnel blasting parameter prediction model is a hybrid network structure oriented towards dual-source input of asymmetric cross-sections, including two input ends corresponding to the asymmetric three-dimensional geological grid and the engineering context sequence, respectively, and a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer and a parameter output layer connected in sequence. The input end of the asymmetric three-dimensional geological grid is connected to a convolutional neural network layer to encode the spatial geological distribution information of the target tunnel cross-section. The input end of the engineering context sequence is connected to a long short-term memory network layer to encode the cross-sectional correlation information of the target tunnel along the advancement direction. The output ends of the convolutional neural network layer and the long short-term memory network layer are jointly connected to a feature fusion layer. The feature fusion layer is used to splice and jointly map spatial features and temporal features. The feature fusion layer is connected to a fully connected regression layer. The fully connected regression layer is connected to a parameter output layer. The parameter output layer includes two parallel output sub-layers corresponding to the weak side and the hard side, respectively, to output the initial blasting parameters of the weak side and the initial blasting parameters of the hard side.

5. The method for predicting tunnel blasting parameters for asymmetric cross-sections according to claim 1, characterized in that, The construction process of the tunnel blasting parameter prediction model includes: Acquire historical training samples, which include asymmetric three-dimensional geological grids, engineering context sequences, blasting parameter labels, and engineering effect labels; Historical training samples are input into a deep learning model with a pre-defined structure for training to obtain predicted values ​​of blasting parameters on the weak side and blasting parameters on the hard side. The parameter prediction loss is calculated based on the blasting parameter label, and the engineering effect prediction loss is calculated based on the engineering effect label. The parameter prediction loss, engineering effect prediction loss and regularization term are combined to form the total loss function, and the trainable parameters in the tunnel blasting parameter prediction model are updated according to the total loss function. When the preset training termination condition is met, a pre-constructed tunnel blasting parameter prediction model is obtained.

6. The method for predicting tunnel blasting parameters for asymmetric cross-sections according to claim 5, characterized in that, The total loss function is composed of the weak-side parameter prediction loss, the hard-side parameter prediction loss, the engineering effect prediction loss, and a regularization term, and its expression is: ； in, Indicates the total loss; The weighting coefficients represent the predicted loss of the weak-side parameters; This represents the loss predicted by the weak-side parameters; The weighting coefficients represent the predicted loss for the hard-side parameters; This indicates the loss predicted by the hard-side parameters; Indicates the first Weighting coefficients for predicting the loss of project outcomes; Indicates the first Predicted losses corresponding to each project performance indicator; Represents the weight coefficient of the regularization term; Represents the regularization term; This represents the total number of engineering performance indicators included in the training. Among them, the weak-side parameter prediction loss This indicates the deviation between the weak-side explosion parameters output by the model and the true weak-side parameter labels. Represented as: ； in, Indicates the number of output parameters on the weak side; Indicating the weak side One real parameter label value; Indicating the weak side One predicted parameter value; Among them, the hard side parameter predicts the loss. This indicates the deviation between the hard-side blasting parameters output by the model and the true hard-side parameter labels. Represented as: ； in, Indicates the number of output parameters on the hard side; Indicates the hard side One real parameter label value; Indicates the hard side One predicted parameter value; Among them, the Project effect prediction loss Indicates the first The deviation between the predicted and actual values ​​of each project performance indicator Represented as: ； in, This indicates the number of samples in the current training batch; Indicates the first The corresponding sample of the th sample The true values ​​of each project performance indicator; Indicates the first The corresponding sample of the th sample Predicted values ​​of each project performance indicator; Among them, the regularization term Used to constrain the magnitude of model parameters. Represented as: ； in, This represents the set of all trainable parameters in the tunnel blasting parameter prediction model; Represents the parameter set Any trainable parameter in the dataset.

7. The method for predicting tunnel blasting parameters for asymmetric cross-sections according to claim 1, characterized in that, The process of performing blasting simulation and multi-objective optimization on the initial blasting parameters to obtain the final blasting parameters for the target tunnel cross-section includes: A numerical calculation model for blasting is constructed based on an asymmetric three-dimensional geological grid of the target tunnel section and initial blasting parameters. Based on the aforementioned blasting numerical calculation model, blasting dynamic response simulation was performed to obtain stress wave propagation results, vibration velocity field results, and plastic zone distribution results. Based on the stress wave propagation results, vibration velocity field results, and plastic zone distribution results, the hard side fracture effect index, the weak side surrounding rock damage index, and the vibration index of the tunnel side are determined respectively. Using the charge amount, hole spacing, and detonation time difference as optimization variables, and taking the maximization of the hard side fracture effect index, the minimization of the soft side surrounding rock damage index, and the satisfaction of the safety threshold of the vibration index at the adjacent existing engineering object as optimization objectives, the initial blasting parameters are optimized in a multi-objective manner to obtain the final blasting parameters of the target tunnel section.

8. A tunnel blasting parameter prediction system for asymmetric cross-sections, characterized in that, include: The data acquisition module is used to acquire the asymmetric three-dimensional geological grid and engineering context sequence of the target tunnel cross section. The asymmetric three-dimensional geological grid is used to characterize the spatial geological distribution characteristics of the target tunnel cross section, and the engineering context sequence is used to characterize the cross section correlation characteristics of the target tunnel along the advancement direction. The model prediction module is used to input the asymmetric three-dimensional geological grid and the engineering context sequence into a pre-constructed tunnel blasting parameter prediction model. The tunnel blasting parameter prediction model includes a convolutional neural network layer, a long short-term memory network layer, a feature fusion layer, a fully connected regression layer, and a parameter output layer. The convolutional neural network layer is used to extract spatial features from the asymmetric three-dimensional geological grid. The long short-term memory network layer is used to extract temporal features from the engineering context sequence. The feature fusion layer is used to concatenate and jointly map the spatial features and the temporal features. The fully connected regression layer is used to perform regression calculations on the fused features. The parameter output layer is used to output the initial blasting parameters of the target tunnel section. The initial blasting parameters include at least the weak side blasting parameters and the hard side blasting parameters. The parameter determination module is used to perform blasting simulation and multi-objective optimization on the initial blasting parameters to obtain the final blasting parameters of the target tunnel cross section.

9. An electronic device, characterized in that, The device includes a processor and a memory; the memory stores a program that is loaded and executed by the processor to implement a method for predicting tunnel blasting parameters for asymmetric cross sections as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a program that, when executed by a processor, is used to implement a method for predicting tunnel blasting parameters for asymmetric cross sections as described in any one of claims 1 to 7.

Images