A tunnel jumbo rock breaking seismic source advanced prediction method and system

Abstract

The application discloses a tunnel rock drilling jumbo rock breaking seismic source advanced prediction method and system, and relates to the technical field of tunnel engineering, and comprises the following steps: adaptively coding obtained rock breaking seismic data to determine the position of effective reflection information and extract the effective reflection information; converting clean rock breaking seismic records into a gray-scale image, detecting the sudden edges of the gray-scale gradients of each channel in the gray-scale image in the time direction, and extracting the first arrival wave according to the detected edge points; and obtaining the inversion imaging result of the wave velocity field by using the mapping model of the trained travel time information and the velocity field according to the travel time information of the first arrival wave. The signal-to-noise ratio of the seismic record is improved through the reconstruction and noise reduction processing of the rock breaking seismic data. The inversion accuracy is enhanced by adding a physical equation constraint, and high-precision mapping of the velocity field is realized.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering technology, and in particular to a method and system for predicting the seismic source of rock breaking by a tunnel drilling rig. Background Technology

[0002] Among existing advanced prediction methods for drill-and-blast tunnels, such as seismic wave prediction, while effective in forecasting tunnel geological conditions, they occupy significant space in the sidewalls and are time-consuming, impacting construction efficiency and schedule. With the innovation of tunnel construction technology, large-scale mechanized construction equipment, such as rock drilling rigs, is increasingly widely used. As a crucial piece of equipment in mechanized drill-and-blast tunnel construction, the rock drilling rig plays a vital role in processes such as blasting drilling, anchor drilling, and advanced drilling. Its drilling vibrations are characterized by high drilling efficiency and large rock-breaking energy, making it a potential source for advanced geological prediction.

[0003] However, in advanced geological prediction, existing inversion methods rely on data-driven models, and the inversion solutions do not conform to physical laws, making it difficult to meet the high-precision mapping requirements of rock-breaking source seismic data. Furthermore, the use of fixed window selection and conventional location encoding methods fails to accurately represent the spatiotemporal distribution of effective information in rock-breaking seismic data. Secondly, existing denoising techniques focus on single local convolutional features or global modeling, making it difficult to consider the waveform details of rock-breaking source seismic records, leading to waveform and arrival time shifts after denoising. Additionally, existing first-arrival determination methods rely on experience or simple thresholds, making them sensitive to noise and unable to guarantee the accuracy of first-arrival wavenumber results. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a method and system for predicting rock-breaking seismic sources using a tunnel drilling rig. By reconstructing and denoising the seismic data of the rock-breaking source, the signal-to-noise ratio of the seismic record is improved. Furthermore, by incorporating physical equation constraints, the accuracy of the inversion is enhanced, thereby achieving high-precision mapping of the velocity field.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting the seismic source of rock breaking using a tunnel drilling rig, comprising: After performing multi-scale convolution and perceptual coding on the acquired rock-breaking source seismic data, the location of effective reflection information is determined by adaptive location coding. The effective reflection information is then extracted through a cross-attention mechanism. After denoising the effective reflection information, a clean rock-breaking source seismic record is obtained. The clean rock-breaking source seismic record is converted into a grayscale image. The abrupt edges of the grayscale gradient in each channel of the grayscale image are detected along the time direction, so as to extract the first arrival wave based on the detected edge points. Based on the first arrival wave travel time information, the inversion imaging result of the wave velocity field is obtained by using the mapping model of the trained travel time information and the velocity field. During the training process, the wave velocity field is obtained by mapping the travel time information, and the theoretical travel time information is calculated based on the wave velocity field. The mapping model is then updated by wave velocity loss, travel time loss and physical loss.

[0006] As an alternative implementation, the acquired rock-breaking source seismic data is first enhanced by features, then nonlinearly integrated with local features through multi-scale convolution processing, and finally weighted by perceptual coding to generate position-specific weights. These weighted weights are then fused, and adaptive position coding is applied to both the outputs of the multi-scale convolution and the perceptual coding. The adaptive position coding is as follows: ;in, X This represents the input for adaptive positional encoding; G Indicates the weight of the gating mechanism. APE Indicates the adaptive position coding parameters. PE This represents the sinusoidal position encoding parameters.

[0007] As an alternative implementation, after determining the location of the effective reflection information, the effective reflection information is extracted by a network consisting of a multi-layer encoder and a multi-layer decoder. The multi-layer encoder is composed of self-attention and a feedforward neural network, and the multi-layer decoder is composed of self-attention, cross-attention and a feedforward neural network. The encoder output feature sequence after adaptive position encoding is used as the key matrix. K e Sum matrix V e The decoder generated by perceptual encoding is input to the initial feature sequence as the query matrix. Q d ; by query matrix Q d AND key matrix K e The relevance score is calculated by scaling the dot product, and then the attention weight matrix is ​​obtained using the Softmax function. The attention weight matrix is ​​then compared with the value matrix. V e We perform weighted summation to extract effective reflection information.

[0008] As an alternative implementation, the effective reflection information is denoised using a trained CNN-Swin-Transformer network; the CNN-Swin-Transformer network uses a CNN network as a front-end encoder and a back-end decoder, with a Swin-Transformer network fused in between. The noisy effective reflection information is input into the front-end encoder to extract and encode the local waveform and local continuous features of the phase axis. The encoded features are then input into the Swin-Transformer network, which learns and represents the global correlation of seismic information across channels and time through multi-head attention and feedforward neural networks. The represented features are then input into the back-end decoder, which uses convolutional layers to recover the seismic record and reconstruct the clean rock-breaking source seismic record.

[0009] As an alternative implementation, the clean rock-breaking source seismic record is normalized in a single channel and converted into a grayscale image. The first arrival wave is represented on the grayscale image as the first abrupt change edge from low grayscale to continuous high grayscale energy band. Specifically, the region with grayscale value below a preset background threshold is defined as a low grayscale region, the region with grayscale value above a preset energy threshold along the time direction or adjacent gather direction is defined as a continuous high grayscale energy band, and the position where the absolute value of the grayscale gradient between adjacent pixels is greater than a preset gradient threshold is defined as the abrupt change position. Using an edge detection method, abrupt edges of grayscale gradients in each channel are detected along the time direction. The first strong edge point obtained from scanning from shallow to deep in each channel is selected as the location of the first arrival wave, generating preliminary detection results. Among them, the shallow edge corresponds to the time sampling point position where the sampling time is less than the set minimum threshold, and the deep edge corresponds to the time sampling point position where the sampling time is greater than the set maximum threshold. Strong edge points are defined as edge points with gradient amplitude greater than the preset edge threshold and continuity in adjacent channels. Median filtering is used to eliminate outlier interference, and combined with edge tracking algorithm, local deviations are corrected along the in-phase axis extension direction to achieve automatic first arrival picking. Outliers are isolated points whose deviation from the first arrival time of adjacent gathers exceeds a preset time difference threshold and do not satisfy the in-phase axis continuity.

[0010] As an alternative implementation method, the time-travel loss term is: ; in, This is the travel time loss term corresponding to the m-th iteration; The number of seismic sources used in the observation system; Let be the theoretical first arrival travel time vector of the s-th source calculated in the m-th iteration; This is a transpose operation; Let be the observation first arrival travel time corresponding to the s-th earthquake source.

[0011] Secondly, the present invention provides a tunnel drilling rig rock-breaking seismic source prediction system, comprising: The denoising module is configured to perform multi-scale convolution and perceptual coding on the acquired rock-breaking source seismic data, determine the location of effective reflection information through adaptive position coding, extract the effective reflection information through a cross-attention mechanism, and obtain a clean rock-breaking source seismic record after denoising the effective reflection information. The extraction module is configured to convert clean rock-breaking source seismic records into grayscale images, detect abrupt edges of grayscale gradients in each channel of the grayscale image along the time direction, and extract the first arrival wave based on the detected edge points. The inversion module is configured to obtain the inversion imaging result of the wave velocity field based on the first arrival wave travel time information and the mapping model of the trained travel time information and velocity field. During the training process, the wave velocity field is mapped from the travel time information, and the theoretical travel time information is calculated from the wave velocity field. The mapping model is then updated through wave velocity loss, travel time loss and physical loss.

[0012] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0013] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0014] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a method for predicting rock-breaking seismic sources using a tunnel drilling rig. It innovatively proposes a neural network-based method for reconstructing and denoising seismic data from rock-breaking sources. First, the method utilizes a neural network to extract effective information from the direct and reflected waves in the original seismic records of the rock-breaking sources. Based on this, noise interference in the seismic records is identified and suppressed, significantly improving the signal-to-noise ratio of the seismic records. Through intelligent processing methods, rapid and effective interpretation of rock-breaking source seismic data is achieved. Simultaneously, an innovative data- and physical-driven inversion method for rock-breaking source seismic data is proposed. By incorporating physical equation constraints, the accuracy of the inversion is enhanced, achieving high-precision mapping of the velocity field.

[0016] To address the challenges of existing tunnel seismic wave prediction methods, such as large space requirements and long prediction times, this invention proposes a tunnel drilling rig rock-breaking seismic source prediction system. It innovatively utilizes the drilling vibrations of the drilling rig as the seismic source for prediction. Three-component geophones are deployed at the sidewalls and the front end of the drill arm to receive seismic data. By processing the seismic data, adverse geological formations ahead of the tunnel face are predicted, ultimately enabling mechanized tunnel drilling and exploration using the drill-and-blast method.

[0017] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart of the tunnel drilling rig rock breaking seismic source advance prediction method provided in Embodiment 1 of the present invention; Figure 2 This is a structural diagram of the effective information extraction model provided in Embodiment 1 of the present invention; Figure 3 The rock-breaking source seismic data is the input to the effective information extraction model provided in Embodiment 1 of the present invention. Figure 4 The effective reflection information output by the effective information extraction model provided in Embodiment 1 of the present invention; Figure 5 This is a diagram of the denoising neural network structure provided in Embodiment 1 of the present invention; Figure 6 The noisy seismic data is the input to the denoising neural network provided in Embodiment 1 of the present invention; Figure 7 This is a clean rock-breaking source seismic record output by the denoising neural network provided in Embodiment 1 of the present invention; Figure 8 This is a diagram of the inversion imaging network structure provided in Embodiment 1 of the present invention; Figure 9 This is a schematic diagram of the initial wave velocity field provided in Embodiment 1 of the present invention; Figure 10 This is a schematic diagram of a three-component detector provided in Embodiment 2 of the present invention; Figure 11 This is a schematic diagram of a wireless transmission node provided in Embodiment 2 of the present invention; Figure 12This is a schematic diagram of the signal acquisition host provided in Embodiment 2 of the present invention; Figure 13 This is a schematic diagram of a tunnel observation system for predicting rock-breaking seismic sources using a tunnel drilling rig, provided in Embodiment 2 of the present invention. Among them, 1. Three-component detector; 2. Threaded hole; 3. Detector interface; 4. Wireless transmission node; 5. Antenna; 6. Electronic components; 7. Node interface; 8. Signal acquisition host; 9. Pilot observation system; 10. Sidewall observation system; 11. Connecting line; 12. Rock drilling rig; 13. Drill arm; 14. Powerful magnet. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “including”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0024] Example 1 This embodiment provides a method for predicting the seismic source of rock breaking using a tunnel drilling rig, including: After performing multi-scale convolution and perceptual coding on the acquired rock-breaking source seismic data, the location of effective reflection information is determined by adaptive location coding. The effective reflection information is then extracted through a cross-attention mechanism. After denoising the effective reflection information, a clean rock-breaking source seismic record is obtained. The clean rock-breaking source seismic record is converted into a grayscale image. The abrupt edges of the grayscale gradient in each channel of the grayscale image are detected along the time direction, so as to extract the first arrival wave based on the detected edge points. Based on the first arrival wave travel time information, the inversion imaging result of the wave velocity field is obtained by using the mapping model of the trained travel time information and the velocity field. During the training process, the wave velocity field is obtained by mapping the travel time information, and the theoretical travel time information is calculated based on the wave velocity field. The mapping model is then updated by wave velocity loss, travel time loss and physical loss.

[0025] The following is combined with Figure 1 The method of this embodiment will be described in detail.

[0026] In this embodiment, rock-breaking source seismic data are first acquired by collecting signals from the pilot geophone and the sidewall geophone. Then, the effective information extraction model after training is used to extract effective reflection information from the rock-breaking source seismic data.

[0027] Specifically: Based on the principle of random generation, geological models such as faults, fracture zones, and filled karst caves are randomly generated within a certain rule range. Furthermore, a seismic record database of rock-breaking sources and hammer-driven sources for different adverse geological body models is established through forward modeling data simulation. We used the constructed seismic record database to train an effective information extraction model based on an improved Transformer neural network.

[0028] After training is completed, input the time series data of the original rock-breaking source seismic data, and output the time series data of the effective reflection information of the rock-breaking source; the time series data is a txt file with 5000 lines × the number of detector channels.

[0029] In this embodiment, the traditional Transformer neural network is improved to address the problems of cluttered wavefields and difficulty in extracting continuous effective information in rock-breaking source seismic data. Features such as feature enhancement, multi-scale convolution, perceptual decoder, adaptive encoding, self-attention, and cross-attention mechanisms are added. Weighted MSE (Mean Squared Error) loss and cross-correlation loss are used as the loss functions of the network, which can effectively identify effective reflection information in rock-breaking source seismic data.

[0030] like Figure 2 As shown, specifically: (1) such as Figure 3 The multi-channel rock-breaking source seismic data shown is first processed by a feature enhancement module to strengthen local waveform changes and energy information after inputting effective information into the model, highlighting waveform jump and abrupt change characteristics. After feature enhancement, the number of channels is increased from 10 to 40.

[0031] (2) The output of the feature enhancement module is transmitted to the multi-scale convolution module, and the local features are nonlinearly integrated through scale fusion. At the same time, the output of the feature enhancement module is transmitted to the perceptual encoder to generate positional weights for the data sequence, and the weighted fusion is used to obtain the decoder input initialization.

[0032] (3) Before the output of the multi-scale convolution and perceptual decoder enters the main network, the temporal information of the data is labeled by the adaptive coding module. The principle of adaptive positional coding is as follows: (1); in, X Indicates the input tensor; G Indicates the weight of the gating mechanism. APE Indicates the adaptive position coding parameters. PE This represents the sinusoidal position encoding parameters.

[0033] Specifically: First, the input tensor X is mapped to a feature sequence. Then, the importance of each feature at each time step and each channel is adaptively weighted through a gating mechanism G. The position response matrix is ​​constructed by fusing the APE with the sinusoidal position coding parameter PE. After multiplying the response matrix element by element with the branch features, the positions corresponding to continuous effective information such as the first arrival wave and the reflected wave receive higher responses, while the positions corresponding to random noise receive lower responses, thereby completing the effective information position localization.

[0034] (4) The network consists of a multi-layer encoder and a multi-layer decoder. The multi-layer encoder is composed of self-attention and a feedforward neural network (FNN). The multi-layer decoder is composed of self-attention, cross-attention and a feedforward neural network (FNN).

[0035] After receiving the output of the adaptive encoding, the encoder ensures the internal consistency of the data sequence through a self-attention mechanism and performs shallow local representation learning on the effective reflective information features.

[0036] Based on self-attention processing, the decoder retrieves the most relevant and effective information at the current moment through cross-attention, thereby achieving fine alignment and reconstruction of local features.

[0037] (5) In the output stage, the network entity restores the seismic data to the original 10 channels through nonlinear mapping, realizing the accurate extraction of effective reflection information from rock-breaking source seismic data. The output effective reflection information is as follows: Figure 4 As shown.

[0038] In this embodiment, the principle of the cross-attention mechanism is as follows: (2); in, Qd Represents the query matrix. K e Represents the bond matrix. V e Representative value matrix, d k The dimension representing each attention head.

[0039] During the extraction process, the encoder output feature sequence after adaptive position encoding is used as the key matrix. K e Sum matrix V e The feature sequence generated by the perceptual encoder as the input to the decoder is used as the query matrix. Q d First by Q d and K e The relevance score is calculated by scaling the dot product, and then the attention weight matrix is ​​obtained using the Softmax function. This attention weight matrix is ​​then used to... V e By performing a weighted summation, attention is prioritized for locations with effective reflection information, while noise components are suppressed, thereby enabling the extraction of effective reflection information.

[0040] In this embodiment, the weighted MSE loss function enables the network to focus on the identified effective reflection information during training, while the cross-correlation loss function ensures that the waveform results in the seismic records output by the network remain consistent.

[0041] The principles of the two loss functions are as follows: (3); in, B Represents batch size. L Represents the sequence length. C Represents the number of channels. W b,t,c It is a weight matrix. Represents label data, Y b,t,c This indicates the output of the model.

[0042] (4); in, Indicates the first b Label data for each sample, Y b Indicates the first b The model output data for each sample.

[0043] In this embodiment, the effective reflection information is denoised using a trained denoising neural network based on CNN-Swin-Transformer to obtain a clean rock-breaking source seismic record.

[0044] Specifically: When training the CNN-Swin-Transformer network using the constructed seismic record database, effective reflection information is used as the label. Random noise is added to the effective reflection information to simulate the interference of effective reflection information in the actual field, forming a noisy reconstructed seismic record, which is used as input data. After outputting clean rock-breaking source seismic records, they are compared with the labeled data to optimize the hyperparameters of the network training. After training, input effective reflection information containing real data noise, and output clean rock-breaking source seismic records.

[0045] In this embodiment, the CNN-Swin-Transformer network adopts a network architecture that combines a CNN (Convolutional Neural Network) network with a Swin-Transformer network (which combines the local feature extraction capability of a CNN with the global modeling capability of a Transformer). Essentially, the CNN network serves as both the front-end encoder and the back-end decoder, with the Swin-Transformer network integrated in between, constructing a sequential network for local / global feature extraction and fusion.

[0046] like Figure 5 As shown, specifically: (1) The CNN encoder consists of a shallow convolution head and a residual convolution block to extract local features of noisy seismic data. The shallow convolution head consists of a 3×3 2D convolution block to extract local continuous information of the phase axis and preserve phase and arrival details. The residual convolution block consists of two layers of 3×3 2D convolution blocks. Through residual connection, waveform information is better preserved.

[0047] (2) The CNN decoder consists of pooling layers and convolutional layers, which are used to compensate for detailed information and avoid time drift and waveform blunting caused by noise reduction.

[0048] (3) The Swin-Transforme structure consists of four layers, each consisting of multi-head attention, feedforward neural network and residual connection. It is mainly used to perform global attention at low resolution and learn the coherent structure of cross-channel and cross-time seismic records of rock-breaking sources, such as phase axis extension and energy consistency.

[0049] The specific process of network processing is as follows: Figure 6The noisy seismic data shown is input into a CNN encoder to extract and encode local waveforms and local continuous features of the phase axis. These encoded features are then input into a Swin-Transformer network, which uses multi-head attention and a feedforward neural network to learn and fully represent the global correlations of seismic information across channels and time. Finally, the represented features are input into a CNN decoder, which uses convolutional modules to recover the detailed information and waveform morphology of the seismic record, gradually reconstructing the data and outputting the final result. Figure 7 The image shows a clean rock-breaking source earthquake record.

[0050] In this embodiment, the processing of the clean, denoised rock-breaking source seismic record based on digital image processing technology includes: (1) The clean rock-breaking source seismic record after noise reduction is normalized by a single channel and converted into a grayscale image. The first arrival wave is represented on the grayscale image as the first abrupt change edge of the transition from low grayscale to continuous high grayscale energy band.

[0051] Specifically, regions with normalized gray values ​​lower than a preset background threshold are defined as low gray-scale regions; regions with gray values ​​higher than a preset energy threshold along the time direction or adjacent gather direction are defined as continuous high gray-scale energy bands; and locations where the absolute value of the gray-scale gradient between adjacent pixels is greater than a preset gradient threshold are defined as abrupt change locations.

[0052] (2) Using the edge detection method, detect the abrupt edges of gray-scale gradients in each channel along the time direction, select the first strong edge point obtained from the shallow to the deep scan of each channel as the location of the first arrival wave, and generate preliminary detection results.

[0053] Among them, shallow points correspond to sampling points with sampling times less than the set minimum threshold, while deep points correspond to sampling points with sampling times greater than the set maximum threshold. Strong edge points are defined as edge points with gradient magnitudes greater than the preset edge threshold and which are continuous in adjacent channels.

[0054] (3) By eliminating abnormal point interference through median filtering, a stable detection basis is established. Combined with the edge tracking algorithm, the local deviation is corrected along the extension direction of the phase axis to further optimize the picking accuracy and finally realize the automatic picking of the first arrival wave.

[0055] Anomalies are isolated points that deviate from the initial arrival time of adjacent gathers by more than a preset time difference threshold and do not satisfy the continuity of the same phase axis.

[0056] The above thresholds can all be set based on the gray-level statistical characteristics of the background area, the gray-level statistical characteristics of the effective waveform area, and the gray-level gradient statistical characteristics.

[0057] Among them, the edge tracking algorithm connects the identified edge points according to connectivity and directional consistency, while removing unreliable and isolated edge response points as noise. It is mainly used to transform scattered edge responses into continuous and clean contours / boundaries.

[0058] In this embodiment, based on a cascaded deep neural network architecture, an inversion imaging network is formed by an initial modeling network and a physical correction network. The initial arrival wave travel time information of the seismic record is used as input, and the nonlinear mapping relationship between the travel time information and the velocity field is established through the initial modeling network to output the initial wave velocity field.

[0059] The initial wave velocity field is input into the physical correction network, the theoretical travel time information and physical residuals are calculated, and the loss function of the physical correction network is constructed through the equation. When the loss converges to a stable stage, the final wave velocity field is output to form the inversion imaging result.

[0060] like Figure 8 As shown, specifically: (1) The initial modeling network consists of multiple hidden layers and nonlinear activation functions. This system is used to establish a mapping relationship between first-arrival wave travel time information and the velocity field. It takes the spatial location and time parameters of the first-arrival wave travel time information from the seismic record as input data, and utilizes the mapping between the hidden layer and the nonlinear activation function to output the initial wave velocity field, such as... Figure 9 As shown, in this process, the hidden layer is used to learn the nonlinear mapping relationship, the activation function is used for nonlinear representation, and the wave velocity loss term between the mapped wave velocity field and the seismic record wave velocity field is used as the initial modeling network loss function. {ic,bc,data} .

[0061] (2) The physical correction network introduces a physical differential operator layer, embedding the travel time and wave path function equations into the network architecture. First, the theoretical first arrival travel time of each receiver location is solved based on the initial wave velocity field. The travel time loss term is constructed based on the difference between the theoretical first arrival travel time and the first arrival travel time of the seismic record picking observation. At the same time, the predicted wave velocity field and its corresponding travel time field are substituted into the travel time and wave path function equations to construct the physical loss term. The travel time loss term and the physical loss term are weighted and summed to obtain the physical correction network loss function loss. f .

[0062] The initial modeling network loss function and the physically corrected network loss function are summed to form the total loss function of the inversion imaging network. The network parameters are then updated based on this total loss function until the total loss is less than a set threshold. .

[0063] In this embodiment, the travel time loss term between the theoretical first arrival travel time and the observed first arrival travel time is: (5); in, This is the travel time loss term corresponding to the m-th iteration; This represents the number of iterations. The number of seismic sources used in the observation system; Let be the theoretical first arrival travel time vector of the s-th source calculated in the m-th iteration; This is a transpose operation; Let be the observation first arrival travel time corresponding to the s-th earthquake source.

[0064] In this embodiment, the equations (taking the tunnel axis x and vertical z directions as examples) are: (6); in, T Let ρ be the theoretical travel time field, λ be the density, and λ be the density. μ Let be Lamé's constant.

[0065] In this embodiment, the wave path function equation (taking the X-tunnel axis and vertical direction Z as an example) is: (7); in, For density, For displacement, , Let Lamé constant be . , Here is the viscous dissipation coefficient. P To impact pressure, A For contact area, To drive torque, To resist torque.

[0066] In the inversion imaging process of rock-breaking source seismic data in this embodiment, the first arrival travel time information is automatically obtained through edge tracking to obtain the arrival time of the first arrival wave. The velocity model is mapped through a convolutional neural network. The inversion method uses the first arrival travel time calculation equation of rock-breaking source theory embedded in the PINN neural network as a loss function equation for constraint, thereby enhancing the accuracy of the inversion and achieving high-precision mapping of the velocity field.

[0067] Example 2 This embodiment provides a tunnel drilling rig rock-breaking seismic source advance prediction system, including: The denoising module is configured to perform multi-scale convolution and perceptual coding on the acquired rock-breaking source seismic data, determine the location of effective reflection information through adaptive position coding, extract the effective reflection information through a cross-attention mechanism, and obtain a clean rock-breaking source seismic record after denoising the effective reflection information. The extraction module is configured to convert clean rock-breaking source seismic records into grayscale images, detect abrupt edges of grayscale gradients in each channel of the grayscale image along the time direction, and extract the first arrival wave based on the detected edge points. The inversion module is configured to obtain the inversion imaging result of the wave velocity field based on the first arrival wave travel time information and the mapping model of the trained travel time information and velocity field. During the training process, the wave velocity field is mapped from the travel time information, and the theoretical travel time information is calculated from the wave velocity field. The mapping model is then updated through wave velocity loss, travel time loss and physical loss.

[0068] In this embodiment, the system also includes a tunnel observation device for acquiring seismic data from rock-breaking sources, solving the problems of large space requirements and long time consumption in traditional seismic wave advance prediction methods. It includes a three-component geophone 1, a wireless transmission node 4, and a signal acquisition host 8.

[0069] Among them, the three-component geophone 1 is used to collect the rock-breaking source seismic wave signal of the rock drilling rig 12; Wireless transmission node 4 is used to wirelessly transmit seismic wave signals to signal acquisition host 8; The signal acquisition host 8 is used to control the wireless transmission node 4 and to acquire seismic wave signals in real time.

[0070] like Figure 10 As shown, the three-component detector 1 is a cuboid design of 35*35*20mm. The three-component detector 1 has an 8mm threaded hole 2 in the middle, which can be connected to the expansion bolt and then fixed to the tunnel sidewall.

[0071] The three-component geophone 1 has X and Y direction markings on its surface, meaning that the geophone can receive seismic waves in the three main directions of the spatial rectangular coordinate system.

[0072] The three-component detector 1 has a circular detector interface 3 on its top, which can be connected to the wireless transmission node 4 via a connecting cable 11.

[0073] The three-component geophone 1 is fixed to the tunnel sidewall by expansion bolts or adhesive when it is installed on the tunnel sidewall, and is attached to the tunnel sidewall by a strong magnet 14 when it is installed on the drill arm 13 of the rock drilling rig 12.

[0074] like Figure 11 As shown, the wireless transmission node 4 is a cuboid design of 115*90*40mm, and is packaged in a waterproof and dustproof metal shell. Inside, it contains electronic devices 6 such as wireless transmission and automatic time synchronization. It is equipped with two 2.4GHz antennas 5 on the top and a node interface 7 at the bottom for connecting to the three-component detector 1.

[0075] Wireless transmission node 4 is connected to three-component detector 1 and wirelessly transmits the seismic wave signal received by three-component detector 1 to signal acquisition host 8 through wireless transmission node 4.

[0076] like Figure 12 As shown, the signal acquisition host 8 is equipped with acquisition software, which controls the wireless transmission node 4 to start and stop acquiring signals.

[0077] In this embodiment, as Figure 13 As shown, the signal acquisition method of the above-mentioned tunnel observation device includes: (1) Based on the tunnel site sidewall environment and the drilling arm of the rock drilling rig, determine the number of three-component geophones 1 installed in the sidewall observation system 10 and the pilot observation system 9.

[0078] Specifically, for the sidewall observation system 10, 8-10 three-component geophones 1 are installed on one sidewall of the tunnel; for the pilot observation system 9, 1-3 three-component geophones 1 are installed at the front end of the drill arm 13 of the rock drilling rig 12.

[0079] (2) Install expansion bolts at the three-component detector 1 deployment point on the tunnel sidewall and connect them to the three-component detector 1 to fix it on the tunnel sidewall; use the connecting line 11 to connect the detector interface 3 to the node interface 7, and find a position on the sidewall to fix the wireless transmission node 4.

[0080] Among them, the spacing between the three-component geophones 1 in the sidewall observation system 10 is 2m, and the distance between the three-component geophones 1 and the bottom of the sidewall is 1.5m.

[0081] (3) In the pilot observation system 9, the three-component detector 1 is fixed to the front end of the drill arm 13 of the rock drilling rig 12 by a powerful magnet 14.

[0082] In both the sidewall observation system 10 and the pilot observation system 9, the three-component detector 1 must be connected to the wireless transmission node 4, and the X, Y, and Z directions of the three-component detector 1 must be consistent.

[0083] (4) After the drilling arm of the rock drilling rig starts drilling, the signal acquisition host 8 simultaneously starts the three-component geophone 1 on the tunnel sidewall and the drilling arm of the rock drilling rig to start receiving seismic signals. When the drilling arm of the rock drilling rig finishes drilling, the three-component geophone 1 stops receiving seismic signals and saves the collected seismic data.

[0084] This embodiment addresses the challenges of existing tunnel seismic wave method advance prediction, such as large space occupation and long prediction time. It innovatively proposes to use the drilling and rock breaking vibration of a rock drilling rig as the seismic source for advance prediction. Three-component geophones are deployed on the sidewall and the front end of the drill arm to receive seismic data. By processing the seismic data, adverse geological bodies in front of the tunnel face can be predicted, ultimately realizing mechanized tunnel drilling and exploration using the drill-and-blast method.

[0085] It should be noted that the above modules correspond to the steps described in Embodiment 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0086] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0087] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0088] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0089] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0090] The method in Example 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0091] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0092] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0093] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0094] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0095] Those skilled in the art will recognize that the units and algorithm steps described in connection with the various examples of this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0096] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting the seismic source of rock breaking using a tunnel drilling rig, characterized in that, include: After performing multi-scale convolution and perceptual coding on the acquired rock-breaking source seismic data, the location of effective reflection information is determined by adaptive location coding. The effective reflection information is then extracted through a cross-attention mechanism. After denoising the effective reflection information, a clean rock-breaking source seismic record is obtained. The acquired rock-breaking source seismic data is first enhanced by feature enhancement, then nonlinearly integrated with local features through multi-scale convolution processing, and finally weighted by perceptual coding to generate position-wise weights. These weighted weights are then fused, and adaptive position coding is applied to both the outputs of the multi-scale convolution and the perceptual coding. The adaptive position coding is as follows: ;in, X This represents the input for adaptive positional encoding; G Indicates the weight of the gating mechanism. APE Indicates the adaptive position coding parameters. PE Indicates the sinusoidal position encoding parameters; The clean rock-breaking source seismic record is converted into a grayscale image. The abrupt edges of the grayscale gradient in each channel of the grayscale image are detected along the time direction, so as to extract the first arrival wave based on the detected edge points. Based on the first arrival wave travel time information, the inversion imaging result of the wave velocity field is obtained by using the mapping model of the trained travel time information and the velocity field. During the training process, the wave velocity field is obtained by mapping the travel time information, and the theoretical travel time information is calculated based on the wave velocity field. The mapping model is then updated by wave velocity loss, travel time loss and physical loss.

2. The method for predicting the seismic source of rock breaking using a tunnel drilling rig as described in claim 1, characterized in that, After determining the location of the effective reflection information, a network consisting of a multi-layer encoder and a multi-layer decoder is used to extract the effective reflection information. The multi-layer encoder is composed of self-attention and feedforward neural networks, and the multi-layer decoder is composed of self-attention, cross-attention and feedforward neural networks. The encoder output feature sequence after adaptive position encoding is used as the key matrix. K e Sum matrix V e The decoder generated by perceptual encoding is input to the initial feature sequence as the query matrix. Q d ; by query matrix Q d AND key matrix K e The relevance score is calculated by scaling the dot product, and then the attention weight matrix is ​​obtained using the Softmax function. The attention weight matrix is ​​then compared with the value matrix. V e We perform weighted summation to extract effective reflection information.

3. The method for predicting the seismic source of rock breaking using a tunnel drilling rig as described in claim 1, characterized in that, For effective reflection information, a trained CNN-Swin-Transformer network is used for denoising; the CNN-Swin-Transformer network uses a CNN network as a front-end encoder and a back-end decoder, and fuses a Swin-Transformer network in between. The noisy effective reflection information is input into the front-end encoder to extract and encode the local waveform and local continuous features of the phase axis. The encoded features are then input into the Swin-Transformer network, which learns and represents the global correlation of seismic information across channels and time through multi-head attention and feedforward neural networks. The represented features are then input into the back-end decoder, which uses convolutional layers to recover the seismic record and reconstruct the clean rock-breaking source seismic record.

4. The method for predicting the seismic source of rock breaking using a tunnel drilling rig as described in claim 1, characterized in that, Seismic records from clean rock-breaking sources are normalized using a single channel and converted into grayscale images. The first arrival wave is represented on the grayscale image as the first abrupt change edge from low grayscale to a continuous high grayscale energy band. Specifically, regions with grayscale values ​​below a preset background threshold are defined as low grayscale regions, regions with grayscale values ​​above a preset energy threshold along the time direction or adjacent gather direction are defined as continuous high grayscale energy bands, and locations where the absolute value of the grayscale gradient between adjacent pixels is greater than a preset gradient threshold are defined as abrupt change locations. Using an edge detection method, abrupt edges of grayscale gradients in each channel are detected along the time direction. The first strong edge point obtained from scanning from shallow to deep in each channel is selected as the location of the first arrival wave, generating preliminary detection results. Among them, the shallow edge corresponds to the time sampling point position where the sampling time is less than the set minimum threshold, and the deep edge corresponds to the time sampling point position where the sampling time is greater than the set maximum threshold. Strong edge points are defined as edge points with gradient amplitude greater than the preset edge threshold and continuity in adjacent channels. Median filtering is used to eliminate outlier interference, and combined with edge tracking algorithm, local deviations are corrected along the in-phase axis extension direction to achieve automatic first arrival picking. Outliers are isolated points whose deviation from the first arrival time of adjacent gathers exceeds a preset time difference threshold and do not satisfy the in-phase axis continuity.

5. The method for predicting the seismic source of rock breaking using a tunnel drilling rig as described in claim 1, characterized in that, The time loss item is: ； in, This is the travel time loss term corresponding to the m-th iteration; The number of seismic sources used in the observation system; Let be the theoretical first arrival travel time vector of the s-th source calculated in the m-th iteration; This is a transpose operation; Let be the observation first arrival travel time corresponding to the s-th earthquake source.

6. A tunnel drilling rig rock-breaking seismic source advance prediction system, characterized in that, include: The denoising module is configured to perform multi-scale convolution and perceptual coding on the acquired rock-breaking source seismic data, determine the location of effective reflection information through adaptive position coding, extract the effective reflection information through a cross-attention mechanism, and obtain a clean rock-breaking source seismic record after denoising the effective reflection information. The acquired rock-breaking source seismic data is first enhanced by feature enhancement, then nonlinearly integrated with local features through multi-scale convolution processing, and finally weighted by perceptual coding to generate position-wise weights. These weighted weights are then fused, and adaptive position coding is applied to both the outputs of the multi-scale convolution and the perceptual coding. The adaptive position coding is as follows: ;in, X This represents the input for adaptive positional encoding; G Indicates the weight of the gating mechanism. APE Indicates the adaptive position coding parameters. PE Indicates the sinusoidal position encoding parameters; The extraction module is configured to convert clean rock-breaking source seismic records into grayscale images, detect abrupt edges of grayscale gradients in each channel of the grayscale image along the time direction, and extract the first arrival wave based on the detected edge points. The inversion module is configured to obtain the inversion imaging result of the wave velocity field based on the first arrival wave travel time information and the mapping model of the trained travel time information and velocity field. During the training process, the wave velocity field is mapped from the travel time information, and the theoretical travel time information is calculated from the wave velocity field. The mapping model is then updated through wave velocity loss, travel time loss and physical loss.

7. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-5.

9. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, performs the method described in any one of claims 1-5.

Images