Tunnel surrounding rock structure geophysical exploration and intelligent analysis method

Abstract

This invention discloses a geophysical detection and intelligent analysis method for tunnel surrounding rock structures, belonging to the technical field of geophysical detection and geological structure analysis of tunnel surrounding rock. It collects multi-source geophysical data and completes standardized preprocessing and spatial registration, extracts multi-scale surrounding rock structural features and achieves cross-modal fusion, intelligently identifies geological structures through a deep learning network, completes rock mass quality classification, mechanical parameter inversion, location analysis of adverse geological bodies, and stability assessment, outputs three-dimensional visualized detection results, and performs incremental iterative optimization of the model. This invention improves the accuracy and efficiency of tunnel surrounding rock detection and analysis, reduces detection blind spots and multiple solutions, achieves accurate identification of geological structures and adverse geological bodies, enhances the model's generalization ability, and provides reliable support for advanced geological prediction in tunnel construction.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration and geological structure analysis of tunnel surrounding rock, and particularly to a method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure. Background Technology

[0002] During tunnel construction, the surrounding rock geological conditions are complex and variable. The distribution of adverse geological bodies such as stratigraphic interfaces, joints and fissures, fault fracture zones, karst caves, and water-rich zones directly affects construction safety and project quality. Traditional tunnel surrounding rock detection often employs single geophysical methods, acquiring only single-dimensional data such as wave fields or electrical properties. This results in limited data information, significant blind spots, and multiple interpretations, making it difficult to fully reflect the true geological structure and mechanical properties of the surrounding rock. Furthermore, the raw data collected on-site lacks standardized preprocessing procedures, resulting in inconsistent spatial coordinates among multiple data sources and an inability to effectively fuse characteristic information, further reducing the reliability of the detection results.

[0003] Current analysis of surrounding rock structures relies primarily on manual interpretation, which is inefficient and heavily influenced by professional experience, making it difficult to accurately identify and locate stratigraphic interfaces and adverse geological bodies. Rock mass quality grading and mechanical parameter inversion depend on limited field test data, failing to incorporate multi-dimensional geophysical characteristics for intelligent analysis, resulting in insufficient parameter acquisition accuracy. Surrounding rock stability assessments largely employ qualitative analysis methods, lacking a quantitative evaluation system and unable to classify risk levels in real time. The accuracy of attribute analysis and spatial location of adverse geological bodies is insufficient, failing to meet the needs of advanced geological forecasting and construction management.

[0004] Existing detection and analysis methods lack real-time quality control mechanisms for field data, and substandard data quality directly affects subsequent analysis results. Detection models are mostly fixed and lack incremental iterative optimization based on actual geological data revealed during excavation, resulting in weak generalization ability under different geological conditions. Detection results are mostly presented in two-dimensional charts with low three-dimensional visualization capabilities, and the output is not standardized, failing to provide intuitive and comprehensive data support for engineering design and construction. The overall detection and analysis process is fragmented, lacking a fully integrated system from data acquisition, preprocessing, feature fusion, intelligent identification to stability assessment and result output, making it difficult to meet the advanced geological detection needs of modern tunnel engineering, which require high precision, high efficiency, and intelligence. Summary of the Invention

[0005] This invention proposes a geophysical detection and intelligent analysis method for tunnel surrounding rock structures to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for geophysical detection and intelligent analysis of tunnel surrounding rock structures, comprising the following steps: The multi-source geophysical data acquisition steps for tunnel surrounding rock detection areas are used to simultaneously collect multiple types of geophysical data for the surrounding rock area to be detected, and to construct a multi-dimensional detection dataset. The standardization and spatial registration steps for multi-source geophysical data are used to standardize various types of acquired data and complete spatial coordinate registration based on a unified engineering coordinate system. The multi-scale surrounding rock structure geophysical feature extraction and cross-modal fusion steps are used to decompose the preprocessed data at multiple scales, extract multi-dimensional geophysical features, complete cross-modal deep fusion and normalization processing, and generate multi-scale fused feature maps. The intelligent identification step of surrounding rock geological structure based on deep learning network is used to input the fused feature map into the identification network and complete the pixel-level geological structure segmentation through the encoder and decoder. The intelligent inversion steps for the quality classification and physical and mechanical parameters of the surrounding rock mass in tunnels are used to combine geological structure and geophysical characteristics to complete the quality classification of the rock mass and invert the physical and mechanical parameters. The steps for precise location and attribute analysis of unfavorable geological bodies in surrounding rock are used to locate unfavorable geological bodies and analyze their characteristic parameters based on the results of fusion feature and structural identification. The intelligent assessment steps for the stability of tunnel surrounding rock structures are used to combine rock mass quality classification, physical and mechanical parameters and distribution of adverse geological bodies to complete the intelligent assessment of the stability of surrounding rock structures, classify stability levels and identify areas at risk of instability. The steps for 3D visualization of detection results and standardized output are used to 3D model and visualize various detection and analysis results, and generate standardized output reports that conform to engineering survey specifications. The incremental iterative optimization step of the surrounding rock detection and analysis model is used to collect actual geological data after excavation as a verification sample to complete the parameter fine-tuning and incremental iterative optimization of the model.

[0007] Furthermore, it also includes the deployment steps of a multi-method geophysical exploration array for tunnel surrounding rock. This is used to deploy seismic wave detection arrays, ground-penetrating radar survey lines, and high-density electrical resistivity tomography arrays in the tunnel face, tunnel sidewalls, and arch area according to the engineering geological conditions and exploration depth requirements of the tunnel. It determines the spacing between measuring points, the length of the survey lines, and the acquisition parameters of each detection method, and completes the deployment of a joint exploration array that covers the entire cross-section of the tunnel surrounding rock in both the transverse and longitudinal directions.

[0008] Furthermore, it also includes real-time quality control and dynamic supplementary testing steps for surrounding rock detection data, which are used to verify the signal-to-noise ratio, effective signal integrity, and spatial coverage of the acquired data in real time during the data acquisition process, compare it with the preset data quality threshold, identify areas where the data quality does not meet the standards, and automatically generate supplementary testing plans.

[0009] Furthermore, during the multi-source geophysical data acquisition process in the tunnel surrounding rock detection area, seismic wave reflection data acquisition was completed using a multi-channel high-resolution seismograph to collect transverse and longitudinal wave reflection signals of the tunnel surrounding rock. Ground-penetrating radar data acquisition was completed using a combination of antennas with different center frequencies, switching the corresponding frequency antenna for different detection depths. High-density electrical resistivity tomography data acquisition was completed using a Winner device, covering the acquisition of electrical information at different depths of the tunnel surrounding rock. Surrounding rock acoustic wave test data was completed using a combination of single-hole and cross-hole acoustic wave tests to obtain acoustic wave velocity data at different depths of the surrounding rock.

[0010] Furthermore, in the process of standardization preprocessing and spatial registration of multi-source geophysical data, direct wave removal, surface wave suppression, deconvolution, and static correction are performed for seismic wave data; zero-point correction, background denoising, gain restoration, and frequency band filtering are performed for ground-penetrating radar data; distortion data removal, terrain correction, and apparent resistivity calculation are performed for high-density electrical resistivity data; and first wave acquisition, sound velocity calculation, and outlier removal are performed for acoustic wave test data. All preprocessed data are mapped to the unified coordinate system of the tunnel project.

[0011] Furthermore, in the process of multi-scale geophysical feature extraction and cross-modal fusion of surrounding rock structures, a cross-modal feature fusion network is used to achieve deep fusion of different types of features. The expression for calculating the weight of multi-source feature fusion is as follows: , These are the feature fusion weight coefficients at the corresponding scale. This represents the total number of geophysical feature types. Number the feature type. The contribution weight of the i-th type of geophysical feature. The result of calculating the inner product between the i-th type of geophysical feature vector and the standard surrounding rock feature reference vector is given. Let i be the normalized geophysical feature vector of the i-th class. This is a standardized reference vector for surrounding rock characteristics. This is the 2-norm operator for vectors.

[0012] Furthermore, in the intelligent identification of surrounding rock geological structures based on deep learning networks, a hybrid architecture combining U-Net and Transformer is adopted as the backbone of the surrounding rock geological structure identification network. The global context information of surrounding rock features is extracted through the multi-head self-attention mechanism of the Transformer encoder to capture the long-distance spatial dependencies between different geological structures. Through the multi-level upsampling structure of the U-Net decoder, multi-scale feature fusion is completed by combining feature maps of different levels of the encoder, and pixel-level geological structure segmentation results of tunnel surrounding rock are output. Instance division of different geological structures is completed through connected component analysis and spatial clustering. At the same time, low-confidence identification areas are screened through feature similarity calculation, and local feature re-extraction and secondary identification are performed.

[0013] Furthermore, in the process of intelligent inversion of the quality classification and physical and mechanical parameters of the tunnel surrounding rock mass, a quality classification and evaluation system for the surrounding rock mass is constructed by combining the geological structure identification results, geophysical characteristic parameters, and indoor rock mechanics test data. The comprehensive score calculation expression for the quality of the surrounding rock mass is as follows: , A comprehensive score for the quality of the surrounding rock mass. The weight of the rock mass integrity coefficient. This represents the measured value of the integrity coefficient of the surrounding rock mass. This represents the maximum value of the rock mass integrity coefficient. The weight of the uniaxial compressive strength of the rock. This is the inversion value of the uniaxial compressive strength of the rock. This represents the standard maximum value of the uniaxial compressive strength of rock. Weighted by the volumetric density of joints and fractures. This represents the measured value of the bulk density of the joints and fissures in the surrounding rock. This represents the standard maximum value of the bulk density of joints and fissures. The weight of the water-rich state of the surrounding rock. This represents the quantitative value of the water-rich state of the surrounding rock. The maximum value is quantified for the water-rich state of the surrounding rock.

[0014] Furthermore, in the process of precise location and attribute analysis of unfavorable geological bodies in the surrounding rock, a geophysical feature sample library of unfavorable geological bodies is constructed based on the multi-scale fusion feature map and geological structure identification results. The type identification of unfavorable geological bodies is completed through a deep learning classification network. Combined with the forward and inverse results of multi-source geophysical data, the spatial boundary, burial depth, and scale of unfavorable geological bodies are determined. The filling medium type, water-rich state, and degree of fragmentation attribute parameters of unfavorable geological bodies are analyzed through multi-physics field joint inversion. At the same time, the spatial coordinate location of unfavorable geological bodies is completed based on the tunnel engineering coordinate system.

[0015] Furthermore, during the incremental iterative optimization of the surrounding rock detection and analysis model, the actual geological structure revealed after tunnel excavation, rock mass physical and mechanical test data, and surrounding rock stability field monitoring data are compiled into validation samples and incremental training samples. These are divided into training set, validation set, and test set in an 8:1:1 ratio. Using a combination of transfer learning and incremental training, the parameters of the geological structure identification network, parameter inversion model, and stability assessment model are fine-tuned. The pre-training parameters of the network backbone feature extraction layer are frozen. The model accuracy is verified through the validation set, and the model generalization ability is tested through the test set. The iteratively optimized model replaces the original model and is used for subsequent geophysical detection and intelligent analysis of the surrounding rock of the tunnel.

[0016] Compared with existing technologies, the beneficial effects of this invention are: This invention constructs a comprehensive detection system covering seismic waves, ground-penetrating radar, high-density electrical resistivity tomography, and acoustic wave testing through the joint acquisition of multi-source geophysical data. This expands the detection range of surrounding rocks, reduces blind spots and multiple solutions caused by single methods, and improves the richness and completeness of the original detection data, providing a comprehensive data foundation for subsequent analysis.

[0017] This invention eliminates noise, distortion, and spatial bias in various types of data through standardized preprocessing of multi-source data and unified spatial registration, achieving accurate alignment of data from different methods and measurement points, improving data quality and spatial consistency, and providing standardized data support for multi-scale feature extraction and cross-modal fusion.

[0018] This invention enhances the feature representation of surrounding rock geological structures and anomalous areas through multi-scale feature extraction and cross-modal deep fusion, improves the distinguishability of stratigraphic interfaces and adverse geological bodies at different scales, makes geological structure identification more accurate, reduces interpretation errors, and improves the overall reliability of analysis.

[0019] This invention employs a deep learning hybrid architecture to achieve intelligent identification of surrounding rock geological structures. It can automatically extract global contextual features and long-distance spatial dependencies, complete pixel-level geological structure segmentation, improve identification efficiency and accuracy, reduce manual intervention and subjective bias, and adapt to the needs of automated interpretation under complex geological conditions.

[0020] This invention achieves quantitative grading of surrounding rock mass quality through multi-index weighted fusion, and combines intelligent regression model to invert mechanical parameters, thereby improving the comprehensiveness and accuracy of parameter acquisition, providing a reliable basis for surrounding rock stability assessment, and making the evaluation results more in line with engineering practice.

[0021] This invention enables precise location and multi-attribute analysis of adverse geological bodies, clearly defining their spatial location, scale, filling status, and water content. This provides accurate data support for proactive treatment, reduces construction risks, and enhances the safety of tunnel engineering construction.

[0022] This invention continuously ensures the quality of the original data through real-time on-site quality control and dynamic supplementary testing. It also continuously improves the model's generalization ability and analytical accuracy under different geological conditions through incremental iterative optimization, enabling the system to have continuous learning and adaptive upgrade capabilities.

[0023] This invention uses three-dimensional visualization and standardized output to intuitively present the surrounding rock structure, parameter distribution, adverse geological bodies and stability levels, and outputs detection reports that conform to engineering specifications, thereby improving the readability and practicality of the results and providing unified and reliable data support for design, construction and supervision. Attached Figure Description

[0024] Figure 1 This is a schematic block diagram of the overall flowchart for tunnel surrounding rock structure detection and intelligent analysis proposed in this invention; Figure 2 This is a schematic block diagram of the multi-method joint detection array deployment and real-time quality control flowchart proposed in this invention; Figure 3 This is a flowchart of the multi-scale feature extraction and cross-modal fusion technology of the present invention; Figure 4 This is a flowchart of the intelligent geological structure recognition process based on a hybrid architecture according to the present invention; Figure 5 This diagram illustrates the rock mass quality classification, stability assessment, and model optimization of this invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Reference Figures 1 to 5 A method for geophysical exploration and intelligent analysis of tunnel surrounding rock structures, comprising the following steps: The multi-source geophysical data acquisition steps for tunnel surrounding rock detection area are used to simultaneously acquire seismic wave reflection data, ground-penetrating radar electromagnetic wave data, high-density electrical resistivity data, and surrounding rock acoustic wave test data for the surrounding rock area to be detected in the tunnel, and to construct a multi-dimensional geophysical detection dataset for tunnel surrounding rock. The multi-source geophysical data standardization preprocessing and spatial registration steps are used to perform standardization processing such as denoising, gain compensation, frequency band filtering, and outlier removal on various types of collected geophysical data. Based on the tunnel unified engineering coordinate system, the spatial coordinate registration of multi-source data is completed to generate a standardized surrounding rock detection dataset with unified spatial location. The multi-scale geophysical feature extraction and cross-modal fusion steps of the surrounding rock structure are used to decompose the preprocessed multi-source data at multiple scales, extract wave field features, electrical features, and acoustic features at different depths of the surrounding rock, complete the cross-modal deep fusion and normalization of multi-dimensional features, and generate a multi-scale fused feature map of the tunnel surrounding rock. The intelligent identification step of the surrounding rock geological structure based on deep learning network is used to input multi-scale fused feature map into the surrounding rock geological structure identification network, extract global context features through encoder, and complete pixel-level geological structure segmentation through decoder to identify the geological structure of the surrounding rock of tunnel, such as stratigraphic interfaces, joint and fracture development zones, and rock strata contact zones. The intelligent inversion step of the surrounding rock mass quality classification and physical and mechanical parameters is used to combine the identified geological structure and geophysical characteristics of the surrounding rock to complete the quality classification of the surrounding rock mass and invert the integrity coefficient, elastic modulus, Poisson's ratio, compressive strength and other physical and mechanical parameters of the surrounding rock. The steps for precise location and attribute analysis of unfavorable geological bodies in the surrounding rock are used to locate fault fracture zones, karst caves, water-rich zones, and weak interlayers in the tunnel surrounding rock based on the results of fusion features and geological structure identification, and to analyze the spatial location, size, filling attributes, and water-rich state characteristic parameters of the unfavorable geological bodies. The intelligent assessment steps for the stability of tunnel surrounding rock structures are used to combine the quality classification of surrounding rock mass, physical and mechanical parameters, and distribution characteristics of adverse geological bodies to complete the intelligent assessment of the stability of tunnel surrounding rock structures, classify the stability level of surrounding rock, and identify areas at risk of instability of surrounding rock. The three-dimensional visualization and standardized output steps of the detection results are used to perform three-dimensional modeling and visualization of the surrounding rock geological structure, rock mass parameters, adverse geological bodies, and stability assessment results, and generate a standardized detection result report that conforms to the tunnel engineering investigation specifications. The incremental iterative optimization step of the surrounding rock detection and analysis model is used to collect actual geological data revealed after tunnel excavation as verification samples to complete the geological structure identification, parameter inversion, stability assessment model parameter fine-tuning and incremental iterative optimization.

[0027] This invention also includes a step for deploying a multi-method geophysical exploration array for tunnel surrounding rock. This step involves deploying seismic wave detection arrays, ground-penetrating radar survey lines, and high-density electrical resistivity tomography (EDT) electrode arrays at the tunnel face, tunnel sidewalls, and arch area, based on the tunnel's engineering geological conditions and exploration depth requirements. It also involves determining the spacing between measurement points, the length of the survey lines, and the acquisition parameters for each detection method. This achieves a multi-method geophysical exploration array deployment that provides full lateral and longitudinal coverage of the tunnel surrounding rock, eliminating blind spots and multiple solutions inherent in single detection methods.

[0028] This invention also includes real-time quality control and dynamic supplementary measurement steps for surrounding rock detection data. These steps are used to verify the signal-to-noise ratio, effective signal integrity, and spatial coverage of the acquired data in real time during the data acquisition process, compare the data with preset data quality thresholds, identify areas where the data quality does not meet the standards, automatically generate supplementary measurement plans, adjust the acquisition parameters to complete the dynamic supplementary measurement of the corresponding areas, and ensure that the original detection data acquired on-site meets the quality requirements for subsequent analysis and processing.

[0029] In this invention, during the multi-source geophysical data acquisition process in the tunnel surrounding rock detection area, seismic wave reflection data acquisition is completed using a multi-channel high-resolution seismograph to collect transverse and longitudinal wave reflection signals of the tunnel surrounding rock. Ground-penetrating radar data acquisition is completed using a combination of antennas with different center frequencies, switching the corresponding frequency antenna for different detection depths. High-density electrical resistivity tomography data acquisition is completed using a Winner device to collect electrical information covering different depths of the tunnel surrounding rock. Surrounding rock acoustic wave test data is completed using a combination of single-hole and cross-hole acoustic wave tests to obtain acoustic wave velocity data at different depths of the surrounding rock. Through multi-method data acquisition, full-depth coverage detection of the tunnel surrounding rock from shallow to deep is achieved.

[0030] In this invention, during the standardization preprocessing and spatial registration of multi-source geophysical data, direct wave removal, surface wave suppression, deconvolution, and static correction are performed on seismic wave data; zero-point correction, background denoising, gain restoration, and frequency band filtering are performed on ground-penetrating radar data; distortion data removal, terrain correction, and apparent resistivity calculation are performed on high-density electrical resistivity data; and first wave acquisition, sound velocity calculation, and outlier removal are performed on acoustic wave test data. All preprocessed data are mapped to a unified coordinate system for tunnel engineering, completing the spatial stitching and coordinate alignment of data from different survey lines and points, eliminating spatial positional deviations of multi-source data, and ensuring the consistency of data spatial position.

[0031] In this invention, during the extraction and cross-modal fusion of geophysical features of surrounding rock structures at multiple scales, wavelet multi-resolution analysis is used to complete the multi-scale decomposition of multi-source data. Wavefield features, electrical features, and acoustic features of the surrounding rock are extracted at different decomposition scales. Deep fusion of different types of features is achieved through a cross-modal feature fusion network. The multi-source feature fusion weight calculation expression is as follows: ,and The values ​​range from 0 to 1 and are dimensionless. These are the feature fusion weight coefficients for the corresponding scale, ranging from 0 to 1, and are dimensionless. This represents the total number of geophysical feature types. Number the feature type. The contribution weight of the i-th type of geophysical feature is dimensionless. The result of the inner product calculation of the i-th type geophysical feature vector and the standard surrounding rock feature reference vector is dimensionless. Let i be the normalized geophysical feature vector of the i-th class, which is dimensionless. This is a standardized reference vector for surrounding rock characteristics, dimensionless. It is a 2-norm operator for vectors. By calculating the correlation between different types of geophysical features and standard surrounding rock features, it dynamically allocates the fusion weight of each feature, strengthens the feature expression ability of surrounding rock anomaly areas and geological structures, improves the feature differentiation of stratigraphic interfaces and adverse geological bodies at different scales, and provides highly recognizable fusion feature support for subsequent geological structure identification and parameter inversion.

[0032] In this invention, during the intelligent identification of surrounding rock geological structures based on deep learning networks, a hybrid architecture combining U-Net and Transformer is used as the backbone of the surrounding rock geological structure identification network. The global context information of surrounding rock features is extracted through the multi-head self-attention mechanism of the Transformer encoder, capturing the long-distance spatial dependencies between different geological structures. Through the multi-level upsampling structure of the U-Net decoder, multi-scale feature fusion is completed by combining feature maps of different levels of the encoder, outputting pixel-level geological structure segmentation results of tunnel surrounding rock. Instance division of different geological structures is completed through connected component analysis and spatial clustering, accurately identifying the spatial distribution and morphological characteristics of stratigraphic interfaces, joint and fracture development zones, and rock strata contact zones of tunnel surrounding rock. At the same time, low-confidence identification areas are screened through feature similarity calculation, and local feature re-extraction and secondary identification are performed to improve the accuracy of geological structure identification.

[0033] In this invention, during the intelligent inversion of the quality classification and physical and mechanical parameters of the surrounding rock mass in tunnels, a quality classification and evaluation system for the surrounding rock mass is constructed by combining the geological structure identification results, geophysical characteristic parameters, and indoor rock mechanics test data. The comprehensive score calculation expression for the quality of the surrounding rock mass is as follows: ,and , , , The values ​​range from 0 to 1 and are dimensionless. The overall quality score of the surrounding rock mass is given, with a value ranging from 0 to 100, and is dimensionless. The weight of the rock mass integrity coefficient is dimensionless. The measured value of the integrity coefficient of the surrounding rock mass is dimensionless. This represents the maximum value of the rock mass integrity coefficient, which is 1 and dimensionless. The uniaxial compressive strength of rock is a dimensionless weight. This is the inversion value of the uniaxial compressive strength of the rock, in MPa. This represents the standard maximum value of the uniaxial compressive strength of rock, in MPa. The unit of measurement is the weight of the volumetric density of joints and fractures, which is dimensionless. The measured value of the bulk density of joints and fissures in the surrounding rock is given in units of one joint per cubic meter. This represents the maximum standard value of the bulk density of joints and fissures, expressed in units of fissures per cubic meter. The weights for the water-rich state of the surrounding rock are dimensionless. This is a dimensionless quantitative value representing the water-rich state of the surrounding rock. The maximum value for the water-rich state of the surrounding rock is set to 1, which is dimensionless. By weighted integration of multiple indicators such as rock mass integrity, rock strength, joint development degree, and water-rich state, the standardized quantitative classification of the surrounding rock mass quality is achieved. At the same time, based on the mapping relationship between geophysical characteristics and rock mass physical and mechanical parameters, the intelligent inversion of the surrounding rock elastic modulus, Poisson's ratio, and integrity coefficient physical and mechanical parameters is completed through machine learning regression model, providing accurate parameter support for the stability assessment of the surrounding rock.

[0034] In this invention, during the precise location and attribute analysis of unfavorable geological bodies in the surrounding rock, a geophysical feature sample library of unfavorable geological bodies is constructed based on multi-scale fused feature maps and geological structure identification results. The type identification of unfavorable geological bodies is completed through a deep learning classification network. Combined with the forward and inverse results of multi-source geophysical data, the spatial boundary, burial depth, and scale of unfavorable geological bodies are determined. The filling medium type, water-rich state, and degree of fragmentation attribute parameters of unfavorable geological bodies are analyzed through multi-physics field joint inversion. At the same time, the spatial coordinate location of unfavorable geological bodies is completed based on the tunnel engineering coordinate system, generating a spatial distribution model and attribute parameter report of unfavorable geological bodies, providing data support for the advanced treatment of unfavorable geological bodies during tunnel construction.

[0035] In this invention, during the incremental iterative optimization of the surrounding rock detection and analysis model, the actual geological structure revealed after tunnel excavation, rock mass physical and mechanical test data, and on-site monitoring data of surrounding rock stability are organized into validation samples and incremental training samples. These are divided into training set, validation set, and test set in an 8:1:1 ratio. A transfer learning combined with incremental training method is used to fine-tune the parameters of the geological structure identification network, parameter inversion model, and stability assessment model. The pre-training parameters of the network backbone feature extraction layer are frozen, and only the parameters of the high-level semantic recognition layer and regression layer are iteratively updated. The model accuracy is verified through the validation set, and the model generalization ability is tested through the test set. The iteratively optimized model replaces the original model and is used for subsequent geophysical detection and intelligent analysis of tunnel surrounding rock, continuously improving the model's adaptability and analysis accuracy for different geological conditions and different types of tunnel projects.

[0036] The following two examples further illustrate the specific implementation of this system: This section, combined with the practical application scenario of advanced geological prediction in tunnel engineering, provides a complete implementation description of the geophysical detection and intelligent analysis method for tunnel surrounding rock structure. The described embodiments fully cover the entire process steps and technical solutions of the method. Other implementation methods obtained by those skilled in the art based on this embodiment are all within the protection scope of this invention.

[0037] Example 1: Implementation of Detection and Intelligent Analysis of Surrounding Rock in Limestone Strata for Deeply Buried Mountain Highway Tunnels This embodiment is applied to a deep-buried tunnel project of a secondary highway in a mountainous and hilly area. The tunnel is a single-bore two-way traffic tunnel with a net excavation width of 12m, a net height of 10m, and a maximum burial depth of 380m. The target of this implementation is the surrounding rock area to be explored 100m in front of the tunnel face at K3+240. The strata in the area are mainly medium-thick limestone. Geological survey data shows that karst is developed in the area, and there are fault fracture zones, hidden karst caves, and water-rich zones. The tunnel construction faces engineering risks such as water and mud inrush and surrounding rock instability. By fully implementing the entire process of this method, high-precision detection and intelligent analysis of the tunnel surrounding rock structure can be achieved.

[0038] This embodiment first executes the deployment steps of a multi-method combined geophysical exploration array for the tunnel surrounding rock. Based on the geological conditions of the tunnel's limestone strata and the requirement of a 100m exploration depth, a 24-channel seismic wave reflection detection array is deployed at the tunnel face, with a channel spacing of 1m and an offset distance of 5m, covering the entire cross-section of the tunnel face. Three ground-penetrating radar lines are deployed at the tunnel face and on each of the left and right tunnel sidewalls, with a line length of 100m and a measurement point spacing of 0.2m, using a combination of 100MHz and 200MHz center frequency antennas to balance exploration depth and resolution. Two high-density electrical resistivity tomography lines are deployed on the tunnel floor, with a line length of 120m and an electrode spacing of 2m, using a Wenner device to deploy the electrode array. Six acoustic wave test holes are deployed on the tunnel sidewalls, with a hole depth of 30m, using a combination of single-hole and cross-hole methods to complete the acoustic wave test array deployment, achieving full cross-sectional coverage of the surrounding rock area to be explored in both the lateral and longitudinal directions, eliminating the blind spots of a single detection method.

[0039] Subsequently, multi-source geophysical data acquisition steps were performed for the tunnel surrounding rock detection area. A multi-channel high-resolution seismograph was used to acquire shear and longitudinal wave reflection signals from the surrounding rock, with a sampling interval of 0.25 ms and a recording length of 1024 ms. Ground-penetrating radar was used to acquire data for each survey line, with 512 sampling points and a time window of 800 ns. A high-density electrical resistivity meter was used to acquire resistivity data, with a power supply time of 400 ms and a sampling period of 100 ms. An acoustic wave meter was used to acquire acoustic wave test data for the surrounding rock, with a sampling interval of 0.1 μs, obtaining acoustic wave velocity data at different depths of the surrounding rock. The simultaneous acquisition and storage of these four types of data were completed, constructing a multi-dimensional geophysical detection dataset for the tunnel surrounding rock.

[0040] During data acquisition, real-time quality control and dynamic supplementary measurement steps for surrounding rock detection data are performed simultaneously. The signal-to-noise ratio, effective signal integrity, and spatial coverage of the acquired data are verified in real time. The data is compared with preset acquisition data quality thresholds. For gathers with a seismic wave data signal-to-noise ratio of less than 30dB, sections with incomplete effective signals in ground-penetrating radar data, and measurement points with excessive distortion data in high-density electrical resistivity tomography, supplementary measurement plans are automatically generated, and acquisition parameters are adjusted to complete dynamic supplementary measurements in the corresponding areas, ensuring that the original detection data meets the quality requirements for subsequent analysis and processing.

[0041] Next, multi-source geophysical data standardization preprocessing and spatial registration steps are performed. For seismic wave data, direct wave removal, surface wave suppression, deconvolution, and static correction are performed to improve the resolution of effective reflected signals. For ground-penetrating radar data, zero-point correction, background denoising, gain restoration, and band filtering are performed to eliminate random noise and background interference. For high-density electrical resistivity tomography (EPT) data, distortion data removal, terrain correction, and apparent resistivity calculation are performed to eliminate the influence of terrain undulations on electrical data. For acoustic wave test data, first wave acquisition, sound velocity calculation, and outlier removal are performed to obtain acoustic parameters related to surrounding rock integrity. All preprocessed data are mapped to the tunnel engineering independent coordinate system, and spatial stitching and coordinate alignment of data from different survey lines and points are completed to eliminate spatial positional deviations of multi-source data and generate a standardized surrounding rock detection dataset with unified spatial position.

[0042] Subsequently, the multi-scale geophysical feature extraction and cross-modal fusion steps of the surrounding rock structure were performed. Wavelet multi-resolution analysis was used to complete the 5-scale decomposition of multi-source data. Wave field propagation characteristics, resistivity electrical characteristics, and acoustic wave velocity characteristics of the surrounding rock were extracted at different decomposition scales. The deep fusion of different types of features was completed through a cross-modal feature fusion network. The fusion weight of each feature was dynamically allocated to enhance the feature expression ability of the surrounding rock anomaly area and geological structure, and generate a multi-scale fusion feature map of the tunnel surrounding rock.

[0043] Next, a deep learning network-based intelligent identification step for surrounding rock geological structures is executed. The multi-scale fused feature map is input into the pre-trained surrounding rock geological structure identification network. The network adopts a hybrid architecture combining U-Net and Transformer. The Transformer encoder extracts global context features to capture long-distance spatial dependencies between different geological structures. The U-Net decoder completes pixel-level geological structure segmentation and outputs the spatial distribution and morphological features of surrounding rock stratigraphic interfaces, joint and fracture development zones, and rock strata contact zones. The geological structure instance division is completed through connected component analysis. At the same time, low-confidence identification areas are screened, and local feature re-extraction and secondary identification are performed to improve the accuracy of geological structure identification.

[0044] Subsequently, the process of classifying the quality of the surrounding rock mass and intelligently inverting its physical and mechanical parameters was carried out. Combining the geological structure identification results, geophysical characteristic parameters, and indoor rock mechanics test data, a quality classification and evaluation system for the surrounding rock mass was constructed, and the standardized and quantitative classification of the surrounding rock mass quality was completed. At the same time, based on the mapping relationship between geophysical characteristics and rock mass physical and mechanical parameters, the physical and mechanical parameters of the surrounding rock, such as integrity coefficient, elastic modulus, Poisson's ratio, and uniaxial compressive strength, were obtained through machine learning regression model.

[0045] Next, the steps of precise location and attribute analysis of unfavorable geological bodies in the surrounding rock were carried out. Based on the multi-scale fusion feature map and geological structure identification results, the type of unfavorable geological body was identified through a deep learning classification network. Combined with the forward and inverse results of multi-source data, the spatial boundaries, burial depth and scale of two karst caves and one fault fracture zone in the area to be explored were determined. The filling medium type, water-rich state and degree of fracture of the unfavorable geological body were analyzed through multi-physics field joint inversion. The spatial coordinate location of the unfavorable geological body was completed based on the tunnel engineering coordinate system, and a spatial distribution model and attribute parameter report of the unfavorable geological body were generated.

[0046] Next, the intelligent assessment of the stability of the tunnel surrounding rock structure is carried out. Combining the results of the surrounding rock mass quality classification, physical and mechanical parameters, and distribution characteristics of adverse geological bodies, the intelligent assessment of the stability of the tunnel surrounding rock structure is completed, the stability level of the surrounding rock is classified, and the fault fracture zone and water-rich karst cave area are identified as high instability risk areas, and the spatial range of the risk area is clarified.

[0047] The subsequent steps involve 3D visualization and standardized output of the detection results. The geological structure of the surrounding rock, the physical and mechanical parameters of the rock mass, the distribution of adverse geological bodies, and the stability assessment results are modeled and visualized in 3D to generate a standardized advanced geological prediction report that conforms to the specifications for highway tunnel engineering investigation. This provides data support for the adjustment of tunnel construction plans and the design of advanced support.

[0048] Finally, the incremental iterative optimization step of the surrounding rock detection and analysis model is executed. After the tunnel is excavated to K3+340, the actual geological structure, rock physical and mechanical test data, and field monitoring data of surrounding rock stability are collected and organized into verification samples and incremental training samples. The parameters of the geological structure identification network, parameter inversion model, and stability assessment model are fine-tuned by using transfer learning combined with incremental training. The iteratively optimized model replaces the original model and is used for surrounding rock detection and analysis of subsequent tunnel sections.

[0049] Table 1 Comparison of the identification effects of different detection methods on adverse geological bodies Table 1 compares the core performance indicators of single geophysical exploration methods and the multi-source combined method of this invention. Single exploration methods, limited by their inherent principles, suffer from low accuracy in identifying stratigraphic interfaces, large errors in locating adverse geological bodies, and a high proportion of blind spots. Furthermore, manual data interpretation is time-consuming, failing to meet the needs of rapid forecasting for tunnel construction. This invention, through joint acquisition of multi-source geophysical data and cross-modal feature fusion, significantly improves the accuracy of stratigraphic interface identification, reduces the location error of adverse geological bodies, and effectively reduces blind spots. Simultaneously, the intelligent identification network greatly shortens data interpretation time, achieving high-precision and high-efficiency detection and analysis of tunnel surrounding rock structures, better adapting to the advanced geological forecasting needs under complex geological conditions in mountain tunnels.

[0050] Example 2: Implementation of Intelligent Analysis of Soft Rock and Water-Rich Strata Surrounding Rock in Shallow-Buried Tunnels of Urban Subways This embodiment is applied to a subway tunnel project in a certain city. The tunnel is constructed using the shield tunneling method, with an excavation diameter of 6.4m and a tunnel depth of 12m-18m. The target of this implementation is the surrounding rock area to be explored within a 50m range of the tunnel sidewall and arch of the ZDK12+180ZDK12+230 section. The strata in this area are mainly composed of silty clay, fine sand, and silty soil, which are typical soft rock water-rich strata. There are water-rich weak interlayers, liquefiable layers of fine sand, and confined aquifers, which are unfavorable geological bodies. Tunnel construction faces engineering risks such as water inrush, face instability, and surface subsidence. By fully implementing the entire process of this method, the surrounding rock structure of the tunnel in the soft rock water-rich strata can be accurately detected and intelligently analyzed.

[0051] This embodiment first executes the deployment steps of a multi-method combined geophysical exploration array for the tunnel surrounding rock. Based on the geological conditions of soft rock and water-rich strata in urban subways and the requirement of a 50m exploration depth, one ground-penetrating radar line is deployed at the tunnel arch, left and right arch waists, and left and right tunnel sidewalls. The line length is 50m, and the measurement point spacing is 0.1m. A combination of 400MHz and 100MHz center frequency antennas is used to take into account both fine exploration of shallow surrounding rock and exploration of deep strata. Three high-density electrical resistivity tomography lines are deployed on the tunnel floor, with a line length of 60m and an electrode spacing of 1m. The electrode array is deployed using a Winner-Schlumberger device. A seismic surface wave detection array is deployed on the tunnel sidewalls, with a channel spacing of 0.5m, covering the entire tunnel cross section. Four acoustic test holes are deployed at the arch, with a hole depth of 20m. Single-hole acoustic testing is used to complete the array deployment, achieving joint exploration coverage of the entire cross section and all depths of the surrounding rock area to be explored.

[0052] Subsequently, multi-source geophysical data acquisition steps were performed for the tunnel surrounding rock exploration area. Ground-penetrating radar equipment was used to complete data acquisition for each survey line, with 1024 sampling points and a time window of 400 ns. A high-density electrical resistivity meter was used to complete resistivity data acquisition, with a power supply time of 200 ms and a sampling period of 50 ms. A multi-channel seismograph was used to complete surface wave data acquisition, with a sampling interval of 0.5 ms and a recording length of 512 ms. An acoustic wave tester was used to complete acoustic wave test data acquisition for the surrounding rock, with a sampling interval of 0.2 μs, to obtain acoustic wave velocity data of soft rock strata. Simultaneously, multi-source data acquisition and storage were completed to construct a multi-dimensional geophysical exploration dataset for soft rock strata.

[0053] During the data acquisition process, real-time quality control and dynamic supplementary measurement steps for surrounding rock detection data are performed simultaneously. The signal-to-noise ratio and effective signal integrity of the acquired data are verified in real time. For the ground radar data noise exceeding the standard range caused by electromagnetic interference in the urban environment and the seismic wave data signal-to-noise ratio insufficient gather caused by ground vibration, a supplementary measurement plan is automatically generated. The acquisition parameters are adjusted during low-interference periods to complete the dynamic supplementary measurement, ensuring that the quality of the original data meets the analysis requirements.

[0054] Next, the standardization and spatial registration steps for multi-source geophysical data are performed. For ground-penetrating radar data, zero-point correction, background denoising, and bandpass filtering are completed to eliminate electromagnetic interference and random noise from the urban environment. For high-density electrical resistivity data, distortion data removal, terrain correction, and apparent resistivity calculation are performed to eliminate the influence of surface structures on electrical data. For seismic surface wave data, surface wave extraction, dispersion analysis, and denoising are performed. For acoustic test data, first wave picking, sound velocity calculation, and outlier removal are performed. All preprocessed data are mapped to the unified coordinate system of the subway project, completing the spatial coordinate registration and stitching of multi-source data to generate a standardized surrounding rock detection dataset with unified spatial location.

[0055] Then, the multi-scale geophysical feature extraction and cross-modal fusion steps of the surrounding rock structure are performed. Wavelet multi-resolution analysis is used to complete the multi-scale decomposition of multi-source data. Surface wave dispersion features, resistivity and electrical properties, and acoustic wave velocity features of soft rock strata are extracted at different decomposition scales. The deep fusion of multi-dimensional features is completed through a cross-modal feature fusion network. The fusion weight of each feature is dynamically allocated to enhance the feature expression of weak interlayers and water-rich zones in soft rock strata and generate a multi-scale fused feature map of soft rock surrounding rock.

[0056] Next, a deep learning network-based intelligent identification step for surrounding rock geological structures is executed. Multi-scale fused feature maps are input into a pre-trained soft rock strata geological structure identification network. Global context features are extracted through a Transformer encoder to capture the spatial dependencies between different structures in soft rock strata. Pixel-level geological structure segmentation is completed through a decoder to identify stratigraphic interfaces, bedding structures, and joint and fracture development zones in soft rock strata. Geological structure instance division is completed through connected component analysis, and secondary identification is performed on low-confidence areas to improve the accuracy of soft rock strata geological structure identification.

[0057] Subsequently, the steps of classifying the quality of the surrounding rock mass and intelligently inverting its physical and mechanical parameters were carried out. Combining the geological structure identification results, geophysical characteristics, and indoor soft rock mechanical test data, the standardized quantitative classification of the quality of the soft rock surrounding rock mass was completed. The integrity coefficient, elastic modulus, Poisson's ratio, cohesion, and internal friction angle of the soft rock surrounding rock were obtained through machine learning regression model inversion.

[0058] Next, the steps of precise location and attribute analysis of unfavorable geological bodies in the surrounding rock are carried out. Based on the fusion of feature maps and geological structure identification results, the type of unfavorable geological body is identified, and the spatial location and size of three water-rich soft interlayers and one silty fine sand liquefaction layer in the area to be explored are determined. The water-rich state, soil density and mechanical strength attribute parameters of the unfavorable geological body are analyzed through multi-physics field joint inversion, and the spatial coordinate location of the unfavorable geological body is completed, generating a spatial distribution model and attribute report.

[0059] Next, the intelligent assessment of the stability of the tunnel surrounding rock structure is carried out. Combining the quality classification of the surrounding rock mass, physical and mechanical parameters, and distribution characteristics of adverse geological bodies, the intelligent assessment of the stability of the soft rock surrounding rock structure is completed, the stability level of the surrounding rock is classified, the water-rich soft interlayer area is identified as a high instability risk area, and the scope of surface subsidence risk is clarified.

[0060] The subsequent steps involve 3D visualization and standardized output of the detection results. The geological structure, mechanical parameters, adverse geological bodies, and stability assessment results of the surrounding rock are modeled and visualized in 3D to generate a standardized advanced geological forecast report that conforms to the urban rail transit engineering survey specifications. This provides data support for adjusting shield tunneling construction parameters and designing advanced stratum reinforcement schemes.

[0061] Finally, the incremental iterative optimization step of the surrounding rock detection and analysis model is executed. After the shield tunnel section is completed, the geological radar scanning data after the surrounding rock behind the segments is grouted and reinforced, the geological data of the surrounding rock exposed in the tunnel, and the on-site monitoring data are collected and organized into incremental training samples to complete the fine-tuning and iterative optimization of the model parameters, thereby improving the model's adaptability and analysis accuracy to urban soft rock and water-rich strata.

[0062] Table 2 Comparison of the analytical effects of the method of the present invention and traditional manual interpretation methods Table 2 compares the core analytical performance indicators of the intelligent analysis method of this invention with those of the traditional manual interpretation method. The traditional manual interpretation method is significantly affected by the professional experience of the interpreters, resulting in low accuracy rates in rock mass quality classification, mechanical parameter inversion, and matching of adverse geological body attributes. The accuracy of surrounding rock stability assessment is also insufficient, and the report generation cycle is long, failing to meet the rapid construction requirements of urban subway projects. This invention, through an intelligent recognition network, machine learning parameter inversion model, and stability assessment model, achieves intelligent processing of the entire surrounding rock structure analysis process. This significantly improves the accuracy of rock mass quality classification, parameter inversion, adverse geological body analysis, and stability assessment, while also greatly shortening the output cycle. It is better suited to the refined detection and rapid analysis needs of soft rock and water-rich strata tunnels in urban subways.

[0063] Reference Figure 2 This diagram details the physical deployment logic and data quality assurance mechanism of the detection end. To eliminate the "multiple solutions" and spatial blind spots of a single detection method, this invention employs a combined array covering the tunnel face, tunnel sidewalls, and arch area. A seismic wave array handles deep detection, a ground-penetrating radar line handles shallow, fine scanning, and a high-density electrical resistivity tomography (EDT) array and acoustic wave testing provide supplementary parameters in electrical and acoustic dimensions, respectively. During deployment, the system dynamically adjusts the acquisition parameters according to geological conditions to achieve full cross-sectional coverage in both the horizontal and vertical directions. The diagram particularly highlights the "real-time quality control and dynamic supplementary measurement" loop: the signal-to-noise ratio and signal integrity are verified in real time at the acquisition site. Once an area with substandard data quality is identified, the system automatically generates a supplementary measurement plan and adjusts the parameters for targeted supplementary measurement. This "immediate assessment upon acquisition, supplementary measurement as needed" mechanism ensures the reliability and completeness of the input data required for subsequent deep learning analysis from the source.

[0064] Reference Figure 3 This paper focuses on the deep fusion process of multi-source heterogeneous data at the feature level, a key preprocessing step for intelligent analysis. The system first employs wavelet multi-resolution analysis to decompose the preprocessed seismic, radar, resistivity, and acoustic data at multiple scales, thereby extracting subtle wavefield features, electrical variations, and acoustic responses at different depths of the surrounding rock. The core logic lies in the "dynamic weight allocation mechanism": the system assigns contribution weights to each type of feature by calculating the inner product (correlation) between various geophysical feature vectors and standard surrounding rock feature reference vectors. Through normalization, these cross-modal features are mapped to a unified feature space, generating a "multi-scale fused feature map." This method not only strengthens the signal representation of anomaly areas in the surrounding rock (such as fracture zones or water-rich areas) but also improves the subsequent identification network's ability to distinguish features of complex geological structures such as stratigraphic interfaces, providing high-fidelity data input for "image-based object identification."

[0065] Reference Figure 4 This figure illustrates the deep learning network architecture logic for pixel-level segmentation of geological structures. The recognition network employs a hybrid model combining U-Net and Transformer, balancing local accuracy with global context awareness. The Transformer encoder captures long-distance spatial dependencies between features at different scales through a multi-head self-attention mechanism, enabling the identification of large-scale stratigraphic boundaries and tectonic trends. Meanwhile, the U-Net decoder structure, through upsampling and multi-level skip connections, preserves high-resolution spatial information at the lower levels, ensuring accurate identification of fine structures such as joints, fractures, and contact zones. The recognition results are then processed through spatial clustering and connected component analysis to achieve instance segmentation of geological structures. Furthermore, the system incorporates a low-confidence filtering mechanism, triggering local feature re-extraction and secondary recognition logic for areas with ambiguous recognition results. This dual-engine driven architecture significantly improves the robustness of automatic segmentation in complex tunnel geological environments.

[0066] Reference Figure 5 This diagram illustrates the logical flow of transforming identified geometric structures into engineering decision parameters. The system integrates the identified structural interfaces, inverted mechanical parameters (such as elastic modulus and Poisson's ratio), and attribute characteristics of adverse geological bodies (such as faults, karst caves, and water-rich zones) to construct a multi-dimensional rock mass quality evaluation system. Through comprehensive weighted calculations of integrity coefficient, compressive strength, joint density, and water-rich state, the system achieves quantitative grading of surrounding rock quality. Based on this, the system automatically assesses the stability of the surrounding rock structure, classifies risk levels, locates instability areas, and generates a three-dimensional visualized exploration report. The end of the process is also the beginning: using geological samples exposed in real time after tunnel excavation as "true values" for verification, the model is incrementally iterated through transfer learning technology, with only minor adjustments to the semantic recognition layer parameters. This allows the system to adaptively match the detection needs under different geological conditions, continuously improving analytical accuracy.

[0067] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for geophysical exploration and intelligent analysis of tunnel surrounding rock structures, characterized in that, Includes the following steps: The multi-source geophysical data acquisition steps for tunnel surrounding rock detection areas are used to simultaneously collect multiple types of geophysical data for the surrounding rock area to be detected, and to construct a multi-dimensional detection dataset. The standardization and spatial registration steps for multi-source geophysical data are used to standardize various types of acquired data and complete spatial coordinate registration based on a unified engineering coordinate system. The multi-scale surrounding rock structure geophysical feature extraction and cross-modal fusion steps are used to decompose the preprocessed data at multiple scales, extract multi-dimensional geophysical features, complete cross-modal deep fusion and normalization processing, and generate multi-scale fused feature maps. The intelligent identification step of surrounding rock geological structure based on deep learning network is used to input the fused feature map into the identification network and complete the pixel-level geological structure segmentation through the encoder and decoder. The intelligent inversion steps for the quality classification and physical and mechanical parameters of the surrounding rock mass in tunnels are used to combine geological structure and geophysical characteristics to complete the quality classification of the rock mass and invert the physical and mechanical parameters. The steps for precise location and attribute analysis of unfavorable geological bodies in surrounding rock are used to locate unfavorable geological bodies and analyze their characteristic parameters based on the results of fusion feature and structural identification. The intelligent assessment steps for the stability of tunnel surrounding rock structures are used to combine rock mass quality classification, physical and mechanical parameters and distribution of adverse geological bodies to complete the intelligent assessment of the stability of surrounding rock structures, classify stability levels and identify areas at risk of instability. The steps for 3D visualization of detection results and standardized output are used to 3D model and visualize various detection and analysis results, and generate standardized output reports that conform to engineering survey specifications. The incremental iterative optimization step of the surrounding rock detection and analysis model is used to collect actual geological data after excavation as a verification sample to complete the parameter fine-tuning and incremental iterative optimization of the model.

2. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, It also includes the deployment steps of a multi-method geophysical exploration array for tunnel surrounding rock, which is used to deploy seismic wave detection arrays, ground-penetrating radar survey lines, and high-density electrical resistivity tomography arrays in the tunnel face, tunnel sidewalls, and arch area according to the engineering geological conditions and exploration depth requirements of the tunnel. It determines the spacing between measuring points, the length of the survey lines, and the acquisition parameters of each detection method, and completes the deployment of a joint exploration array that covers the entire cross-section of the tunnel surrounding rock in both the transverse and longitudinal directions.

3. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, It also includes real-time quality control and dynamic supplementary testing steps for surrounding rock detection data, which are used to verify the signal-to-noise ratio, effective signal integrity, and spatial coverage of the acquired data in real time during the data acquisition process, compare it with the preset data quality threshold, identify areas where the data quality does not meet the standards, and automatically generate supplementary testing plans.

4. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, During the multi-source geophysical data acquisition process in the tunnel surrounding rock detection area, seismic wave reflection data acquisition was completed using a multi-channel high-resolution seismograph to collect transverse and longitudinal wave reflection signals of the tunnel surrounding rock. Ground-penetrating radar data acquisition was completed using a combination of antennas with different center frequencies, switching the corresponding frequency antenna for different detection depths. High-density electrical resistivity tomography data acquisition was completed using a Winner device, covering the acquisition of electrical information at different depths of the tunnel surrounding rock. Surrounding rock acoustic wave test data was completed using a combination of single-hole and cross-hole acoustic wave tests to obtain acoustic wave velocity data at different depths of the surrounding rock.

5. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, During the standardization preprocessing and spatial registration of multi-source geophysical data, direct wave removal, surface wave suppression, deconvolution, and static correction were performed on seismic wave data; zero-point correction, background denoising, gain restoration, and frequency band filtering were performed on ground-penetrating radar data; distortion data removal, terrain correction, and apparent resistivity calculation were performed on high-density electrical resistivity data; and first wave acquisition, sound velocity calculation, and outlier removal were performed on acoustic wave test data. All preprocessed data were mapped to the unified coordinate system of the tunnel project.

6. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, In the process of multi-scale geophysical feature extraction and cross-modal fusion of surrounding rock structures, a cross-modal feature fusion network is used to achieve deep fusion of different types of features. The expression for calculating the weight of multi-source feature fusion is as follows: , These are the feature fusion weight coefficients at the corresponding scale. This represents the total number of geophysical feature types. Number the feature type. The contribution weight of the i-th type of geophysical feature. The result of calculating the inner product between the i-th type of geophysical feature vector and the standard surrounding rock feature reference vector is given. Let i be the normalized geophysical feature vector of the i-th class. This is a standardized reference vector for surrounding rock characteristics. This is the 2-norm operator for vectors.

7. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, In the intelligent identification of surrounding rock geological structures based on deep learning networks, a hybrid architecture combining U-Net and Transformer is adopted as the backbone of the surrounding rock geological structure identification network. The global context information of surrounding rock features is extracted through the multi-head self-attention mechanism of the Transformer encoder to capture the long-distance spatial dependencies between different geological structures. Through the multi-level upsampling structure of the U-Net decoder, multi-scale feature fusion is completed by combining feature maps of different levels of the encoder, and pixel-level geological structure segmentation results of tunnel surrounding rock are output. Instance division of different geological structures is completed through connected component analysis and spatial clustering. At the same time, low-confidence identification areas are screened through feature similarity calculation, and local feature re-extraction and secondary identification are performed.

8. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, In the process of intelligent inversion of the quality classification and physical and mechanical parameters of tunnel surrounding rock mass, a quality classification and evaluation system for surrounding rock mass is constructed by combining the geological structure identification results of the surrounding rock, geophysical characteristic parameters, and indoor rock mechanics test data. The comprehensive score calculation expression for the quality of surrounding rock mass is as follows: , A comprehensive score for the quality of the surrounding rock mass. The weight of the rock mass integrity coefficient. This represents the measured value of the integrity coefficient of the surrounding rock mass. This represents the maximum value of the rock mass integrity coefficient. The weight of the uniaxial compressive strength of the rock. This is the inversion value of the uniaxial compressive strength of the rock. This represents the standard maximum value of the uniaxial compressive strength of rock. Weighted by the volumetric density of joints and fractures. This represents the measured value of the bulk density of the joints and fissures in the surrounding rock. This represents the standard maximum value of the bulk density of joints and fissures. The weight of the water-rich state of the surrounding rock. This represents the quantitative value of the water-rich state of the surrounding rock. The maximum value is quantified for the water-rich state of the surrounding rock.

9. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, In the process of accurately locating and analyzing the attributes of unfavorable geological bodies in the surrounding rock, a geophysical feature sample library of unfavorable geological bodies is constructed based on the multi-scale fusion feature map and geological structure identification results. The type identification of unfavorable geological bodies is completed through a deep learning classification network. Combined with the forward and inverse results of multi-source geophysical data, the spatial boundary, burial depth, and size of unfavorable geological bodies are determined. The filling medium type, water-rich state, and degree of fragmentation attribute parameters of unfavorable geological bodies are analyzed through multi-physics field joint inversion. At the same time, the spatial coordinate location of unfavorable geological bodies is completed based on the tunnel engineering coordinate system.

10. The method for geophysical exploration and intelligent analysis of tunnel surrounding rock structure according to claim 1, characterized in that, During the incremental iterative optimization of the surrounding rock detection and analysis model, the actual geological structure revealed after tunnel excavation, rock mass physical and mechanical test data, and surrounding rock stability field monitoring data were compiled into validation samples and incremental training samples. These samples were divided into training set, validation set, and test set in an 8:1:1 ratio. Transfer learning combined with incremental training was used to fine-tune the parameters of the geological structure identification network, parameter inversion model, and stability assessment model. The pre-training parameters of the network backbone feature extraction layer were frozen. The model accuracy was verified through the validation set, and the model generalization ability was tested through the test set. The iteratively optimized model replaced the original model and was used for subsequent geophysical detection and intelligent analysis of the surrounding rock of the tunnel.

Images