A method for constructing and applying a digital twin model of tunnel surrounding rock, and a digital twin model system for tunnel surrounding rock.

Abstract

This invention relates to a method and system for constructing and applying a digital twin model of tunnel surrounding rock. Based on the RFM model, tunnel surrounding rock data is labeled and then hierarchically clustered to output surrounding rock grading results. A model-driven architecture, combined with deep learning algorithms, is used to construct a digital twin model. The surrounding rock grading results are used to virtually replicate the tunnel surrounding rock conditions, forming a virtual model. The data from the digital twin model is stored, retrieved, and transmitted through a relational database, mapping the virtual model to the construction site and updating the tunnel surrounding rock data in real time. The system execution method includes a data perception module, a twin model construction module, a data management module, and an application interaction module. This invention significantly improves the efficiency of tunnel surrounding rock data acquisition, achieves high accuracy in surrounding rock grade identification, provides real-time data support for construction parameter optimization, and achieves high accuracy in model data acquisition, enabling intelligent decision support throughout the entire process.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, and in particular to a method for constructing and applying a digital twin model of tunnel surrounding rock that combines tunnel engineering and digital twin technology, as well as a digital twin model system for tunnel surrounding rock. Background Technology

[0002] With the rapid development of transportation infrastructure construction, the scale of mountain tunnel projects is constantly expanding, and geological conditions are becoming increasingly complex. As the main construction method for mountain tunnel excavation, the drilling and blasting method directly affects the overall project progress and safety in terms of its construction quality and efficiency. During drilling and blasting construction, accurate collection and timely updating of tunnel surrounding rock data are crucial prerequisites for scientifically setting drilling and blasting parameters, controlling the smooth blasting effect, optimizing the excavation outline, and reserving deformation allowances.

[0003] In existing technologies, tunnel surrounding rock data acquisition mainly relies on manual operation of cross-section scanners. Construction personnel need to collect data multiple times based on site conditions to ensure accuracy. This approach often suffers from the following technical drawbacks: (1) Manual operation of the cross-section scanner requires construction personnel to repeatedly measure, record and verify. The time for a single acquisition is usually between 47 and 70 minutes, with an average time of about 57 minutes. Taking a certain long tunnel as an example, more than 30,000 data collections of surrounding rock are required during the construction period. The overall workload is huge and seriously restricts the efficiency of drilling and blasting cycle operations. (2) In the traditional model, the collected surrounding rock data are mostly stored in paper records or discrete spreadsheets, lacking systematic data labels and classification mechanisms, making it difficult to form a structured dataset that can be used for intelligent analysis. Construction personnel mainly rely on experience to judge the surrounding rock grade, which is highly subjective. The consistency and accuracy of the grading results are difficult to guarantee, and the interpretability is poor. (3) Tunnel construction is generally dynamic and the surrounding rock conditions change continuously with the excavation progress. Traditional methods can only provide static data at a certain time point, and cannot realize real-time monitoring and dynamic updating of the surrounding rock condition, resulting in parameter adjustment lagging behind the actual working conditions, affecting blasting effect and construction safety. (4) The existing technology has not established a mechanism that can link the physical surrounding rock entity with the virtual model in real time. Construction personnel cannot predict the trend of surrounding rock changes through the virtual environment, and it is also difficult to simulate and optimize the scheme in the virtual space.

[0004] In recent years, digital twin technology, by constructing digital mappings of physical objects to achieve synchronization between the virtual and the real world, has been applied in fields such as intelligent manufacturing and urban management. However, existing digital twin technologies are mainly geared towards industrial equipment or urban buildings, and there are technical obstacles to directly applying them to tunnel surrounding rock data management, including: the lack of tagged models and intelligent classification algorithms suitable for the characteristics of tunnel surrounding rock data, the lack of a data transmission architecture that can drive real-time updates of digital twin models, and the absence of a complete technical closed loop from surrounding rock perception to model mapping. Summary of the Invention

[0005] This invention solves the problems existing in the prior art and provides a method for constructing and applying a digital twin model of tunnel surrounding rock, as well as a digital twin model system for tunnel surrounding rock.

[0006] The technical solution adopted in this invention is a method for constructing and applying a digital twin model of tunnel surrounding rock, comprising the following steps: S1 uses the RFM model to label tunnel surrounding rock data, performs hierarchical clustering on the labeled data, and outputs the surrounding rock classification results. S2 is based on a model-driven architecture and combines deep learning algorithms to build a digital twin model. It uses the surrounding rock classification results to virtually replicate the surrounding rock conditions of the tunnel and form a virtual model. S3 uses a relational database to store, retrieve, and transmit data from digital twin models; S4 maps the virtual model to the construction site and updates the tunnel surrounding rock data in real time.

[0007] Preferably, in S1, the tunnel surrounding rock data includes the straight lines, curves, turning angles, slopes, vertical curves, and cross-sectional surface features of the tunnel surrounding rock.

[0008] Preferably, in S1, the tunnel surrounding rock data is identified by performing feature analysis on the cluster centers of each category of data in the RFM model.

[0009] Preferably, in S2, the model-driven architecture includes a perception layer, a digital twin network layer, and a physical data layer, used to drive the transmission flow of tunnel surrounding rock data; the digital twin network layer introduces a Mamba module and a hybrid attention mechanism; The Mamba module is used to receive video data of the surrounding rock of the tunnel collected by the perception layer and output the first feature vector, which is used to update the geometry of the three-dimensional virtual model. The hybrid attention module receives multimodal data collected by the perception layer and outputs a second feature vector for intelligent rock classification or drilling and blasting parameter decision-making.

[0010] Preferably, the digital twin network layer includes: The input layer has a first input interface and a second input interface, which receive video data and text data respectively. The encoding module includes a visual encoder and a word segmenter, which output first encoded information and second encoded information, respectively. The Mamba module receives the first encoded information and outputs the first feature vector for real-time updates of the virtual model. The cross-attention module outputs a fused token based on the first and second encoded information. The self-attention module receives the second encoded information and outputs self-attention features. The second encoded information, fusion token, and self-attention features are added together and output as a second feature vector through an MLP layer, which is used for surrounding rock classification and decision-making.

[0011] Preferably, in S2, the deep learning algorithm uses a semi-automatic annotation method to annotate the data of each layer of the digital twin model, and trains the model through forward propagation and backward propagation.

[0012] Preferably, in S4, the model application layer is constructed based on B / S architecture, linear structure, and WebGL library; The B / S architecture includes a basic environment layer, an information resource layer, an application support layer, and a business application layer, which are used to integrate model business modules and facilitate data interaction. Linear structures are used to logically connect the various business modules of a digital twin model; The WebGL library is used to convert digital twin model code into a 3D graphics mode, generating tunnel views.

[0013] Preferably, the virtual model data output by the digital twin model is compared with the actual measured data on site to calculate the accuracy of model data acquisition; When the accuracy of model data acquisition is lower than a preset threshold, the model calibration process is triggered. The model calibration process includes: re-importing tunnel surrounding rock data, adjusting the network weights of the deep learning algorithm, and updating the parameters of the digital twin model.

[0014] A digital twin model system for tunnel surrounding rock, used to execute the method for constructing and applying a digital twin model for tunnel surrounding rock, includes: The data perception module is used to label tunnel surrounding rock data based on the RFM model, and to intelligently classify the labeled data through clustering algorithms, outputting the surrounding rock classification results. The twin model construction module is used to construct a digital twin model based on a model-driven architecture and combined with deep learning algorithms. It uses the surrounding rock classification results to virtually replicate the surrounding rock conditions of the tunnel and generate a virtual model. The data management module is used to store, retrieve, and transmit digital twin model data through a relational database; An application interaction module is used to map the virtual model to the construction site and update the tunnel surrounding rock data in real time; A data synchronization interface is provided between the data perception module and the twin model construction module; a two-way data channel is provided between the application interaction module and the data management module.

[0015] Preferably, the application interaction module includes: The surrounding rock intelligent identification unit is used to display lithology, integrity, groundwater outflow status, and drilling parameters; Excavation and support decision-making unit, used to provide excavation and support schemes and tunnel structure safety monitoring and assessment; The blasting unit is used to output the drilling and blasting hole layout design drawing and drilling and blasting parameters; Equipped with intelligent sensing units for real-time data collection from construction equipment; Safety and quality monitoring unit, used to monitor safety and quality indicators during construction; The decision-making unit is used to integrate data from various units and output comprehensive decision-making instructions.

[0016] This invention relates to a method for constructing and applying a digital twin model of tunnel surrounding rock, and a system for such a model. The method involves labeling tunnel surrounding rock data based on an RFM model, performing hierarchical clustering on the labeled data, and outputting surrounding rock grading results. A model-driven architecture is used, combined with deep learning algorithms, to construct a digital twin model. The surrounding rock grading results are then used to virtually replicate the tunnel's surrounding rock conditions, forming a virtual model. The data from the digital twin model is stored, retrieved, and transmitted through a relational database. The virtual model is mapped to the construction site, and the tunnel surrounding rock data is updated in real time. The system execution method includes a data perception module, a twin model construction module, a data management module, and an application interaction module.

[0017] The beneficial effects of this invention are as follows: (1) Automatic labeling and intelligent classification of surrounding rock data are achieved through RFM model and clustering algorithm. Combined with real-time mapping of digital twin model, the data acquisition time of tunnel surrounding rock is reduced from an average of 57 minutes in traditional methods to 7.8 minutes, with an efficiency improvement of about 630%, and the acquisition efficiency is significantly improved. (2) Establish a structured surrounding rock data labeling system, and realize the automatic identification of surrounding rock grade through cluster center feature analysis. The identification accuracy rate reaches 99.73%, avoiding the subjectivity of manual judgment; (3) Through model-driven architecture and bidirectional data channels, real-time synchronization between physical surrounding rock entities and virtual models is achieved. The digital twin model can be dynamically updated with the construction progress, providing real-time data support for the optimization of construction parameters. (4) Scientifically allocate computing resources to ensure that the digital twin model operates efficiently while meeting real-time requirements, and the model data acquisition accuracy reaches 99.87%; (5) The integrated functional modules of surrounding rock intelligent identification, excavation support decision-making, blasting optimization, safety and quality monitoring provide intelligent decision support for tunnel construction from surrounding rock classification to drilling and blasting parameter output. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention; Figure 3 This is a schematic diagram of the hierarchical clustering of tunnel surrounding rock in this invention; Figure 4 This is a schematic diagram of the digital twin network layer in this invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0020] This invention relates to a method for constructing and applying a digital twin model of tunnel surrounding rock, comprising the following steps: S1 uses the RFM model to label tunnel surrounding rock data, performs hierarchical clustering on the labeled data, and outputs the surrounding rock classification results. S2 is based on a model-driven architecture and combines deep learning algorithms to build a digital twin model. It uses the surrounding rock classification results to virtually replicate the surrounding rock conditions of the tunnel and form a virtual model. S3 uses a relational database to store, retrieve, and transmit data from digital twin models; S4 maps the virtual model to the construction site and updates the tunnel surrounding rock data in real time.

[0021] The following uses a particularly long tunnel project as an example to illustrate the steps.

[0022] S1 uses the RFM model to label tunnel surrounding rock data, performs hierarchical clustering on the labeled data, and outputs the surrounding rock classification results. First, data acquisition and feature extraction are performed. The tunnel surrounding rock data includes straight lines, curves, turning angles, slopes, vertical curves, and cross-sectional surface features. Specifically, a cross-sectional scanner is deployed near the tunnel excavation face to collect geometric feature data of the tunnel surrounding rock. The collected feature parameters include straight line segment length (unit: m), curve segment radius of curvature (unit: m), turning angle (unit: °), slope value (unit: %), vertical curve radius (unit: m), and cross-sectional surface data (unit: m²), totaling six feature parameters. In this embodiment, a total of 5000 sets of surrounding rock data are collected as a sample set.

[0023] The RFM model is then constructed and labeled; the three dimensions of the RFM model include: R (proximity), the time interval since the last data acquisition, reflects the timeliness of the surrounding rock data; F (frequency) is the frequency of data collection for the surrounding rock in this area, reflecting the data update density. M (value score) is the overall quality score of surrounding rock data, reflecting the reliability of the data. Using the average values ​​of R, F, and M as boundaries, the surrounding rock data were divided into three categories: high value, medium value, and low value. The RFM model was implemented using Python programming, running on an Intel Xeon Gold 6248R processor and 64GB of memory. Testing showed that the average runtime of the 10 computing power configurations was 1.8 minutes, the average accuracy of the computing power configuration was 99.45%, and the average data classification time was 3202ms, meeting the design requirements.

[0024] The K-means clustering algorithm is used to perform hierarchical clustering on the labeled data output by the RFM model. The input is a labeled surrounding rock dataset with a preset number of clusters K=3 (corresponding to Class I, II, and III surrounding rock). K cluster centers are randomly selected, and the Euclidean distance from each sample to each cluster center is calculated. Samples are assigned to the nearest cluster center, and the cluster center is updated to the mean of all samples in that cluster. The iteration stops when the change in cluster center is less than a threshold. By performing feature analysis on the cluster centers of each category of data in the RFM model, tunnel surrounding rock data can be identified.

[0025] In this embodiment, it was verified that the average accuracy of the clustering algorithm in identifying surrounding rocks was 99.73%.

[0026] S2 is based on a model-driven architecture and combines deep learning algorithms to build a digital twin model. It uses the surrounding rock classification results to virtually replicate the surrounding rock conditions of the tunnel and form a virtual model. (2-1) Building a model-driven architecture The model-driven architecture includes a perception layer, a digital twin network layer, and a physical data layer, which drive the transmission flow of tunnel surrounding rock data; the digital twin network layer introduces Mamba modules and a hybrid attention mechanism. The Mamba module is used to receive video data of the surrounding rock of the tunnel collected by the perception layer and output the first feature vector, which is used to update the geometry of the three-dimensional virtual model. The hybrid attention module receives multimodal data collected by the perception layer and outputs a second feature vector for intelligent rock classification or drilling and blasting parameter decision-making.

[0027] Furthermore, the digital twin network layer includes: The input layer has a first input interface and a second input interface, which receive video data and text data respectively. The encoding module includes a visual encoder and a word segmenter, which output first encoded information and second encoded information, respectively. The Mamba module receives the first encoded information and outputs the first feature vector for real-time updates of the virtual model. The cross-attention module outputs a fused token based on the first and second encoded information. The self-attention module receives the second encoded information and outputs self-attention features. The second encoded information, fusion token, and self-attention features are added together and output as a second feature vector through an MLP layer, which is used for surrounding rock classification and decision-making.

[0028] In this invention, a model-driven architecture is used to construct the underlying framework of the digital twin model. The perception layer is equipped with sensors and data acquisition devices, including but not limited to cross-section scanners, displacement sensors, and stress sensors, to acquire physical parameters such as the geometric shape, deformation, and stress of the tunnel surrounding rock in real time. The digital twin network layer constructs a virtual mapping model of the surrounding rock data to realize the conversion from physical entities to digital space, including functional modules such as data fusion, feature extraction, and model inference. The physical data layer interacts with a relational database through a data interface to realize the persistent storage and retrieval of model data.

[0029] In this embodiment, the digital twin network layer first constructs a multimodal feature extraction network to extract depth features from the original video and text data. The digital twin network layer adopts a dual-output structure to simultaneously meet the requirements of intelligent rock mass classification and real-time updates of the 3D model. Specifically: The network input layer contains two interfaces: the first interface receives video data (frame sequence, sampling rate 30fps) of the tunnel surrounding rock collected by the cross-section scanner and camera; the second interface receives text description data, including surrounding rock labels, drilling and blasting records, support parameters, etc. In the coding layer, the video data is processed by a visual encoder (using ResNet-50, pre-trained on the ImageNet dataset) to extract spatial features and output the first encoded information A, which is 256-dimensional. Each frame of the image generates a 256-dimensional feature vector after passing through the encoder. The entire video sequence forms a feature sequence A = [a1, a2, ..., aT], where T is the sequence length. The text data is transformed by a word segmenter (WordPiece) and an embedding layer (256 dimensions) to output the second encoded information B, which has a size of 256 dimensions. Each text record corresponds to a 256-dimensional feature vector. The first encoded information A is fed into both the Mamba module and the cross-attention module in the hybrid attention module; the Mamba module is built based on a state-space model, and its core computation is h. t =A*h t-1 +B*x t y t =C*h t , where x t For the input at time step t, h t In the hidden state, y t The output at time t is A, B, and C, which are learnable parameter matrices. The final output Y=[y1, y2, ..., yT] of the Mamba module after processing the entire sequence is obtained by global average pooling to obtain the first feature vector (dimension 128). This output is used to drive the real-time update of the geometry of the 3D virtual model, such as tunnel contour deformation and over- or under-excavation area labeling. Compared with Transformer, the computational complexity of the Mamba module is linearly related to the sequence length T, making it suitable for processing long video streams. The cross-attention module uses the first encoded information A as the query and the second encoded information B as the key and value. After calculating the attention weights, it outputs a fusion token to enhance the text features with visual information. Meanwhile, the second encoded information B is also fed into the self-attention module in the hybrid attention module to capture the dependencies within the text sequence and output self-attention features; The adder adds the second encoded information B, the fusion token, and the self-attention feature to obtain the combined feature, which is then output as a second feature vector through two layers of MLP (512-dimensional hidden layer) for intelligent classification of surrounding rock (outputting the probability of level I-V) and decision-making on drilling and blasting parameters.

[0030] In the above dual-output structure, the two feature vectors focus on semantic understanding and decision-making, and real-time mapping of geometric shapes, respectively. Together, they constitute the complete output of the digital twin network layer and are input to the corresponding units in the application interaction module.

[0031] In this embodiment, the MQTT protocol is used for data transmission between the perception layer and the digital twin network layer to ensure low latency and high reliability; the ODBC interface is used for data interaction between the digital twin network layer and the physical data layer; the model-driven architecture uses JavaScript to write the data source code, collects surrounding rock data through the perception layer, processes it through the digital twin network layer, and then transmits it to the physical data layer.

[0032] (2-2) Deep Learning Algorithm Construction and Training In this invention, a deep learning algorithm is used to construct a digital twin model. The input layer receives the surrounding rock grading results (second feature vector) output by the RFM model and clustering algorithm. The input layer has a dimension of 6 and corresponds to six feature parameters: straight line, curve, turning angle, slope, vertical curve, and cross-sectional surface. The hidden layer has three fully connected layers with 128, 64, and 32 neurons respectively, and the activation function is ReLU. The output layer outputs the parameter set of the virtual model with a dimension of 8, corresponding to the geometric shape and mechanical parameters of the surrounding rock.

[0033] During the training process, the deep learning algorithm inputs tunnel surrounding rock data into the network in batches, with each batch size set to 32. The prediction results are calculated, and the prediction accuracy is evaluated using the loss function (mean squared error MSE). The gradient is calculated, and the network weights are updated. Optimization strategies include: using the Adam optimizer for hyperparameter tuning with a learning rate set to 0.001; employing Dropout (dropout rate set to 0.3) and Early Stopping (stopping training when the validation set loss does not decrease for 10 consecutive rounds) to enhance the model's generalization ability.

[0034] (2-3) Implementation of semi-automatic annotation The deep learning algorithm uses a semi-automatic annotation method to annotate the data of each layer of the digital twin model, and trains the model through forward and backward propagation. First, typical surrounding rock sections are manually annotated, and 30 representative Class I, II, and III surrounding rock sections are selected as seed samples. Unannotated samples are automatically annotated through transfer learning. The model pre-trained on the seed samples is used to predict new data and generate a preliminary sample dataset. The annotation results are then combined with the deep learning algorithm for predictive analysis and verification. Annotations with high confidence are included in the training set, and the model is iteratively optimized.

[0035] (2-4) Optimization of computing power configuration To ensure that the digital twin model meets real-time requirements, the minimum computing power requirement is determined according to the following formula: F min_unified =max(C total / T compute , λ×Ctotal ) Among them, T compute =L target -t in -t out ; In this embodiment, the target total time L is set. target The time taken for data input is t, which is 30 minutes. in The runtime is 1.8 min (based on the RFM model running time), and the data output time is t. out =1.9min (based on the running time of the digital twin model), then T compute The time was 26.3 minutes, and the total computation time was C. total The model complexity is estimated to be 3.2 * 10. 9 With a redundancy coefficient λ of 0.2 for each floating-point operation, the calculated minimum computing power requirement is approximately 2.4 * 10^6. 7 Floating-point operations per second.

[0036] (2-5) Data Acquisition Model Construction The digital twin model (input-output relationship) adopts the data acquisition model y=f(x)+ε, where x is the input feature parameter (vector), x=[x1,x2,x3,x4,x5,x6]. T , respectively, correspond to straight line length, radius of curvature, turning angle, slope value, vertical curve radius, and cross-sectional area; y is the output data (the set of surrounding rock parameters of the virtual model), ε is the error term, which follows a normal distribution with a mean of 0 and a standard deviation of 0.01; The data storage set is represented as S={(x i ,y i x | i=1,2,...,n} i Let y be the input feature vector of the i-th sample. i Let n be the output label of the i-th sample, and n be the total number of samples, which is 5000 in this embodiment. Data analysis uses least squares estimation to determine the function f; the optimization problem f(x) = argmin is solved. f∈F ∑ i n =1 (y i -f(x i Obtain the optimal model parameters.

[0037] The model fit R was calculated to be... 2 A value of 0.96 or higher meets the requirements for engineering applications.

[0038] S3 uses a relational database to store, retrieve, and transmit data from digital twin models; Specifically, a relational database is built using SQL Server 2019, and the following core data tables are created: The original data table of the surrounding rock is used to store the original characteristic data such as straight lines, curves, turning angles, slopes, vertical curves, and cross-sectional tables. The label data table is used to store the R, F, and M values ​​generated by the RFM model, as well as the label classification results. The grading results table is used to store the surrounding rock grades output by the clustering algorithm; The virtual model parameter table is used to store the set of surrounding rock parameters output by the digital twin model; The drilling and blasting parameter table is used to store the hole layout scheme and charge parameters output by the blasting unit.

[0039] Furthermore, to improve data retrieval efficiency, this invention establishes a data caching cluster, using Redis as the caching middleware; in this embodiment, the caching strategy is configured as follows: The cache expiration time for frequently accessed data (such as the latest surrounding rock classification results and current drilling and blasting parameters) is set to 3600 seconds. Low-frequency access data (such as historical raw data) is not cached and is read directly from the database; The cache refresh mechanism is automatically triggered when cached data expires or is updated.

[0040] Tests showed that the average data retrieval speed was 3300ms, the average data storage accuracy was 99.64%, and the average data packet loss rate was 0.19%.

[0041] S4 maps the virtual model to the construction site and updates the tunnel surrounding rock data in real time; To be precise, this refers to matching and mapping the virtual model with the spatial location of the construction site under a unified coordinate system, and updating the tunnel surrounding rock data in real time.

[0042] The model application layer is built based on B / S architecture, linear structure, and WebGL library; The B / S architecture includes a basic environment layer, an information resource layer, an application support layer, and a business application layer, which are used to integrate model business modules and facilitate data interaction.

[0043] Specifically, the basic environment layer deploys a CentOS 7.9 operating system, an Nginx 1.20 web server, and a Node.js 14 runtime environment; the information resource layer integrates a relational database (SQL Server), a file storage system (MinIO), and a caching service (Redis); the application support layer provides public services such as identity authentication (JWT), access control (RBAC), and log auditing (ELK); and the business application layer carries business functions such as intelligent rock identification, excavation and support decision-making, blasting optimization, and safety and quality monitoring.

[0044] The linear structure is used to logically connect the various business modules of the digital twin model. Specifically, the business modules are organized in the form of a singly linked list, and the modules interact with each other through a RESTful API. The linear structure sets the logical order of the business modules through pointer fields, so as to realize the sequential access and data flow of the business modules.

[0045] The WebGL library is used to convert digital twin model code into 3D graphics mode to generate tunnel views. Specifically, it includes tunnel 3D model rendering, visualization of surrounding rock grades (using different colors to indicate the distribution of Class I, II, and III surrounding rock), and parameter annotation. The parameter annotation involves directly annotating parameters such as borehole location, depth, and charge amount on the model, and dynamically updating the model display by establishing a long connection with the backend and refreshing the surrounding rock monitoring data every preset time, such as 30 seconds.

[0046] In fact, the present invention also includes model verification and correction in application, and realizes its update and iteration; Specifically, the virtual model data output by the digital twin model is compared with the on-site measured data to calculate the accuracy of model data acquisition. In this embodiment, 30 verification points are selected, and on-site measured data and virtual model output data are collected respectively. The absolute error and relative error of each verification point are calculated, and the accuracy is statistically analyzed. When the accuracy of model data acquisition is lower than a preset threshold, such as 99%, the model calibration process is triggered. The model calibration process includes: Re-import the tunnel surrounding rock data, prioritizing data collected within the most recent construction period; Adjusting the network weights of the deep learning algorithm: The incremental learning approach is adopted to retain the effective knowledge of the original model and only make minor adjustments to the weights for the new data. Update the parameters of the digital twin model, write the updated model parameters into a relational database, and replace the original virtual model; Once completed, recalculate the accuracy of the model data acquisition, and complete the correction after confirming that it has reached or exceeded the threshold.

[0047] This invention also relates to a digital twin model system for tunnel surrounding rock, used to execute the aforementioned method for constructing and applying a digital twin model of tunnel surrounding rock, comprising: The data perception module is deployed at the tunnel construction site. It collects surrounding rock data through cross-section scanners and sensors. It is used to label the surrounding rock data based on the RFM model and intelligently classifies the labeled data through clustering algorithms, outputting the surrounding rock classification results. The twin model construction module is deployed on the information center server. It is used to construct a digital twin model based on a model-driven architecture and combined with deep learning algorithms. The surrounding rock classification results are used to virtually replicate the surrounding rock conditions of the tunnel and generate a virtual model. The data management module, built with SQL Server and Redis, is used to store, retrieve, and transmit digital twin model data through relational databases. The application interaction module, based on B / S architecture and WebGL library, is used to map the virtual model to the construction site and update the tunnel surrounding rock data in real time. A data synchronization interface is set up between the data perception module and the twin model construction module, which uses the WebSocket protocol to achieve real-time data push. After the data perception module completes the acquisition and classification of surrounding rock data, it pushes the classification results to the twin model construction module in JSON format through this interface. The push frequency is once every 30 seconds to ensure that the synchronization delay between the virtual model and the physical entity is less than 500ms. A bidirectional data channel is set up between the application interaction module and the data management module, and the gRPC protocol is used to achieve efficient data interaction. The application interaction module queries historical surrounding rock data, drilling and blasting parameters, etc. from the data management module through this channel, and sends user operation instructions, such as model cutting and parameter adjustment, to the data management module through this channel, which triggers the corresponding operation of the twin model construction module.

[0048] The application interaction module includes: The surrounding rock intelligent identification unit uses a support vector machine classification algorithm to display lithology, integrity, groundwater discharge status, and drilling parameters; in fact, it simultaneously outputs the surrounding rock grade. The excavation and support decision unit, based on the surrounding rock classification results, is used to provide excavation and support schemes (shotcrete, anchor bolts, steel arches, etc. and corresponding support parameters) and tunnel structure safety monitoring and assessment. The blasting unit, based on a digital twin model, simulates the blasting effect through finite element analysis, optimizes the hole layout scheme and charge parameters, and is used to output the drilling and blasting hole layout scheme design drawing and drilling and blasting parameters; The equipment is equipped with an intelligent sensing unit, which uses an Internet of Things platform to collect real-time data on construction equipment, including working status, operation parameters, energy consumption information, fault codes, etc., to achieve remote monitoring and fault early warning of the equipment. The safety and quality monitoring unit integrates data from monitoring devices such as displacement sensors, stress sensors, and convergence meters to monitor safety and quality indicators during the construction process, including real-time monitoring of safety indicators such as tunnel deformation, lining stress, and arch settlement, and triggering alarms when necessary. The decision-making unit is used to integrate data from various units and output comprehensive decision-making instructions.

[0049] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0053] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0054] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for constructing and applying a digital twin model of tunnel surrounding rock, characterized in that, Includes the following steps: S1 uses the RFM model to label tunnel surrounding rock data, performs hierarchical clustering on the labeled data, and outputs the surrounding rock classification results. S2 is based on a model-driven architecture and combines deep learning algorithms to build a digital twin model. It uses the surrounding rock classification results to virtually replicate the surrounding rock conditions of the tunnel and form a virtual model. S3 uses a relational database to store, retrieve, and transmit data from digital twin models; S4 maps the virtual model to the construction site and updates the tunnel surrounding rock data in real time.

2. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 1, characterized in that, In S1, the tunnel surrounding rock data includes straight lines, curves, turning angles, slopes, vertical curves, and cross-sectional surface features of the tunnel surrounding rock.

3. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 1, characterized in that, In S1, the surrounding rock data of the tunnel is identified by performing feature analysis on the cluster centers of each category of data in the RFM model.

4. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 1, characterized in that, In S2, the model-driven architecture includes a perception layer, a digital twin network layer, and a physical data layer, which drive the transmission flow of tunnel surrounding rock data; the digital twin network layer introduces a Mamba module and a hybrid attention mechanism; The Mamba module is used to receive video data of the surrounding rock of the tunnel collected by the perception layer and output the first feature vector, which is used to update the geometry of the three-dimensional virtual model. The hybrid attention module receives multimodal data collected by the perception layer and outputs a second feature vector for intelligent rock classification or drilling and blasting parameter decision-making.

5. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 4, characterized in that, The digital twin network layer includes: The input layer has a first input interface and a second input interface, which receive video data and text data respectively. The encoding module includes a visual encoder and a word segmenter, which output first encoded information and second encoded information, respectively. The Mamba module receives the first encoded information and outputs the first feature vector for real-time updates of the virtual model. The cross-attention module outputs a fused token based on the first and second encoded information. The self-attention module receives the second encoded information and outputs self-attention features. The second encoded information, fusion token, and self-attention features are added together and output as a second feature vector through an MLP layer, which is used for surrounding rock classification and decision-making.

6. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 1, characterized in that, In S2, the deep learning algorithm uses a semi-automatic annotation method to annotate the data of each layer of the digital twin model, and trains the model through forward propagation and backward propagation.

7. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 1, characterized in that, In S4, the model application layer is built based on B / S architecture, linear structure and WebGL library; The B / S architecture includes a basic environment layer, an information resource layer, an application support layer, and a business application layer, which are used to integrate model business modules and facilitate data interaction. Linear structures are used to logically connect the various business modules of a digital twin model; The WebGL library is used to convert digital twin model code into a 3D graphics mode, generating tunnel views.

8. The method for constructing and applying a digital twin model of tunnel surrounding rock according to claim 1, characterized in that, The virtual model data output by the digital twin model is compared with the actual measured data on site to calculate the accuracy of the model data acquisition. When the accuracy of model data acquisition is lower than a preset threshold, the model calibration process is triggered. The model calibration process includes: re-importing tunnel surrounding rock data, adjusting the network weights of the deep learning algorithm, and updating the parameters of the digital twin model.

9. A digital twin model system for tunnel surrounding rock, characterized in that, A method for constructing and applying a digital twin model of tunnel surrounding rock as described in any one of claims 1 to 8 includes: The data perception module is used to label tunnel surrounding rock data based on the RFM model, and to intelligently classify the labeled data through clustering algorithms, outputting the surrounding rock classification results. The twin model construction module is used to construct a digital twin model based on a model-driven architecture and combined with deep learning algorithms. It uses the surrounding rock classification results to virtually replicate the surrounding rock conditions of the tunnel and generate a virtual model. The data management module is used to store, retrieve, and transmit digital twin model data through a relational database; An application interaction module is used to map the virtual model to the construction site and update the tunnel surrounding rock data in real time; A data synchronization interface is provided between the data perception module and the twin model construction module; a two-way data channel is provided between the application interaction module and the data management module.

10. A digital twin model system for tunnel surrounding rock according to claim 9, characterized in that, The application interaction module includes: The surrounding rock intelligent identification unit is used to display lithology, integrity, groundwater outflow status, and drilling parameters; the excavation and support decision unit is used to provide excavation and support schemes and tunnel structure safety monitoring and assessment. The blasting unit is used to output the drilling and blasting hole layout design drawing and drilling and blasting parameters; Equipped with intelligent sensing units for real-time data collection from construction equipment; Safety and quality monitoring unit, used to monitor safety and quality indicators during construction; The decision-making unit is used to integrate data from various units and output comprehensive decision-making instructions.

Images