A high-precision velocity modeling method and system for deep long tunnels

Abstract

The application discloses a high-precision velocity modeling method and system for a deep-buried long tunnel, relates to the engineering geophysical prospecting technical field, and comprises the following steps: acquiring wave velocity data in a drill hole of the deep-buried long tunnel and resistivity data at a drill hole position; constructing a selective depth state space model; subsequently, adjusting hyperparameters through cross iteration, taking the minimization of the difference between predicted velocity and actual velocity as the target, and iteratively optimizing the deep learning network model; applying the trained model to a test area, calculating and outputting a deep-buried long tunnel velocity prediction profile. According to the scheme, the deep learning model is introduced, combined with multi-source geological information such as drill hole wave velocity data and three-dimensional resistivity data, high-precision velocity modeling is realized, the limitations of traditional methods are overcome, the precision and reliability of deep-buried long tunnel velocity modeling are improved, and a brand-new solution is provided for deep-buried long tunnel velocity modeling.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engineering geophysical prospecting technology, in particular to a deep-buried long tunnel high-precision velocity modeling method and system. BACKGROUND

[0002] In the field of engineering construction, especially in the exploration and design of deep-buried long tunnels, accurately obtaining the velocity information of the strata along the tunnel is crucial for assessing the stability of the tunnel, designing a reasonable support structure, and predicting potential geological disasters. Traditional velocity modeling methods mainly rely on seismic wave exploration technology, which measures the propagation time of seismic waves in underground media, combines known medium density, and uses physical formulas to calculate the velocity distribution. However, this method faces many challenges and limitations in practical application.

[0003] Firstly, the geological conditions of deep-buried long tunnels are complex and variable, and factors such as stratum structure, lithology distribution, and groundwater conditions will significantly affect the propagation of seismic waves, resulting in large uncertainties in the velocity modeling results of traditional methods. Secondly, seismic wave exploration technology usually requires the deployment of a large number of geophones and sources, which is costly and complex to operate, especially in rugged terrain and inaccessible mountainous areas. In addition, the collection and processing of seismic wave exploration data is tedious and susceptible to noise interference, further limiting its application effect in practical engineering. Generally, it is convenient to collect geomagnetic data, and although inversion can provide large-scale resistivity data, it lacks completeness in low-frequency and high-frequency information, and its ability to depict underground conditions is insufficient. Therefore, it is necessary to obtain velocity information to depict the underground conditions, but the collection of large-scale velocity data is costly and influenced by many factors. Therefore, it is necessary to realize velocity modeling through existing data. In the velocity modeling process, the following problems exist:

[0004] (1) Complex data processing: Drilling wave velocity data and resistivity data come from different measurement methods, and the resolution and quality of the data are inconsistent. Complex preprocessing steps are needed to effectively integrate and analyze the data. Traditional data processing methods cannot effectively handle nonlinear and complex data patterns, affecting the accuracy and reliability of velocity modeling.

[0005] (2) Difficulty in model construction: In deep-buried long tunnels, the non-homogeneous and anisotropic characteristics of the geological structure make it difficult for traditional linear models to fully capture the physical properties of the underground medium. Traditional velocity modeling methods are usually based on empirical formulas and simplified assumptions, making it difficult to accurately reflect the complexity of the underground medium.

[0006] (3) Limited prediction accuracy: Traditional velocity modeling methods are prone to model distortion and prediction bias when predicting deep geological structures due to the lack of sufficient low-frequency information. For deep long tunnels under complex geological conditions, traditional models often fail to provide high-precision velocity prediction, affecting the scientificity and safety of engineering decision-making.

[0007] (4) Insufficient multi-source data fusion: Geological investigation of deep long tunnels often involves multiple geophysical data such as borehole wave velocity data, resistivity data, and geological data. Traditional modeling methods lack effective multi-source data fusion techniques, making it difficult to fully utilize the advantages of various data and improve the comprehensive performance of the model.

[0008] In summary, although deep learning models have strong data fitting and feature extraction capabilities, they can automatically learn and extract useful information from large-scale data, providing a new approach to solving the problems existing in traditional velocity modeling methods. However, there are still many technical difficulties in applying deep learning technology to deep long tunnel velocity modeling, such as how to build a deep learning model suitable for deep long tunnel geological conditions, how to effectively integrate and utilize multi-source geological data to improve the accuracy of velocity modeling, and how to optimize the model training process to shorten the calculation time. SUMMARY

[0009] To solve the problems of the prior art, the present application provides a high-precision velocity modeling method and system for deep long tunnels. By introducing a deep learning model and combining borehole wave velocity data and three-dimensional resistivity data and other multi-source geological information, high-precision velocity modeling is achieved, overcoming the limitations of traditional methods and improving the accuracy and reliability of deep long tunnel velocity modeling, providing a new solution for deep long tunnel velocity modeling.

[0010] The present application is achieved by the following technical solutions:

[0011] A high-precision velocity modeling method for deep long tunnels, comprising the following steps:

[0012] Obtain wave velocity data in the borehole of the deep long tunnel and resistivity data at the borehole location;

[0013] Analyze the spatial variability of the wave velocity data and resistivity data, and construct a three-dimensional velocity model and a three-dimensional resistivity model through random simulation to form a training set data;

[0014] Preprocess the three-dimensional velocity model and the three-dimensional resistivity model; the preprocessing includes resolution coordination and normalization processing;

[0015] An initial framework of a state space model is built by defining initial parameters, a deep learning network model is constructed according to the initial framework of the discretized state space model and by using a selective state space model;

[0016] The deep learning network model is trained and verified by randomly selecting training set data and verification set data according to a certain proportion, cross-iteration is used to adjust hyperparameters, and the deep learning network model is iteratively optimized with the goal of minimizing the difference between the predicted speed and the actual speed;

[0017] The resistivity data of the measured area are input into the deep learning network model optimized by training, and a velocity prediction profile of the deep long tunnel is output.

[0018] In the prior art, when facing the complex and variable geological environment of deep long tunnels, the traditional velocity modeling method often appears to be inadequate and cannot accurately depict the velocity distribution inside the tunnel. The geological conditions of deep long tunnels are complex, including stratum velocity, thickness, lithology distribution, and underground water conditions, and many other factors, which pose high requirements on the accuracy of velocity modeling. The present application provides a high-precision velocity modeling method for deep long tunnels, which realizes high-precision prediction of the velocity distribution of deep long tunnels by introducing a selective state space model. Compared with traditional methods, the modeling accuracy is significantly improved, highlighting the shortcomings of traditional methods in high-precision modeling. The present application can automatically adjust model parameters and structures according to different geological characteristics and multi-source data, thereby enhancing the adaptability of the model to different geological conditions. The present application comprehensively utilizes multi-source data such as drilling and resistivity to provide more comprehensive information for the characterization of deep long tunnels. The comprehensive application of such multi-source data reduces the errors and uncertainties that may be caused by a single data source, and improves the reliability and robustness of the velocity model construction. In the velocity modeling process, computational efficiency is a key factor. The present application also establishes a connection between key parameters and input, and adaptively adjusts the key parameters of the model according to the characteristics and importance of the input data, reduces unnecessary calculation steps and redundant information, optimizes the calculation process, and thus improves the real-time performance and efficiency of modeling. This is of great significance for tunnel engineering that requires quick decision-making and response, and contributes to the sustainable development of tunnel engineering.

[0019] In a further aspect, the specific steps of obtaining resistivity data at the drilling location include:

[0020] Three-dimensional resistivity data of the deep long tunnel are obtained by a geophysical inversion method;

[0021] In the three-dimensional resistivity data, the resistivity data at the drilling location are extracted.

[0022] Further, the specific step of preprocessing the three-dimensional velocity model and the three-dimensional resistivity model comprises:

[0023] The three-dimensional resistivity model and the three-dimensional wave velocity model have consistent resolution by sequentially performing coordinate alignment, clipping, resampling and interpolation on the three-dimensional velocity model and the three-dimensional resistivity model.

[0024] Subsequently, the resistivity simulation result and the velocity simulation result are normalized to map them to a specific interval, so as to ensure that the magnitudes of the resistivity and the velocity are consistent.

[0025] Further, the initial parameters include a state matrix A, an input matrix B, an output matrix C, a direct transfer matrix D, a time step and a convolution kernel size, and the initial framework of the state space model is:

[0026] ;

[0027] In the formula, represents the spatial state of the next step; represents the resistivity data; is the wave velocity data; is the input signal.

[0028] Further, the state transition equation of the discretized initial framework of the state space model is:

[0029] ;

[0030] ;

[0031] In the formula, is a discretized state transition matrix; is a discretized input matrix; is a discretized output matrix; y t is a discretized output, t specifically refers to a discretized time point; h(t) represents the hidden state at time t, and h(t-1) represents the hidden state at the previous time.

[0032] Further, the selective state space model adopted is:

[0033] ;

[0034] In the formula, represents a matrix function dynamically generated according to the input x, which allows the model to adjust its behavior according to different characteristics of the input; represents a learnable parameter for estimating the discrete interval of the input sample; is a proportional coefficient; A is a gain parameter; Parameter represents a system parameter.

[0035] Further, after the deep learning network model is constructed, the deep learning network model is iteratively trained using the constructed three-dimensional velocity model and three-dimensional resistivity model, so as to ensure that the deep learning network model learns the nonlinear relationship in the input data.

[0036] Further, the deep learning network model is trained and verified according to the proportion of the randomly selected training set data and verification set data, the hyperparameters are adjusted through cross iteration, the specific steps of iteratively optimizing the deep learning network model include:

[0037] The training set data and the verification set data are randomly selected according to a certain proportion to train the deep learning network model;

[0038] During the forward propagation of the neural network, the output signal of each layer is transmitted to the next layer as an input signal, and the input signal sequentially passes through the network model and the additionally added hidden layer, and finally reaches the output layer, and the nonlinear conversion is completed through the activation function, and the prediction result is output;

[0039] The error between the output value of the output layer and the actual value is calculated, and the hyperparameters of the model are adjusted according to the loss error;

[0040] The above steps are repeatedly executed, and the deep learning network model is optimized for multiple times until the value of the loss function meets the error requirement or the iteration number reaches the maximum value set in advance.

[0041] Further, after the deep learning network model outputs the velocity prediction profile of the deep-buried long tunnel, the velocity prediction profile is verified by using the measured wave velocity data of the drill hole or other independent data sources, and the model is fine-tuned or optimized according to the verification result.

[0042] Further, the application further provides a high-precision velocity modeling system for a deep-buried long tunnel, comprising:

[0043] A data collection module is configured to obtain wave velocity data in a drill hole of a deep-buried long tunnel and resistivity data at a drill hole position.

[0044] A data analysis module is configured to analyze spatial variability of the wave velocity data and the resistivity data, and construct a three-dimensional velocity model and a three-dimensional resistivity model through random simulation to form training set data.

[0045] a data processing module for preprocessing the three-dimensional velocity model and the three-dimensional resistivity model;

[0046] a model construction and processing module for building an initial framework of a state space model by defining initial parameters, constructing a deep learning network model according to the discretized initial framework of the state space model and using a selective state space model, then training and verifying the deep learning network model according to a certain proportion of randomly selected training set data and verification set data, adjusting hyperparameters through cross iteration, and iteratively optimizing the deep learning network model to minimize the difference between the predicted velocity and the actual velocity;

[0047] a velocity prediction module for inputting the resistivity data of the measured area into the trained and optimized deep learning network model to output a velocity prediction profile of the deep long tunnel;

[0048] a result verification and optimization module for verifying the velocity prediction profile with drilling measured wave velocity data or other independent data sources, and fine-tuning or optimizing the model according to the verification result.

[0049] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0050] 1. The deep long tunnel high-precision velocity modeling method and system provided by the present application can significantly improve the modeling accuracy by introducing a selective state space model, effectively solving the problem that traditional velocity modeling methods cannot accurately depict the velocity distribution inside the tunnel in the face of complex and variable geological environment of deep long tunnels.

[0051] 2. The deep long tunnel high-precision velocity modeling method and system provided by the present application can automatically adjust model parameters and structures according to different geological features and multi-source data, thereby realizing flexible response to different geological conditions and improving the universality and practicality of the model.

[0052] 3. The deep long tunnel high-precision velocity modeling method and system provided by the present application can comprehensively utilize drilling and resistivity multi-source data to provide more comprehensive and accurate information for the feature description of deep long tunnels. This method reduces the errors and uncertainties that may be caused by a single data source, improves the reliability and robustness of the velocity model construction, and provides a more solid foundation for engineering decision-making.

[0053] 4. The deep-buried long tunnel high-precision velocity modeling method and system provided by the present application can significantly improve the real-time performance and efficiency of modeling by optimizing the calculation process, establishing a connection between key parameters and input, and adaptively adjusting the key parameters of the model according to the characteristics and importance of the input data, thereby reducing unnecessary calculation steps and redundant information. This is of great significance for tunnel engineering which requires quick response and decision-making.

[0054] 5. The deep-buried long tunnel high-precision velocity modeling method and system provided by the present application can provide more reliable technical support for the planning, design and construction of tunnel engineering by improving the modeling accuracy and calculation efficiency. This not only helps to reduce engineering risks and construction costs, but also helps to improve engineering quality and promote the sustainable development of tunnel engineering. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as limiting the scope. Other related drawings can also be obtained by those skilled in the art without creative labor. In the drawings:

[0056] Figure 1 The flowchart of the deep-buried long tunnel high-precision velocity modeling method provided by the present application;

[0057] Figure 2 The resistivity and wave velocity diagram of the deep-buried long tunnel high-precision velocity modeling system provided by the present application;

[0058] Figure 3 The schematic diagram of the deep-buried long tunnel high-precision velocity modeling system provided by the present application;

[0059] Figure 4 The measured data provided by the present application (a is the measured borehole wave velocity data, and b is the measured resistivity data);

[0060] Figure 5 The three-dimensional wave velocity model provided by the present application;

[0061] Figure 6 The three-dimensional resistivity model provided by the present application;

[0062] Figure 7 The resistivity inversion profile schematic diagram of the input in the prediction stage provided by the present application;

[0063] Figure 8A schematic diagram of a predicted velocity profile output by the present application. DETAILED DESCRIPTION

[0064] In order to make the objectives, technical solutions and advantages of the present application clearer, further detailed description will be given below in combination with embodiments and drawings, and the schematic embodiments of the present application and the description thereof are only used to explain the present application, and do not limit the present application.

[0065] Embodiment 1: The present embodiment 1 provides a high-precision velocity modeling method for deep long tunnel, as shown in the following formula (1), including the following specific steps: Figure 1

[0066] S1: Collecting wave velocity data from the drill hole of the deep long tunnel, and obtaining three-dimensional resistivity data through geophysical inversion method, then extracting the resistivity data at the drill hole position to form a wave velocity data set and a resistivity data set.

[0067] Further, as shown in the following formula (2), the input data is resistivity data and wave velocity data, wherein the resistivity data is obtained by extracting the corresponding value at the drill hole position from the resistivity inversion result, and the input data of the area to be predicted is resistivity data, including but not limited to direct current method, controllable source audio magnetotelluric sounding and natural source electromagnetic sounding method. Wherein the drill hole wave velocity data includes but is not limited to P wave velocity and S wave velocity. Figure 4

[0068] S2: Analyzing the spatial variability of the above-mentioned velocity data and resistivity data, constructing three-dimensional velocity model and three-dimensional resistivity model as shown in the following formula (3) and (4) through random simulation method to form training data, ensuring to meet the requirement of deep learning on training data amount, and ensuring that the features have good consistency with the features of the collected data. Figure 5 Figure 6

[0069] Specifically, the spatial variability analysis includes but is not limited to using variogram, data template, probability distribution and the like to analyze the key structural features of the drill hole wave velocity data and the resistivity data at the corresponding position.

[0070] Specifically, the random simulation method includes but is not limited to obtaining three-dimensional velocity simulation results and three-dimensional resistivity simulation results on the results of the above-mentioned spatial variability analysis through sequential simulation, indicator simulation and simulated annealing simulation and the like, forming training data to meet the requirement of deep learning on training data amount, and ensuring that the features have good consistency with the features of the collected data.

[0071] ​​​​S3: Preprocessing the wave velocity dataset and the resistivity dataset, the preprocessing including resolution coordination and normalization processing, to ensure that the data obtained in different ways have high quality and consistent resolution, forming a more refined and stable modeling framework.

[0072] Further, the preprocessing includes resolution coordination and normalization processing.

[0073] Specifically, through coordinate alignment, clipping, resampling and interpolation, the three-dimensional resistivity model and the three-dimensional wave velocity model have consistent resolution capability, ensuring the accuracy and consistency of the data, so as to better reveal the correlation between the two.

[0074] Further, the normalization includes normalization processing of the resistivity simulation results and the velocity simulation results, which are mapped to a specific interval, ensuring the consistency of the magnitude of resistivity and velocity, and facilitating the rapid acquisition of the characteristics of the two types of data.

[0075] In addition, it also includes a multi-source data fusion step, in which, in addition to wave velocity data and resistivity data, other geological knowledge or engineering data are integrated to further improve the accuracy of velocity prediction.

[0076] S4: Defining the initial parameters and initial model of the model, i.e. the initial framework of the state space model, establishing the deep learning network model structure according to the discretized initial framework of the state space model, the model being capable of receiving wave velocity data and resistivity data as input and outputting velocity prediction values, and through iterative training using the constructed three-dimensional velocity model and three-dimensional resistivity model, ensuring that the model learns the nonlinear relationship in the input data. Wherein, the initial framework of the state space model is selected as follows:

[0077] ;

[0078] In the formula, A is the state matrix, B is the input matrix, C is the output matrix, D is the direct transfer matrix, represents the hidden state, describing the current internal state of the system, represents the next spatial state; represents the resistivity data, affecting the behavior of the system; is the target parameter, i.e. wave velocity data; is the input signal.

[0079] Discretize the initial architecture of deep learning, so that deep learning can effectively process input data from drilling or resistivity, and the discretized state transition equation is:

[0080] ;

[0081] ;

[0082] wherein, is a discrete state transition matrix; is a discrete input matrix; is a discrete output matrix; y t denotes the discretized output, t denotes the discretized time point; h(t) represents the hidden state at time t, and h(t-1) represents the hidden state at the previous time.

[0083] The selective state space model is adopted, so that key model parameters can be dynamically adjusted according to the characteristics of input data, the calculation difficulty is reduced, and the calculation efficiency is improved.

[0084] ;

[0085] wherein, represents a matrix function dynamically generated according to the input x, and allows the model to adjust its behavior according to different characteristics of the input; represents a learnable parameter for estimating the discrete interval of the input sample; is a proportional coefficient; A is a gain parameter; Parameter represents a system parameter.

[0086] S5: training and verifying the model according to the training set data and the validation set data selected randomly in a certain proportion, adjusting the state size and the sequence length and other hyperparameters through cross iteration, taking minimizing the difference between the predicted speed and the actual speed as the goal, and constantly optimizing the model until the loss function meets the pre-set error range or reaches the maximum number of iterations.

[0087] Specifically, the model training method in step S6 specifically includes the following contents: considering that different input attributes have different contribution rates to the prediction target, inputting the wave velocity and resistivity and other data into the network model to obtain a prediction result, calculating the error between the output value of the output layer and the actual value, adjusting the hyperparameters of the model such as the convolution kernel size, the learning rate, the state size and the sequence length according to the loss error. Repeat the above steps to optimize the deep learning network model multiple times until the value of the loss function meets the error requirement or the number of iterations reaches the maximum value set in advance.

[0088] Specifically, the training set data and the validation set data are randomly selected in a proportion of 8:2 for model training. During the forward propagation process of the neural network, the output signals of each layer are transmitted to the next layer as input signals. These input signals pass through the network model and the additional hidden layer in turn, and finally reach the output layer. Through the activation function, the nonlinear conversion is completed, and the prediction result is output.

[0089] Specifically, in one embodiment, the network model architecture is composed of three network blocks, each containing a convolutional layer, a linear projection layer, and a selective state space module. In this model, the convolutional unit is composed of a convolutional layer containing 32 3x3 convolutional kernels with a stride of 1 and using a SiLU activation function. In the selective state space module, the state number is set to 128 and the sequence length is set to 80.

[0090] S6: as shown in Figure 7 and Figure 8 , the trained deep learning network model is applied to the resistivity data of the test area to calculate and output the velocity prediction profile of the deep long tunnel.

[0091] S7: optionally, the velocity prediction profile is verified using borehole measured wave velocity data or other independent data sources, and the model is fine-tuned or optimized based on the verification results.

[0092] The above scheme provides a high-precision velocity modeling method for deep long tunnels. By introducing a selective state space model, high-precision prediction of the velocity distribution of deep long tunnels is achieved. Compared with traditional methods, the modeling accuracy is significantly improved, highlighting the shortcomings of traditional methods in high-precision modeling. The invention can automatically adjust model parameters and structures according to different geological features and multi-source data, thereby enhancing the adaptability of the model to different geological conditions. The invention comprehensively utilizes multi-source data such as boreholes and resistivity to provide more comprehensive information for the characterization of deep long tunnels. The comprehensive application of multi-source data reduces the errors and uncertainties that may be caused by a single data source, improving the reliability and robustness of the velocity model construction. In the velocity modeling process, computational efficiency is a key factor. The invention also establishes a connection between key parameters and inputs, and adjusts the key parameters of the model according to the characteristics and importance of the input data, reducing unnecessary calculation steps and redundant information, optimizing the calculation process, and improving the real-time performance and efficiency of the modeling. This is of great significance for tunnel projects that require quick decision-making and response, and contributes to the sustainable development of tunnel projects.

[0093] Embodiment 2: This embodiment 2 is further optimized based on embodiment 1, as shown in Figure 2 and Figure 3 , a high-precision velocity modeling system for deep long tunnels is provided, comprising:

[0094] A data collection module is used to obtain wave velocity data in a deep long tunnel borehole and resistivity data at the borehole location.

[0095] a data analysis module, configured to analyze spatial variability of the wave velocity data and the resistivity data, and construct a three-dimensional velocity model and a three-dimensional resistivity model by means of random simulation to form training set data;

[0096] a data processing module, configured to pre-process the three-dimensional velocity model and the three-dimensional resistivity model;

[0097] a model construction and processing module, configured to build an initial framework of a state space model by defining initial parameters, construct a deep learning network model according to the discretized initial framework of the state space model and the selected state space model, and then train and verify the deep learning network model by randomly selecting training set data and verification set data at a certain proportion, and iteratively optimize the deep learning network model by adjusting hyperparameters to minimize the difference between the predicted velocity and the actual velocity;

[0098] a velocity prediction module, configured to input the measured resistivity data of a region into the trained and optimized deep learning network model to output a velocity prediction profile of the deep-buried long tunnel;

[0099] a result verification and optimization module, configured to verify the velocity prediction profile by drilling measured wave velocity data or other independent data sources, and fine-tune or optimize the model according to the verification result.

[0100] In some example embodiments, the embodiment also provides a deep-buried long tunnel high-precision velocity modeling terminal device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the minimum technical solution of a deep-buried long tunnel high-precision velocity modeling method for achieving the purpose of "combining drilling wave velocity data and three-dimensional resistivity data and other multi-source geological information to achieve high-precision velocity modeling, overcoming the limitations of traditional methods, and improving the precision and reliability of deep-buried long tunnel velocity modeling" as described in Embodiment 1 when executing the computer program.

[0101] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0102] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0103] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device that implements the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0105] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned facts and methods can be completed by a program instructing relevant hardware, and the program involved or the program can be stored in a computer-readable storage medium. When the program is executed, the steps include the corresponding method steps, and the storage medium can be a ROM / RAM, a magnetic disk, an optical disk, and the like.

[0106] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A high-precision velocity modeling method for deep long tunnels, characterized in that, The method comprises the following steps: obtaining wave velocity data in a deep long tunnel drilling hole and resistivity data at a drilling hole position; analyzing spatial variability of the wave velocity data and the resistivity data, and constructing a three-dimensional velocity model and a three-dimensional resistivity model by means of random simulation to form training set data; preprocessing the three-dimensional velocity model and the three-dimensional resistivity model; the preprocessing comprises resolution coordination and normalization processing; building an initial framework of a state space model by defining initial parameters, and constructing a deep learning network model according to the discretized initial framework of the state space model and by using a selective state space model; training and verifying the deep learning network model by randomly selecting training set data and verification set data at a certain ratio, adjusting hyperparameters through cross iteration, and iteratively optimizing the deep learning network model with the aim of minimizing the difference between predicted velocity and actual velocity; inputting measured resistivity data of a region into the deep learning network model after training and optimization to output a velocity prediction profile of the deep long tunnel.

2. The method according to claim 1, wherein, The specific steps for obtaining resistivity data at a drilling hole position comprise: obtaining three-dimensional resistivity data of the deep long tunnel by a geophysical inversion method; extracting resistivity data at the drilling hole position from the three-dimensional resistivity data.

3. The method of claim 1, wherein, The specific steps for preprocessing the three-dimensional velocity model and the three-dimensional resistivity model comprise: aligning coordinates, clipping, resampling and interpolating the three-dimensional velocity model and the three-dimensional resistivity model in sequence to make the three-dimensional resistivity model and the three-dimensional wave velocity model have consistent resolution capabilities; then, performing normalization processing on the resistivity simulation results and the velocity simulation results to map them to a specific interval, and ensuring that the magnitudes of resistivity and velocity are consistent.

4. The method of claim 1, wherein, The initial parameters comprise a state matrix A, an input matrix B, an output matrix C, a direct transfer matrix D, a time step and a convolution kernel size, and the initial framework of the state space model is: ； wherein represents the spatial state of the next step; represents resistivity data; is wave velocity data; is the input signal.

5. The method of claim 4, wherein, the state transition equation of the discretized initial framework of the state space model is: ； ； wherein is a discrete state transition matrix; is a discrete input matrix; is a discrete output matrix; y t denotes the discretized output, t specifically refers to the discretized time point; h(t) represents the hidden state at time t, and h(t-1) represents the hidden state at the previous time.

6. The method of claim 5, wherein, the selective state space model used is: ； wherein: represents a matrix function dynamically generated according to the input x, allowing the model to adjust its behavior according to different features of the input; where denotes a learnable parameter for estimating the discrete interval of the input sample; is a proportional coefficient; A is a gain parameter; Parameter: a system parameter.

7. The method of claim 1, wherein, After constructing the deep learning network model, the three-dimensional velocity model and the three-dimensional resistivity model are used to iteratively train the deep learning network model to ensure that the deep learning network model learns the nonlinear relationship in the input data.

8. The method of claim 1, wherein, The specific steps for training and verifying the deep learning network model by randomly selecting training set data and verification set data at a certain ratio, adjusting hyperparameters through cross iteration, and iteratively optimizing the deep learning network model with the aim of minimizing the difference between predicted velocity and actual velocity comprise: training the deep learning network model by randomly selecting training set data and verification set data at a certain ratio; in the forward propagation process of the neural network, the output signals of each layer are transmitted to the next layer as input signals, and these input signals pass through the network model and additional hidden layers in sequence, finally reach the output layer, complete nonlinear conversion through an activation function, and output a prediction result; An error between an output value of the output layer and an actual value is calculated, and a hyperparameter of the model is adjusted according to a loss error; The above steps are repeatedly performed to optimize the deep learning network model multiple times until a value of the loss function meets an error requirement or an iteration number reaches a maximum value set in advance.

9. The method of claim 1, wherein, After the deep learning network model outputs the velocity prediction profile of the deep long tunnel, the velocity prediction profile is verified by using the measured wave velocity data of the drill hole or other independent data sources, and the model is fine-tuned or optimized according to the verification result.

10. A high-precision velocity modeling system for deep long tunnels, characterized in that, Comprise: A data collection module is configured to acquire wave velocity data in a drill hole of a deep long tunnel and resistivity data at a position of the drill hole; A data analysis module is configured to analyze spatial variability of the wave velocity data and the resistivity data, and construct a three-dimensional velocity model and a three-dimensional resistivity model by means of random simulation to form training set data; A data processing module is configured to preprocess the three-dimensional velocity model and the three-dimensional resistivity model; A model construction and processing module is configured to build an initial framework of a state space model by defining initial parameters, construct a deep learning network model according to the discretized initial framework of the state space model and the selective state space model, then randomly select training set data and verification set data at a certain proportion to train and verify the deep learning network model, adjust a hyperparameter by cross iteration to minimize a difference between a predicted velocity and an actual velocity, and iteratively optimize the deep learning network model; A velocity prediction module is configured to input measured resistivity data of a region into the trained and optimized deep learning network model to output a velocity prediction profile of the deep long tunnel; A result verification and optimization module is configured to verify the velocity prediction profile by using measured wave velocity data of a drill hole or other independent data sources, and fine-tune or optimize the model according to a verification result.

Images