A tunnel karst development degree intelligent evaluation method, device and medium

Abstract

The present application relates to the technical field of geological engineering, and specifically relates to a tunnel karst development degree intelligent evaluation method, equipment and medium, comprising obtaining a multi-source fusion data set; generating a karst development fractal evaluation comprehensive index; determining a tunnel karst key detection area; constructing a three-dimensional medium distribution model of the karst area; classifying the risk grade of the karst area, and establishing a geological risk area prediction model; the present application significantly improves the comprehensiveness and accuracy of karst development characteristics by combining multi-source data and fractal theory; the detection path is dynamically optimized using a reinforcement learning model, reducing the detection blind area and improving the detection efficiency of the key area; the data is denoised and fused by a generative adversarial network, enhancing the accuracy of model construction; based on the three-dimensional medium distribution model and the risk prediction model, the risk of the karst area is effectively evaluated and dynamically predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological engineering, in particular to a tunnel karst development degree intelligent evaluation method, device and medium. BACKGROUND

[0002] In tunnel engineering construction, the complexity of karst geological conditions poses a major challenge to engineering safety and stability. Accurate evaluation of karst development degree is a key link in tunnel design and construction, directly affecting risk control and cost management of the project. Traditional karst evaluation methods mainly rely on geological exploration data and analysis of water chemical parameters, but these methods usually have the following limitations:

[0003] Traditional methods rely on single type of geological data (such as electromagnetic field data or elastic wave velocity data), lack of multi-source data fusion analysis, resulting in insufficient comprehensiveness and accuracy of evaluation results. Existing technologies rely on manual intervention and simple statistical analysis for processing of geological data and water chemical parameters, which is difficult to deal with large-scale, high-dimensional data, and is easily disturbed by noise and outliers. Existing methods rely on empirical formulas or simple statistical models for risk prediction in karst areas, lack of deep learning and dynamic modeling ability for complex geological conditions, resulting in insufficient prediction accuracy.

[0004] In recent years, with the rapid development of artificial intelligence technology, advanced algorithms such as reinforcement learning and generative adversarial networks have been preliminarily applied in geological exploration and engineering risk assessment. However, these technologies have not fully combined multi-source data fusion and fractal theory, making it difficult to achieve intelligent and dynamic evaluation of karst development degree. Therefore, there is an urgent need for a technical solution that can integrate multi-source data, dynamically optimize evaluation models and achieve high-precision risk prediction to address the challenges of complex karst geological conditions in tunnel engineering. SUMMARY

[0005] The present application solves the problems of single data source, low processing efficiency, static evaluation model, limited risk prediction ability and insufficient three-dimensional modeling technology in traditional karst development degree evaluation methods, and aims to provide a tunnel karst development degree intelligent evaluation method, device and medium, which realizes dynamic evaluation and risk prediction of karst development degree based on multi-source data fusion, fractal theory, reinforcement learning and generative adversarial networks.

[0006] The present application is achieved by the following technical solutions:

[0007] A tunnel karst development degree intelligent evaluation method, comprising:

[0008] Collecting multi-source geological data and water chemical parameters of the karst area, and preprocessing the multi-source geological data and water chemical parameters to obtain a multi-source fusion data set;

[0009] The geological fractal dimension and hydrochemical fractal dimension were calculated using a multi-source fusion dataset, and a comprehensive index for evaluating the fractal characteristics of karst development was generated.

[0010] The comprehensive index of karst development fractal evaluation is input into the reinforcement learning model, and the key detection areas of tunnel karst are dynamically determined using the deep deterministic strategy gradient algorithm.

[0011] Data was acquired from key detection areas, and generative adversarial networks were used for noise reduction and high-dimensional fusion processing to construct a three-dimensional medium distribution model of the karst region.

[0012] Based on the three-dimensional medium distribution model and the comprehensive index of karst development fractal evaluation, the risk level of karst areas is classified, and a geological risk area prediction model is established.

[0013] Specifically, the multi-source geological data includes electromagnetic field data, elastic wave velocity field data, and current field data, while the hydrochemical parameters include pH value, mineralization degree, and concentration of major ions.

[0014] Methods for preprocessing data include:

[0015] The collected multi-source geological data were verified to remove outliers and incomplete data, and noise was removed through filtering and wavelet transform. The boundary features were refined using a gradient enhancement algorithm.

[0016] The integrity of water chemical parameters was checked and outliers were removed. After filling in the missing data using interpolation, the data was normalized.

[0017] Spatial coordinate alignment is performed on multi-source geological data and hydrochemical parameters, and the hydrochemical parameters are mapped to the spatial grid of multi-source geological data to form a unified multi-source fusion dataset.

[0018] Optionally, methods for generating a comprehensive index for fractal evaluation of karst development include:

[0019] By utilizing multi-source geological data and hydrochemical parameters from a multi-source fusion dataset, a multidimensional feature matrix of karst development areas is established. ,in, Geological spatial coordinates For the time of data collection;

[0020] The multidimensional feature matrix is ​​divided into multiple equally spaced grids according to the geological data dimensions, and the geological fractal dimension of the multi-source geological data is calculated using the box counting method based on the gridding results. , ,in, The number of boxes in the grid that contain the feature regions. The scale of the grid cell;

[0021] Extract time series of water chemical parameters from a multidimensional feature matrix and map them to corresponding spatial coordinates;

[0022] For each time series, calculate the cumulative bias. and standard deviation And the Hearst exponent was obtained by fitting. The hydrochemical fractal dimension of the hydrochemical parameters was calculated. , ;

[0023] Comprehensive index for fractal evaluation of karst development , ,in, and The weighting coefficients were determined based on historical data.

[0024] Specifically, the methods for determining key areas for karst exploration in tunnels include:

[0025] Define the state space, action space, and reward function for reinforcement learning;

[0026] Policy network for building reinforcement learning models And determine its optimization objective. ,in, For the parameters of the policy network, The state distribution generated for the policy network, Let the expected value be the state distribution based on the current policy. State;

[0027] Building a value network for reinforcement learning models And determine its optimization objective. ,in, As a discount factor, For the parameters of the value network, For the quaternions in the experience replay pool Expected value For instant rewards, For action;

[0028] Construct a sample quadruple The experience replay pool, in which For the new state;

[0029] Samples are randomly drawn in batches from the experience replay pool, and then the parameters are optimized and updated using the objective function. ,in, This is the soft update coefficient. For the parameters of the target policy network, The parameters of the target value network;

[0030] The current state is determined by the updated policy network. Optimal action , This refers to the displacement increment of the detection device in each direction;

[0031] Update the location of the detection equipment Add the fractal evaluation comprehensive index corresponding to the new position to the state. The process is repeated until the detection stopping condition is met.

[0032] All historical locations of the detection equipment constitute the key detection areas for tunnel karst.

[0033] Specifically, the state space of reinforcement learning is defined as the fractal evaluation comprehensive index distribution of the tunnel karst region and the current location of the detection equipment, and the state of the detection equipment. ,in, Spatial coordinates The comprehensive index for evaluating the fractal characteristics of karst development. This indicates the current spatial location of the detection equipment;

[0034] Define the motion space as the direction and step size of the movement of the detection device, and the motion... ,in, This refers to the displacement increment of the detection device in each direction;

[0035] Determine the reward function for reinforcement learning ,in, To emphasize the weighting coefficients of the target region, The weighting factor is the penalty path length. This represents the path length of the current action.

[0036] Specifically, methods for constructing a three-dimensional media distribution model of karst regions include:

[0037] The multi-source fusion dataset corresponding to each detection point in the intermediate detection area of ​​the tunnel karst is obtained and mapped onto a three-dimensional spatial grid to form an initial three-dimensional feature matrix. ;

[0038] Construct a generative adversarial network model and determine its adversarial loss function. ,in, For the discriminator to the feature matrix The judgment result, For the discriminator to the generator The discrimination result of the output feature matrix. The expectation of the feature matrix sampled from the real data distribution. For the true data distribution, The data distribution generated by the generator;

[0039] Determine the reconstruction error of the generator Generator loss function Discriminator loss function ; These are the weighting coefficients for the generator reconstruction error. It is an L2 norm;

[0040] The discriminator and generator are trained iteratively, and the loss function is optimized until the generator loss function and the discriminator loss function converge.

[0041] Update discriminator parameters Update generator parameters ;in, The learning rate of the discriminator. The learning rate of the generator;

[0042] The initial 3D feature matrix is ​​input into the generator and the denoised feature matrix is ​​output. And perform multi-dimensional feature fusion to obtain a three-dimensional feature matrix. ;

[0043] A three-dimensional medium distribution model of the karst region was generated using volume rendering technology. ,in, This is a volume rendering algorithm function that maps the feature matrix to a 3D model.

[0044] Specifically, the methods for classifying the risk levels of karst areas include:

[0045] Constructing a risk input matrix ,in, A three-dimensional medium distribution model. It serves as a comprehensive index for evaluating the fractal characteristics of karst development.

[0046] Extracting local medium gradients Local rate of change of fractal composite index ;

[0047] Determine spatial points risk factors ,in, and These are weighting coefficients determined based on historical data;

[0048] Determine multiple risk level thresholds for risk factors ;

[0049] Construct a deep residual network for dynamic classification of risk factors and determine the loss function. ,in, The number of spatial points, The number of risk levels, For the first The true risk level of each spatial point For classification networks to the first Predicted risk level for each spatial point;

[0050] Using the risk factor matrix as input and the true risk level label as output, a deep residual network is trained until the loss function converges to obtain a risk level classification model.

[0051] The risk factors of spatial points are input into the risk level classification model to calculate the probability of each point belonging to each risk level. The maximum probability criterion is then used to classify the risk level of each spatial point to obtain the final risk level. .

[0052] Optionally, methods for establishing geological risk area prediction models include:

[0053] By integrating real-time monitoring data and historical risk distribution data, a dynamic risk assessment system is established. ,in, Current time Risk level distribution The final risk level output by the risk level classification model. Historical risk level distribution Weighting for real-time data;

[0054] The risk level distribution across multiple time steps is represented as a time series. ,in, The time window length is specified; and the risk level and volatility of the time series are obtained.

[0055] The time series data, final risk level, assessed risk level, and risk volatility are used as inputs to a spatiotemporal convolutional network model to train a geological risk area prediction model. ;

[0056] The final predicted risk level is obtained by revising the predicted data according to the final risk level. ,in, Geological risk area prediction model Predicted risk level This is a correction factor.

[0057] A smart evaluation device for the degree of karst development in tunnels includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a smart evaluation method for the degree of karst development in tunnels as described above.

[0058] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the intelligent evaluation method for the degree of karst development in tunnels as described above.

[0059] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0060] This invention collects and fuses electromagnetic field data, elastic wave velocity field data, current field data, and hydrochemical parameters to form a unified multi-source fusion dataset. Based on this dataset, it calculates the geological fractal dimension and hydrochemical fractal dimension to generate a comprehensive fractal evaluation index for karst development. A reinforcement learning model is used to identify key karst detection areas in tunnels. A generative adversarial network is then used to denoise and perform high-dimensional fusion processing on the data from these key areas, constructing a high-precision three-dimensional media distribution model. Finally, based on the three-dimensional media distribution model and the comprehensive fractal evaluation index, combined with a deep residual network and a spatiotemporal convolutional network, it achieves risk level classification and dynamic prediction for karst areas.

[0061] This invention significantly improves the comprehensiveness and accuracy of karst development characteristics by combining multi-source data with fractal theory; it uses a reinforcement learning model to dynamically optimize the detection path, reducing blind spots and improving detection efficiency in key areas; it enhances the accuracy of model construction by using generative adversarial networks to denoise and fuse data; and it effectively realizes the hierarchical assessment and dynamic prediction of risks in karst areas based on a three-dimensional medium distribution model and a risk prediction model. Attached Figure Description

[0062] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, but do not constitute a limitation on the embodiments of the present invention.

[0063] Figure 1 This is a flowchart illustrating an intelligent evaluation method for the degree of karst development in tunnels according to the present invention.

[0064] Figure 2 This is a schematic flowchart of the method for generating a comprehensive index for fractal evaluation of karst development according to the present invention.

[0065] Figure 3 This is a flowchart illustrating the method for constructing a three-dimensional media distribution model of a karst region according to the present invention.

[0066] Figure 4 This is a flowchart illustrating the method for classifying the risk level of karst areas according to the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0068] It should also be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.

[0069] Where there is no conflict, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0070] Example 1

[0071] like Figure 1 As shown, an intelligent evaluation method for the degree of karst development in tunnels is provided, mainly including five steps: data acquisition and processing, characteristic index calculation, key detection area determination, three-dimensional model construction, and risk level classification and prediction. The following is a breakdown and explanation of the specific technical content:

[0072] S1. Collect multi-source geological data and hydrochemical parameters of karst areas, and preprocess the multi-source geological data and hydrochemical parameters to obtain a multi-source fusion dataset;

[0073] Multi-source geological data (such as electromagnetic field, elastic wave velocity field, and current field data) and hydrochemical parameters (such as pH value, mineralization, and major ion concentration) of karst areas were collected. Subsequently, outliers were removed and noise was reduced through verification, filtering, and wavelet transform. Interpolation was used to fill in missing data, and the geological data and hydrochemical parameters were spatially aligned to form a unified multi-source fusion dataset.

[0074] S2. Calculate the geological fractal dimension and hydrochemical fractal dimension using multi-source fusion datasets, and generate a comprehensive index for evaluating karst development fractals.

[0075] Based on a multi-source fusion dataset, the geological fractal dimension and the hydrochemical fractal dimension were calculated separately. The geological fractal dimension was calculated using grid partitioning and box counting; the hydrochemical fractal dimension was derived from the Hearst exponent obtained through time series analysis. Finally, using weighting coefficients, the two were fused to generate a comprehensive index for evaluating the fractal characteristics of karst development, serving as the basis for subsequent analyses.

[0076] S3. Input the comprehensive index of karst development fractal evaluation into the reinforcement learning model, and use the deep deterministic strategy gradient algorithm to dynamically determine the key detection areas of tunnel karst.

[0077] Using the fractal evaluation comprehensive index as input, a reinforcement learning model is constructed through a deep deterministic policy gradient algorithm. The model optimizes the detection path through a policy network and a value network, dynamically determining key detection areas in tunnel karst, significantly improving detection efficiency and reducing blind spots.

[0078] S4. Acquire data within the key detection area and use generative adversarial networks for noise reduction and high-dimensional fusion processing to construct a three-dimensional medium distribution model of the karst area;

[0079] New data was acquired in key exploration areas, and generative adversarial networks (GANs) were used for noise reduction and high-dimensional fusion processing. The generator produced an optimized feature matrix, the discriminator ensured data quality, and finally, volume rendering techniques were used to construct a three-dimensional medium distribution model of the karst region, visually demonstrating the characteristics of karst development.

[0080] S5. Based on the three-dimensional medium distribution model and the comprehensive index of karst development fractal evaluation, classify the risk level of karst areas and establish a geological risk area prediction model.

[0081] By combining a three-dimensional medium distribution model and a fractal evaluation comprehensive index, local risk factors are calculated, and a deep residual network classification model is used to dynamically classify risk levels. Furthermore, a spatiotemporal convolutional network model is employed to predict the future development trend of geological risk areas, achieving dynamic prediction of the spatiotemporal distribution of risk.

[0082] In summary, the method in this embodiment comprehensively characterizes karst development features through multi-source data fusion and fractal index analysis; it improves the identification efficiency of key areas by optimizing the detection path with reinforcement learning; it generates a high-precision medium distribution model by combining generative adversarial networks and 3D modeling technology; and finally, it realizes intelligent and dynamic risk assessment of karst areas through risk classification and prediction models.

[0083] Example 2

[0084] Multi-source geological data includes electromagnetic field data, elastic wave velocity field data, and current field data. Electromagnetic field data is obtained by detecting the electrical distribution of strata using geoelectromagnetism, which is used to identify differences in conductivity among different media. Elastic wave velocity field data of strata is obtained through seismic wave measurements, reflecting the mechanical properties of the medium in karst areas. Current field distribution characteristics are used to identify current field data in karst regions, aiding in the analysis of subsurface medium distribution.

[0085] Hydrochemical parameters include pH value, mineralization, and concentration of major ions. pH value is used to determine the acidity or alkalinity of a water body and is related to the dissolution reactions during karst development. Mineralization reflects the total content of dissolved solids in the water body and can indirectly assess the intensity of karst activity. The concentration of major ions directly reflects the chemical characteristics of the karst dissolution process, such as calcium ions and carbonate ions.

[0086] Methods for preprocessing data include:

[0087] The collected multi-source geological data were verified, and outliers and incomplete data were identified and removed using statistical methods. Filtering and wavelet transform were then used to denoise the multi-source geological data while preserving the main features of the original data. A gradient enhancement algorithm was employed to highlight the boundary features of medium changes in the geological data, refining these boundary features.

[0088] The integrity of water chemical parameters is checked and outliers are removed by outlier detection. After filling in the missing data using interpolation methods (such as Kriging interpolation or inverse distance weighting), the data is normalized to map each parameter value to the same dimension.

[0089] Spatial coordinate alignment is performed on multi-source geological data and hydrochemical parameters. Coordinate transformation techniques (such as affine transformation or interpolation transformation) are used to match hydrochemical parameters with geological grids, mapping hydrochemical parameters to the spatial grids of multi-source geological data to form a unified multi-source fusion dataset.

[0090] Example 3

[0091] like Figure 2 As shown, this embodiment provides a method for generating a comprehensive index for fractal evaluation of karst development through multi-source fusion datasets, including:

[0092] By utilizing multi-source geological data and hydrochemical parameters from a multi-source fusion dataset, a multidimensional feature matrix of karst development areas is established. ,in, Geological spatial coordinates For the time of data collection;

[0093] The multidimensional feature matrix is ​​divided into multiple equally spaced grids according to the geological data dimensions, and the geological fractal dimension of the multi-source geological data is calculated using the box counting method based on the gridding results. , ,in, The number of boxes in the grid that contain the feature regions. The scale of the grid cells; the number of boxes at a small scale can reflect the complexity of geological features, and the fractal dimension describes the spatial irregularity and distribution complexity of karst development areas.

[0094] Extract time series of water chemical parameters (such as pH value, mineralization, and ion concentration) from a multidimensional feature matrix and map them to the corresponding spatial coordinates;

[0095] For each time series, calculate the cumulative bias. and standard deviation And the Hearst exponent was obtained by fitting. The hydrochemical fractal dimension of the hydrochemical parameters was calculated. , ;

[0096] Cumulative deviation represents the cumulative trend of change over time. Standard deviation represents the degree of fluctuation over time. Fractal dimension describes the variation and complexity of water chemical parameters over time; the closer the value is to 2, the more irregular the water chemical process.

[0097] Comprehensive index for fractal evaluation of karst development , ,in, and The weighting coefficients are determined based on historical data. The comprehensive index for fractal evaluation of karst development serves as a comprehensive evaluation indicator of the degree of karst development, quantitatively describing the spatial complexity and temporal evolution characteristics of karst development within a region.

[0098] Example 4

[0099] This embodiment provides a method for dynamically determining key detection areas in tunnel karst based on a reinforcement learning model. Specifically, by constructing a state space, action space, reward function, policy network, and value network using reinforcement learning, the optimal path for the detection equipment is determined, and the detection area is optimized. The method for determining key detection areas in tunnel karst includes:

[0100] Define the state space, action space, and reward function for reinforcement learning; the state space represents the reinforcement learning model's description of the detection environment, the action space describes the displacement increment of the detection device in each direction, and the reward function guides the device to select the optimal path.

[0101] The state space of reinforcement learning is defined as the fractal evaluation comprehensive index distribution of the tunnel karst region and the current location of the detection equipment, and the state of the detection equipment. ,in, Spatial coordinates The comprehensive index for evaluating the fractal characteristics of karst development. This indicates the current spatial location of the detection equipment;

[0102] Define the motion space as the direction and step size of the movement of the detection device, and the motion... ,in, This refers to the displacement increment of the detection device in each direction;

[0103] Determine the reward function for reinforcement learning ,in, To emphasize the weighting coefficients of the target region, The weighting factor is the penalty path length. This represents the path length of the current action.

[0104] Weighting coefficient Prioritizing key exploration areas, regions with higher fractal evaluation comprehensive indices are assigned higher weights. Weighting coefficients. By penalizing the path length, the device's movement path is prevented from being too long, thus improving detection efficiency.

[0105] Policy network for building reinforcement learning models And determine its optimization objective. ,in, For the parameters of the policy network, The state distribution generated for the policy network, Let the expected value be the state distribution based on the current policy. The policy network generation device is in the current state. The following action The goal is to optimize network parameters. Maximize long-term cumulative rewards.

[0106] Building a value network for reinforcement learning models And determine its optimization objective. ,in, As a discount factor, For the parameters of the value network, For the quaternions in the experience replay pool Expected value For instant rewards, For action; For value network to state and actions Value estimation. Value networks are used to evaluate the contribution of specific states and actions to long-term cumulative rewards, with the goal of optimizing network parameters to minimize its loss function.

[0107] Construct a sample quadruple The experience replay pool, in which This is the new state; the experience replay pool is used to store samples generated during the exploration process, and parameters are optimized by randomly selecting batches of samples from it, thereby improving model stability and training efficiency.

[0108] Samples are randomly drawn in batches from the experience replay pool, and then the parameters are optimized and updated using the objective function. ,in, This is the soft update coefficient. For the parameters of the target policy network, The parameters of the target value network;

[0109] The current state is determined by the updated policy network. Optimal action , This refers to the displacement increment of the detection device in each direction;

[0110] According to the action Update the location of the detection equipment. Add the fractal evaluation comprehensive index corresponding to the new position to the state. The process is repeated until the detection stops (e.g., the area coverage reaches a threshold or the total reward no longer increases significantly).

[0111] All historical locations of the detection equipment constitute the key detection areas for tunnel karst. Through reinforcement learning models, the equipment can prioritize the detection of areas with higher fractal evaluation comprehensive indices, dynamically optimize the detection path, and improve efficiency and accuracy.

[0112] This embodiment defines the state space, action space, and reward function using a reinforcement learning model, and guides the detection equipment to dynamically optimize its detection path using a policy network and a value network. During the iterative process, the equipment updates its policy based on immediate rewards, gradually covering key karst areas. Finally, by using the equipment's historical detection locations, key karst detection areas for the tunnel are generated, providing a scientific basis for subsequent risk assessment and tunnel construction.

[0113] Example 5

[0114] This embodiment provides specific steps for constructing a three-dimensional media distribution model of a karst region using Generative Adversarial Network (GAN) technology. From mapping multi-source data to generating the final three-dimensional model, each step is based on GAN training and feature matrix fusion processing to form a high-precision three-dimensional model.

[0115] like Figure 3 As shown, the method for constructing a three-dimensional medium distribution model of karst regions includes:

[0116] The multi-source fusion dataset corresponding to each detection point in the intermediate detection area of ​​the tunnel karst is obtained and mapped onto a three-dimensional spatial grid to form an initial three-dimensional feature matrix. The matrix is ​​based on a three-dimensional grid and represents the numerical distribution of karst features in each grid cell.

[0117] A Generative Adversarial Network (GAN) model is constructed, consisting of two parts: a generator and a discriminator. These two components are collaboratively optimized through adversarial training to generate a feature matrix that approximates the true distribution.

[0118] The discriminator is used to distinguish whether the input feature matrix is ​​real data or fake data generated by the generator. The generator tries to deceive the discriminator by generating fake data, making it unable to correctly determine the source of the data.

[0119] Determine its adversarial loss function ,in, For the discriminator to the feature matrix The judgment result, For the discriminator to the generator The discrimination result of the output feature matrix. The expectation of the feature matrix sampled from the real data distribution. For the true data distribution, The data distribution generated by the generator;

[0120] The generator must not only generate a realistic feature matrix, but also ensure that its reconstruction error relative to the true matrix is ​​as small as possible. Determining the generator's reconstruction error... Generator loss function Discriminator loss function ; These are the weighting coefficients for the generator reconstruction error. It is an L2 norm;

[0121] Adversarial training of the generative adversarial network (GAN) is achieved by optimizing the loss function of the generator and discriminator parameters respectively: updating the discriminator parameters. Update generator parameters ;in, The learning rate of the discriminator. The learning rate of the generator;

[0122] Through multiple iterations of training, the loss functions of the generator and discriminator converge, meaning that the feature matrix generated by the generator is difficult to distinguish from the true feature matrix.

[0123] The initial 3D feature matrix is ​​input into the generator and the denoised feature matrix is ​​output. And perform multi-dimensional feature fusion to obtain a three-dimensional feature matrix. ;

[0124] A three-dimensional medium distribution model of the karst region was generated using volume rendering technology. ,in, This is a volume rendering algorithm function that maps the feature matrix to a 3D model.

[0125] A volume rendering algorithm is a technique for directly converting three-dimensional volume data into two-dimensional images. It comprehensively processes each voxel (the basic unit in three-dimensional data, similar to a pixel in a two-dimensional image) in the three-dimensional volume data, taking into account the optical properties of the voxel such as color and transparency, and simulates the process of light passing through the volume data field. Finally, it generates a projected image on a two-dimensional image plane, thereby realizing the visualization of three-dimensional data.

[0126] Example 6

[0127] This embodiment constructs a risk level classification method for karst areas in tunnels by calculating the change rate of medium gradient and fractal comprehensive index, combined with the classification of risk factors using a depth residual network. Figure 4 As shown, the methods for classifying the risk level of karst areas include:

[0128] Constructing a risk input matrix ,in, A three-dimensional medium distribution model. The comprehensive index for fractal evaluation of karst development; the three-dimensional medium distribution model represents the spatial distribution characteristics of the medium in the karst region; the comprehensive index for fractal evaluation of karst development reflects the spatial distribution characteristics of the degree of karst development.

[0129] Extracting local medium gradients Local rate of change of fractal composite index The gradient represents the rate of change in different directions, reflecting the spatial variation of the local medium distribution. The local rate of change reflects the degree of spatial variation of the fractal index of karst development.

[0130] Determine spatial points based on the rate of change of local gradients and fractal composite indices. risk factors ,in, and These are weighting coefficients determined based on historical data; This is the medium gradient weighting coefficient, reflecting the importance of the gradient to risk assessment. The fractal change rate weighting coefficient reflects the importance of the fractal composite index for risk assessment.

[0131] Determine multiple risk level thresholds for risk factors Different thresholds divide spatial points into L risk levels.

[0132] A deep residual network is constructed for dynamic classification of risk factors, and the cross-entropy loss function is used. As the optimization objective, among which... The number of spatial points, The number of risk levels, For the first The true risk level of each spatial point For classification networks to the first Predicted risk level for each spatial point;

[0133] Using the risk factor matrix as input and the true risk level label as output, a deep residual network is trained until the loss function converges (i.e., the prediction result is highly consistent with the true label) to obtain a risk level classification model.

[0134] The risk factors of spatial points are input into the risk level classification model to calculate the probability of each point belonging to each risk level. The maximum probability criterion is then used to classify the risk level of each spatial point to obtain the final risk level. .

[0135] Example 7

[0136] This embodiment details the method for establishing a geological risk area prediction model. This model dynamically assesses and predicts the distribution of geological risk levels by integrating real-time monitoring data and historical risk distribution data. The method for establishing the geological risk area prediction model includes:

[0137] By integrating real-time monitoring data and historical risk distribution data, a dynamic risk assessment system is established. ,in, Current time Risk level distribution The final risk level output by the risk level classification model. Historical risk level distribution Weighting for real-time data;

[0138] Reflecting the current risk status, Based on long-term accumulated risk data statistics. Controlling the contribution ratio of real-time data and historical data, through Adjustments are made to balance the impact of real-time performance and historical experience: Higher It focuses more on real-time risk, is suitable for rapidly changing geological conditions, and has lower... It relies more on historical data and is suitable for stable geological environments.

[0139] The risk level distribution across multiple time steps is represented as a time series. ,in, The time window length is specified; and the risk level and volatility of the time series are obtained.

[0140] The time series data, final risk level, assessed risk level, and risk volatility are input into the spatiotemporal convolutional network model (ST-ConvNet), which captures the spatial and temporal dependencies of the risk data. The spatial convolutional module extracts the spatial features of geological risks, such as the spatial patterns of risk level distribution. The temporal convolutional module captures the dynamic patterns of risk changes in the time series, such as short-term and long-term trends.

[0141] Training to obtain a geological risk area prediction model ;

[0142] The final predicted risk level is obtained by revising the predicted data according to the final risk level. ,in, Geological risk area prediction model Predicted risk level This is a correction factor. When When the value is large, the model relies more on the prediction data and is suitable for situations where the prediction results are stable; when When the risk level is lower, the model relies more on the real-time risk level and is suitable for rapidly changing environments.

[0143] In summary, the method for establishing a geological risk area prediction model includes the following core steps:

[0144] Dynamic risk assessment system: By combining real-time monitoring data with historical risk distribution data in a weighted manner, the risk level distribution for the current time is generated.

[0145] Time series feature extraction: Based on the risk distribution at multiple time steps, features such as risk level and risk volatility are obtained.

[0146] Spatiotemporal convolutional network modeling: Construct and train a spatiotemporal convolutional network model to capture the spatial-temporal correlation features of geological risks and predict the future risk level distribution.

[0147] Prediction result correction: The prediction results are corrected based on the current real-time risk level to generate a more reliable distribution of future risk levels.

[0148] Example 8

[0149] A smart evaluation device for the degree of karst development in tunnels includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a smart evaluation method for the degree of karst development in tunnels as described above.

[0150] Memory is used to store software programs and modules. The processor executes various terminal functions and data processing by running the software programs and modules stored in memory. Memory can mainly consist of a program storage area and a data storage area. The program storage area can store the operating system, at least one executable program required for a given function, etc.

[0151] The storage data area can store data created based on the use of the terminal. Furthermore, the memory can include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory, or other volatile solid-state storage devices.

[0152] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the intelligent evaluation method for the degree of karst development in tunnels as described above.

[0153] Computer program products include computer programs or instruction sets used to perform specific tasks or achieve specific functions. These programs or instructions are designed to be executed by a processor to implement a series of predefined steps or operations. The program product may be stored in various forms of computer storage media, such as memory, hard disks, solid-state drives, optical discs, or other forms of digital storage devices. It may exist in the form of compiled binary code or in the form of scripts or bytecode that can be executed by an interpreter. Through carefully designed algorithms and logical instructions, the program product enables the processor to process data in a specific order and manner, performing various functions such as data analysis, user interaction, and device control.

[0154] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0155] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0156] Those skilled in the art should understand that the above embodiments are merely for illustrating the present invention and are not intended to limit the scope of the invention. Those skilled in the art can make other changes or modifications based on the above invention, and these changes or modifications still fall within the scope of the present invention.

Claims

1. A tunnel karst development degree intelligent evaluation method, characterized in that, The method comprises the following steps: Collecting multi-source geological data and water chemical parameters of karst areas, and preprocessing the multi-source geological data and water chemical parameters to obtain a multi-source fusion data set; Calculating the geological fractal dimension and the water chemical fractal dimension through the multi-source fusion data set, and generating a karst development fractal evaluation comprehensive index; Inputting the karst development fractal evaluation comprehensive index into a reinforcement learning model, and dynamically determining a tunnel karst key detection area by using a deep deterministic policy gradient algorithm; Obtaining data in the key detection area, and performing noise reduction and high-dimensional fusion processing by using a generative adversarial network to construct a three-dimensional medium distribution model of the karst area; Based on the three-dimensional medium distribution model and the karst development fractal evaluation comprehensive index, classifying the risk level of the karst area, and establishing a geological risk area prediction model; The method for generating the karst development fractal evaluation comprehensive index comprises the following steps: A multi-dimensional feature matrix of a karst development region is established by using multi-source geological data and water chemical parameters in a multi-source fusion data set wherein, is a geological space coordinate, is a collection time; The multi-dimensional feature matrix is divided into multiple equal-interval grids according to geological data dimensions, and the geological fractal dimension of the multi-source geological data is calculated based on the grid result by using a box counting method , wherein, is the number of boxes containing the feature region in the grid, is the scale of the grid unit; Extracting the time series of water chemical parameters from the multi-dimensional feature matrix and mapping them to the corresponding spatial coordinates; For each time series, the cumulative deviation is calculated and the standard deviation and the Hurst exponent is obtained by fitting The hydrochemical fractal dimension of the hydrochemical parameter is calculated , ; Generating a karst development fractal evaluation comprehensive index , wherein, and are weight coefficients determined according to historical data verification. 2.The intelligent evaluation method for tunnel karst development degree according to claim 1, characterized in that, The multi-source geological data includes electromagnetic field data, elastic wave velocity field data and current field data, and the water chemical parameters include pH value, salinity and main ion concentration; The method for preprocessing the data comprises the following steps: Checking the collected multi-source geological data, eliminating abnormal values and incomplete data, and performing noise reduction on the multi-source geological data through filtering and wavelet transform; and refining the boundary features by using a gradient enhancement algorithm; Performing integrity check on the water chemical parameters and eliminating abnormal values, filling in the missing data by using an interpolation method, and then performing normalization processing; Aligning the spatial coordinates of the multi-source geological data and the water chemical parameters, mapping the water chemical parameters to the spatial grid of the multi-source geological data, and forming a unified multi-source fusion data set. 3.The intelligent evaluation method for tunnel karst development degree according to claim 1, characterized in that, The method for determining the tunnel karst key detection area comprises the following steps: Defining the state space, action space and reward function of reinforcement learning; Policy network for building a reinforcement learning model and determine its optimization objective wherein are parameters of the policy network, is a state distribution generated by the policy network, is an expected value for the state distribution based on the current policy, is a state; Value network for building reinforcement learning models and determine its optimization objective wherein is a discount factor, are parameters of the value network, is an expected value for a quadruple in an experience replay pool, is an immediate reward, is an action; constructing an experience replay pool comprising a sample quadruple wherein, is a new state; randomly batch samples from the experience replay pool and optimize the update parameters using the optimization objective function wherein, is a soft update coefficient, is a parameter of the target policy network, is a parameter of the target value network; determining an optimal action for the current state by the updated policy network , to detect the incremental displacement of the device in each direction;​ Updating the position of a probe device adding the fractal evaluation comprehensive index corresponding to the new position to the state , and the loop is iterated until the probe stopping condition is met; All historical positions of the detection equipment are the tunnel karst key detection area.

4. The intelligent evaluation method of the development degree of the tunnel karst according to claim 3, characterized in that, The state space of the definition of the reinforcement learning is the fractal evaluation comprehensive index distribution of the tunnel karst region and the current position of the detection equipment wherein, is the spatial coordinate is the karst development fractal evaluation comprehensive index, is the spatial position of the detection equipment; The action space is defined as the movement direction and step size of the probe device, the action wherein, is the incremental displacement of the probe device in each direction; Determining a reward function for reinforcement learning wherein, is a weight coefficient emphasizing the importance of the target region, is a weight coefficient penalizing the path length, is the path length of the current action. 5.The intelligent evaluation method for tunnel karst development degree according to claim 1, characterized in that, The method for constructing the three-dimensional medium distribution model of the karst area comprises the following steps: Obtain the multi-source fusion data set corresponding to each detection point in the tunnel karst intermediate detection area, and map it to a three-dimensional space grid to form an initial three-dimensional feature matrix ; A generative adversarial network model is constructed, and an adversarial loss function thereof is determined wherein, is a discrimination result of the discriminator on a feature matrix , is a discrimination result of the discriminator on a feature matrix output by the generator , is an expectation of the feature matrix sampled from a real data distribution, is the real data distribution, is a data distribution generated by the generator; reconstruction error of the generator ; generator loss function ; discriminator loss function ; is a weight coefficient for the reconstruction error of the generator, is an L2 norm; Iteratively training the discriminator and the generator, and optimizing the loss function until the generator loss function and the discriminator loss function converge; updating the discriminator parameters updating the generator parameters wherein is a learning rate for the discriminator, is a learning rate for the generator; inputting the initial three-dimensional feature matrix into the generator and outputting a denoised feature matrix , and performing multi-dimensional feature fusion to obtain a three-dimensional feature matrix ; Generating a three-dimensional medium distribution model of a karst region using volume rendering technology wherein, is a volume rendering algorithm function that maps the feature matrix to a three-dimensional model. 6.The intelligent evaluation method for tunnel karst development degree according to claim 1, characterized in that, The method for classifying the risk level of the karst area comprises the following steps: Constructing a risk input matrix wherein, is a three-dimensional medium distribution model, is a karst development fractal evaluation comprehensive index; extracting local medium gradients and local rate of change of fractal synthesis index ; determining a spatial point of risk factors wherein and are weight coefficients determined from historical data; Multiple risk level thresholds for determining risk factors ; constructing a deep residual network for dynamic classification of risk factors and determining a loss function wherein, is the number of spatial points, is the number of risk levels, is the true risk level of the th spatial point, is the predicted risk level of the th spatial point by the classification network; Using a risk factor matrix as input and a real risk level label as output, training a deep residual network until the loss function converges to obtain a risk level classification model; inputting the risk factors of the spatial points into a risk level classification model, calculating the probability of each point belonging to each risk level, and using a maximum probability criterion to classify the risk level of each spatial point to obtain a final risk level .

7. The intelligent tunnel karst development degree evaluation method according to claim 6, characterized in that, The method for establishing the geological risk area prediction model comprises the following steps: Comprehensive real-time monitoring data and historical risk distribution data, establish a dynamic risk assessment system, wherein, is the risk level distribution of the current time , is the final risk level output by the risk level classification model, is the historical risk level distribution, is the real-time data weight; representing the risk level distribution of multiple time steps as a time series wherein, is a time window length; and obtaining an evaluation risk level and a risk volatility rate of the time series; The time series, the final risk level, the evaluation risk level and the risk volatility rate are input into a space-time convolution network model as model inputs to train a geological risk region prediction model ; correcting the prediction data by the final risk level to obtain a final predicted risk level wherein, is a geological risk zone prediction model the predicted risk level, is a correction factor.

8. An intelligent tunnel karst development degree evaluation device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the method of any one of claims 1-7.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to realize the method of any one of claims 1-7.

Images