Soft rock tunnel surrounding rock parameter dynamic identification method and system based on data driving

Abstract

The invention provides a soft rock tunnel surrounding rock parameter dynamic identification method and system based on data driving, and relates to the technical field of underground tunnel mechanical parameter dynamic identification, and the method comprises the steps: obtaining multi-element tunnel surrounding rock parameters, and building a joint probability distribution model of the multi-element surrounding rock parameters based on a Copula theory; performing Monte Carlo simulation, and generating a high-dimensional parameter sample library meeting physical constraints in the parameter constraint space based on the joint probability distribution model; establishing a tunnel three-dimensional numerical model and performing automatic numerical simulation to generate multivariate response data; constructing a Kriging agent model of a Gaussian kernel function based on multivariate response data training, establishing a nonlinear mapping relation between parameter input and deformation output, constructing an inversion objective function by taking the minimum root-mean-square error of multi-measurement-point displacement as an objective, and performing inversion solution by using an adaptive particle swarm optimization algorithm to obtain inversion identification parameters, and a dynamic feedback mechanism is constructed to realize adaptive tracking of the time-varying characteristics of the surrounding rock parameters.

Description

Technical Field

[0001] The present disclosure relates to the technical field of dynamic identification of mechanical parameters of underground tunnels, and in particular to a data-driven method and system for dynamic identification of surrounding rock parameters of soft rock tunnels. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] As deep, long tunnel projects rapidly evolve towards complex geological conditions and intelligent construction, the need for inversion of surrounding rock parameters has evolved from identifying single elastic parameters to refined dynamic identification of multiple coupled parameters (elasticity, plasticity, and rheology). Inversion accuracy requirements have been upgraded from empirical threshold constraints to quantitative evaluation based on multi-source data fusion.

[0004] The inversion process for tunnel surrounding rock parameters is characterized by strong nonlinearity, high-dimensional coupling, and dynamic time-varying characteristics. This is especially true in complex geological sections such as weak fracture zones and high-stress areas. Traditional inversion methods still have the following problems:

[0005] 1) Traditional methods often suffer from non-unique solutions and difficulty converging to the global optimal solution during parameter inversion due to strong model nonlinearity, monitoring data errors, and high parameter coupling. This is especially true when inverting multiple parameters simultaneously, resulting in non-unique inversion results and difficulty in convergence.

[0006] 2) Existing methods are limited by computational efficiency and accuracy, and can usually only invert 2-3 parameters (such as elastic modulus E, Poisson's ratio ν), while in reality, elastic and plastic parameters (such as cohesion c, internal friction angle) need to be inverted simultaneously. The effect is not good when the number of parameters increases, and the inversion efficiency will be significantly reduced;

[0007] 3) Traditional methods rely on high-fidelity numerical models (such as finite element simulation) for single forward calculations, which are time-consuming and resource-intensive. They cannot support the batch calculation requirements of large-scale parameter samples, severely restricting the engineering implementation of data-driven methods, increasing the cost of forward calculation models, and limiting their practicality.

[0008] 4) The measured displacement data are easily disturbed by construction disturbances, instrument errors and uncertainty of geological conditions. Traditional filtering methods (such as uncertainty filtering) are insufficient to handle them, affecting the reliability of the inversion parameters.

[0009] 5) Traditional methods rely on orthogonal experiments or engineering experience to select initial parameter values. The calculation process requires repeated iterative calls to finite element software (such as ABAQUS and FLAC3D), which is time-consuming and difficult to automate.

[0010] 6. Existing algorithms are prone to falling into local optima when parameters are highly coupled. Furthermore, they lack a multi-solution screening mechanism (such as multi-objective joint constraints based on measured data) to address the non-uniqueness of the inversion problem. This results in significant deviations between the inversion parameters and actual operating conditions, and overall, the inversion optimization algorithm lacks robustness.

[0011] 7) Traditional inversion methods are mostly "offline static" modes, and no closed-loop mechanism has been established to dynamically update the proxy model and modify parameters in real time based on monitoring data during construction. They are unable to adapt to the time-varying characteristics of soft rock, and the inversion results lag behind the actual changes in the surrounding rock state, resulting in the lack of dynamic real-time updates and closed-loop verification.

[0012] In summary, existing methods struggle to simultaneously characterize the spatial correlations and temporal evolution of parameters, resulting in significant deficiencies in computational efficiency, solution reliability, and engineering adaptability. When the number of parameters to be inverted is large, the computational time required by traditional methods increases exponentially, and the fluctuations in inversion errors also increase. Existing methods face significant challenges in core areas such as dimensionality reduction of high-dimensional parameters and screening for non-unique solution sets. Summary of the Invention

[0013] In order to solve the above problems, the present disclosure proposes a data-driven dynamic identification method and system for surrounding rock parameters of soft rock tunnels, which supports multi-dimensional parameter and multi-field coupling parameter inversion. The sample boundary coverage rate is more in line with the actual working conditions. The credibility of the inversion parameters is ensured through the dual verification mechanism of root mean square error and geological measured data, providing an efficient and high-precision dynamic intelligent inversion solution for surrounding rock parameters for soft rock tunnel projects.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions:

[0015] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels includes:

[0016] Obtain multivariate tunnel surrounding rock parameters and perform preprocessing;

[0017] The optimal marginal distribution of the pre-processed surrounding rock parameters is estimated, the optimal marginal distribution of each surrounding rock parameter is determined, and a joint probability distribution model of multivariate surrounding rock parameters is constructed based on the Copula theory;

[0018] Monte Carlo simulation is performed based on the optimal Vine Copula structure, and efficient sample sampling is performed based on the joint probability distribution model in the parameter constraint space to generate a high-dimensional parameter sample library that meets physical constraints;

[0019] Establish a three-dimensional numerical model of the tunnel, perform automated numerical simulation on the three-dimensional numerical model of the tunnel, and generate multivariate response data;

[0020] A Kriging proxy model of Gaussian kernel function is constructed based on multivariate response data training, and a nonlinear mapping relationship between parameter input and deformation output is established. Based on the Kriging proxy model, an inversion objective function is constructed with the goal of minimizing the root mean square error of multi-measurement point displacements. The inversion solution is performed using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of surrounding rock parameters.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] The data-driven dynamic identification system for surrounding rock parameters of soft rock tunnels includes:

[0023] Parameter acquisition module, used to obtain multivariate tunnel surrounding rock parameters and perform preprocessing;

[0024] The sample space construction module is used to estimate the optimal marginal distribution of the preprocessed surrounding rock parameters, determine the optimal marginal distribution of each surrounding rock parameter, and construct a joint probability distribution model of multivariate surrounding rock parameters based on Copula theory. Monte Carlo simulation is performed based on the optimal Vine Copula structure, and efficient sample sampling is performed within the parameter constraint space based on the joint probability distribution model to generate a high-dimensional parameter sample library that meets physical constraints.

[0025] Numerical simulation module, used to establish a three-dimensional numerical model of the tunnel, perform automated numerical simulation on the three-dimensional numerical model of the tunnel, and generate multivariate response data;

[0026] The inversion optimization module is used to train and construct a Kriging proxy model of the Gaussian kernel function based on multivariate response data, establish a nonlinear mapping relationship between parameter input and deformation output, and construct an inversion objective function based on the Kriging proxy model with the goal of minimizing the root mean square error of the displacement of multiple measuring points. The inversion solution is performed using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of the surrounding rock parameters.

[0027] According to some embodiments, the present disclosure adopts the following technical solutions:

[0028] A computer program product includes a computer program, which, when executed by a processor, implements the data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels.

[0029] According to some embodiments, the present disclosure adopts the following technical solutions:

[0030] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the data-driven dynamic identification method of surrounding rock parameters of a soft rock tunnel is implemented.

[0031] According to some embodiments, the present disclosure adopts the following technical solutions:

[0032] An electronic device includes: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in this paper innovatively constructs a closed-loop technical architecture of "Vine Copula joint distribution-Kriging proxy model-adaptive particle swarm optimization". Based on the VineCopula theory, a joint probability distribution model is established, the topological structure is optimized by the maximum spanning tree rule, and a high-dimensional parameter sample library that meets physical constraints is generated by combining Monte Carlo simulation; a parameterized numerical model and an automated calculation interface are created to batch generate parameter-displacement mapping data sets and train high-precision Kriging proxy models; a dynamic adaptive particle swarm algorithm is designed; a dynamic closed-loop verification platform driven by monitoring data is constructed, and sliding window filtering and regularization are used to fuse multi-source information to achieve online updating and error correction of surrounding rock time-varying parameters, forming a full-process intelligent closed-loop system from parameter inversion to dynamic verification, providing a new tool for intelligent parameter inversion and risk prevention and control for soft rock tunnel projects. It solves the problems of low parameter coupling solution efficiency, insufficient authenticity of high-dimensional sample generation, and lack of dynamic time-varying characteristic modeling in traditional inversion methods.

[0035] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in this paper adopts a parameter value range that is more consistent with actual geological data (such as the measured statistical interval of cohesion c). The optimal marginal distribution determined based on maximum likelihood estimation and KS test is significantly better than the empirical hypothesis method; multi-parameter collaborative modeling allows the variable dimension of inversion to be expanded to more than 10 parameters (traditional methods only have 2-3 parameters), while supporting the needs of multi-field coupled inversion, breaking through the limitations of traditional methods that only invert a single parameter or do not consider the correlation between multiple parameters.

[0036] The disclosed data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels uses Vine Copula-based Monte Carlo sampling for efficient sampling and physical constraint assurance. Monte Carlo simulation is performed based on the optimal Vine Copula structure to generate simulated data between 0 and 1 that satisfies both joint distribution and physical constraints. This generates high-quality samples that satisfy both non-negativity (e.g., cohesion > 0) and parameter correlation (e.g., nonlinear correlation between E and φ). This method can improve the accuracy of characterizing the correlation between multivariate surrounding rock parameters of soft rock tunnels, ensure that the sample library simultaneously covers the interactive effects of elastic, plastic, and rheological parameters, and provide better sample boundary coverage than traditional orthogonal experiments or Latin sampling, further enhancing the rationality of parameter combinations.

[0037] The data-driven dynamic identification method of surrounding rock parameters of soft rock tunnels disclosed in this paper integrates FLAC 3D Three-dimensional refined modeling and Python automated interface technology are used to establish a dynamically adjustable three-dimensional numerical model of the tunnel, as well as a dynamic simulation system that combines physical realism and computational efficiency. The dynamic parametric modeling technology based on surrounding rock-support and construction steps, combined with real-time calibration of on-site monitoring data, significantly improves the adaptability of the numerical model to complex geological conditions. The developed automated interface realizes a closed-loop process of batch assignment of multivariate parameters and real-time extraction of response data, solving the pain points of low efficiency and data fragmentation in traditional numerical simulation, and constructing an "input variable-output response" database, which provides a standardized, high-density multi-dimensional data source for agent model training, ensuring the accuracy and reliability of subsequent inversion analysis.

[0038] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in this paper uses Z-score normalization processing on the multivariate response data generated by numerical simulation, smoothes the time series response fluctuations through sliding window filtering, eliminates outliers caused by numerical oscillations, and forms standardized, low-redundancy, and high signal-to-noise ratio structured data. It effectively solves key problems such as dimensional aliasing, noise interference, and abnormal oscillations in numerical simulation response data, constructs a high-fidelity, low-redundancy structured data set, and lays a robust and reliable data foundation for parameter inversion.

[0039] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in this paper integrates the Gaussian kernel function Kriging model and the adaptive hyperparameter optimization technology to construct a high-precision and strong generalization nonlinear proxy model; it uses maximum likelihood estimation or Bayesian optimization to dynamically calibrate hyperparameters (such as length scale and variance) to accurately capture the complex nonlinear relationship between surrounding rock mechanical parameters and tunnel deformation; through 10-fold cross validation and multi-index (R 2, RMSE, and MAPE) evaluation mechanisms to avoid the risk of model overfitting. Combined with an active learning strategy, it dynamically adds key samples, overcoming the limitations of traditional static datasets and enabling iterative optimization of surrogate models and efficient exploration of parameter space. The resulting surrogate model combines high computational efficiency and high accuracy, significantly reducing the computational cost of numerical simulations.

[0040] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in this paper innovatively adopts the forward modeling paradigm to construct a Gaussian process proxy model with surrounding rock mechanical parameters as input and tunnel deformation response as output, and establishes a deterministic mapping relationship that strictly conforms to the physical causal law. Compared with traditional inverse modeling strategies, this method improves the uniqueness of the inversion solution and significantly reduces the multi-solution problem of parameter inversion.

[0041] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in the present invention realizes efficient and reliable multi-parameter inversion by integrating multi-source data and intelligent optimization strategies: the objective function is constructed with the minimum root mean square error of displacement of multiple measuring points as the goal, taking into account the collaborative optimization of multiple physical quantities and global error control; regularization constraints and Bayesian prior distributions are introduced to suppress multi-solutions at the mathematical level while integrating geological prior knowledge to ensure that the inversion parameters are both in line with numerical optimality and meet the laws of engineering physics; the Kriging proxy model is combined to replace high-cost numerical calculations, and a dynamic adaptive sampling mechanism is used to achieve an intelligent balance between computing resources and accuracy, forming an inversion framework that can be quantified and verified, which significantly improves the efficiency and credibility of surrounding rock parameter identification under complex geological conditions.

[0042] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in the present invention effectively overcomes the local convergence problem of traditional particle swarm optimization algorithms under complex conditions through a multi-level intelligent optimization mechanism. The dynamic inertia weight mechanism adaptively balances global exploration and local development capabilities based on the population evolution state to avoid premature convergence. The parameter sensitivity guidance strategy drives the search direction through physical mechanisms, significantly improving the identification efficiency of highly sensitive parameters. The cross-perturbation of differential evolution, the strong mutation characteristics of Cauchy jumps, and the probabilistic acceptance mechanism of simulated annealing are integrated to systematically enhance the algorithm's climbing ability in the multi-peak solution space, breaking through the bottleneck of traditional particle swarm optimization algorithms that are prone to falling into local extreme values. While ensuring the physical rationality of the parameters, the global optimization efficiency and convergence stability of the multi-parameter inversion process are greatly improved.

[0043] The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels disclosed in this paper constructs a "double insurance" system for parameter credibility through a multi-dimensional verification mechanism: at the numerical level, high-precision RMSE and R2 indicators are used to ensure the accurate matching of the inversion model with the monitoring data; at the engineering level, physical correlations between parameters and geomechanical properties are established through cross-validation of geological survey measured data such as drilling core strength and elastic wave velocity. This breaks through the limitation of traditional inversion that relies solely on numerical optimization, avoids the non-physical solution that may be caused by pure data drive, and compensates for the subjective bias of single geological experience judgment, significantly improving the dual credibility of parameter inversion results in terms of numerical convergence and engineering feasibility.

[0044] The data-driven dynamic identification method of surrounding rock parameters of soft rock tunnels disclosed in this invention constructs a dynamic inversion system with self-evolution capability through the data-model co-evolution mechanism: based on the fusion of multi-source heterogeneous data, through real-time deviation rate threshold warning (±5%) and iterative update of the proxy model, it breaks through the limitation of traditional static inversion models that cannot respond to geological time changes and construction disturbances. It realizes the real-time correction of model parameters based on monitoring data, captures the time-dependent evolution law of the mechanical properties of the surrounding rock through online parameter identification, and continuously optimizes the generalization ability of the model based on the incremental learning mechanism. The traditional discretized staged inversion is optimized into a continuous time-space coupled intelligent inversion, which significantly improves the timeliness, robustness and engineering applicability of surrounding rock parameter identification in complex construction environments.

[0045] This data-driven dynamic identification method for surrounding rock parameters in soft rock tunnels utilizes multimodal interaction technology to construct an intelligent collaborative platform for deep algorithm-engineering integration. Developed using the PyQt5 framework, the platform employs a high-performance graphical interface (GUI) that integrates parameter management, model computation, and 3D visualization core functions through a modular architecture, enabling batch import of geomechanical parameters and dynamic rendering of 3D geological models. The platform also includes a built-in inversion algorithm progress monitoring module that displays inversion iteration convergence curves in real time and simultaneously displays the evolution trends of inversion parameters at different construction stages. This restructures the traditional discrete engineering analysis process into a real-time, iterative ecosystem platform comprised of "data acquisition → intelligent inversion → 3D verification." BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which constitute a part of the present disclosure, are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation to the present disclosure.

[0047] Figure 1 This is a flow chart of a data-driven method for dynamic identification of surrounding rock parameters of soft rock tunnels according to an embodiment of the present disclosure;

[0048] Figure 2 Schematic diagram of data-driven intelligent inversion of surrounding rock parameters of soft rock tunnels according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0049] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs.

[0051] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0052] Example 1

[0053] In one embodiment of the present disclosure, a data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels is provided, comprising the following steps:

[0054] Step 1: Obtain multivariate tunnel surrounding rock parameters and perform preprocessing;

[0055] Step 2: Estimate the optimal marginal distribution of the pre-processed surrounding rock parameters, determine the optimal marginal distribution of each surrounding rock parameter, and construct a joint probability distribution model of multivariate surrounding rock parameters based on Copula theory;

[0056] Step 3: Perform Monte Carlo simulation based on the optimal Vine Copula structure, perform efficient sample sampling based on the joint probability distribution model within the parameter constraint space, and generate a high-dimensional parameter sample library that meets physical constraints;

[0057] Step 4: Establish a three-dimensional numerical model of the tunnel, perform automated numerical simulation on the three-dimensional numerical model of the tunnel, and generate multivariate response data;

[0058] Step 5: Based on the multivariate response data training, a Kriging proxy model of the Gaussian kernel function is constructed to establish a nonlinear mapping relationship between parameter input and deformation output. Based on the Kriging proxy model, the inversion objective function is constructed with the minimum root mean square error of the displacement of multiple measuring points as the goal. The inversion is solved using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of the surrounding rock parameters.

[0059] As an embodiment, the disclosed data-driven dynamic identification method for surrounding rock parameters in soft rock tunnels innovatively constructs a closed-loop inversion technology architecture combining Vine Copula joint distribution, Kriging proxy model, and adaptive particle swarm optimization. Through intelligent inversion of surrounding rock parameters in soft rock tunnels, online identification, updating, and error correction of time-varying surrounding rock parameters are achieved, forming a full-process intelligent closed-loop system from parameter inversion to dynamic verification, providing a new tool for intelligent parameter inversion and risk prevention and control in soft rock tunnel projects. The specific implementation process is as follows:

[0060] Step 1: Obtain multivariate tunnel surrounding rock parameters and perform preprocessing. This includes extracting key parameters such as the soft rock elastic modulus, Poisson's ratio, and cohesion based on geological survey and laboratory test data. Statistical analysis is used to determine the distribution patterns and correlations between these parameters. A library of parameter value ranges and correlation constraint rules is then established to provide a data foundation for subsequent experimental design.

[0061] Specifically, the on-site geological survey methods include drilling sampling and in-situ stress testing, and the indoor tests include uniaxial or triaxial compression tests, direct shear tests, etc. The system obtains key surrounding rock parameters such as soft rock elastic modulus, Poisson's ratio, and cohesion, and establishes a parameter value range and correlation constraint rule library to provide a data basis for subsequent experimental design.

[0062] Among them, the parameter value range refers to the actual interval range of the measured values ​​of parameters such as elastic modulus, Poisson's ratio, and cohesion; the correlation is calculated based on the measured data using the Pearson correlation coefficient, and the correlation of this actual data can be used as the standard for subsequent simulation data; the correlation constraint rule library refers to the correlation calculation results as the standard, and subsequent simulation data is generated based on the correlation.

[0063] Furthermore, the preprocessing process includes data cleaning to remove outliers and standardization to eliminate dimensional differences.

[0064] Step 2: Use maximum likelihood to estimate the optimal marginal distribution of each surrounding rock parameter, perform KS test fitting to determine the optimal marginal distribution of each surrounding rock parameter, and construct a joint probability distribution model of multivariate surrounding rock parameters based on Copula theory.

[0065] Specifically, several probability distributions are selected, such as normal, lognormal, Gamma distribution, etc., and the function parameters of each distribution are calculated using the maximum likelihood estimation method. Then, the best distribution function is selected from these fitted probability distributions based on the evaluation standard of the KS test (the smaller the KS test value, the better the corresponding distribution function); the joint distribution refers to the joint distribution function between multiple variables, which can simultaneously describe the functions of multiple variables and take into account the correlation between multiple variables. Since the construction of the joint distribution of more than two variables is relatively difficult, the present disclosure introduces the copula function to solve this problem, which can more accurately establish the joint distribution function between multiple variables, that is, the joint distribution model.

[0066] Step 3: Perform Monte Carlo simulation based on the optimal Vine Copula structure, perform efficient sample sampling based on the joint probability distribution model within the parameter constraint space, and generate a high-dimensional parameter sample library that meets physical constraints. This includes: performing efficient sample sampling based on the joint distribution model within the parameter constraint space, ensuring that sample points evenly cover parameter boundaries and interaction-sensitive areas by truncating the marginal distribution, and generating a training dataset containing key operating condition combinations to provide sufficient data support for the proxy model.

[0067] Specifically, the maximum likelihood estimation is used to fit the Vine Copula parameters, and the maximum spanning tree rule is used to determine the optimal Vine Copula structure. Monte Carlo simulation is performed based on the optimal Vine Copula structure to generate simulated data between 0 and 1 that satisfy the joint distribution and physical constraints (such as non-negative cohesion). Then, an equal probability transformation is performed according to the optimal marginal distribution to obtain random samples of multivariate surrounding rock parameters, forming a high-dimensional sample library that characterizes the spatial randomness of surrounding rock parameters, providing more realistic and reliable data support for subsequent proxy models.

[0068] Furthermore, the fitting parameters are obtained based on maximum likelihood estimation. There may be multiple structures when fitting the copula function. The maximum spanning tree rule is used to directly determine which copula structure is optimal. Based on the fitted copula structure and parameters, Monte Carlo simulation is directly used in R or MATLAB to generate simulated data. If there are multiple variables, the simulated data of multiple variables are high-dimensional sample data.

[0069] Step 4: Establish a 3D numerical model of the tunnel, perform automated numerical simulations on the 3D numerical model, and generate multivariate response data. This includes: establishing a dynamically adjustable 3D numerical model of the tunnel; developing an automated interface to implement batch parameter assignment, parallel calculations, and displacement response data extraction to improve the efficiency of large-scale working condition calculations; generating a parameter-deformation mapping relationship dataset to statistically analyze numerical simulation results; constructing a structured parameter-response normalized dataset; and completing data standardization to eliminate dimensional differences and form a high-quality structured dataset to provide reliable data support for surrogate model training.

[0070] Specifically, based on the large-scale numerical software FLAC for geotechnical engineering 3D A dynamically adjustable three-dimensional refined numerical model of the tunnel is established, covering the interaction between surrounding rock and support structure, nonlinear constitutive relations, and construction steps (such as full-section excavation or step method). The initial model is calibrated using field monitoring data (convergent deformation) to ensure its rationality. This involves using the actual measured surrounding rock parameters on site to calculate the tunnel deformation in the numerical simulation software and comparing it with the actual measured deformation. The model is calibrated when the error between the two is within an acceptable range.

[0071] An automated calculation interface was developed based on Python, integrating parametric modeling, dynamic configuration of boundary conditions, and result post-processing functions. This enabled batch assignment of the multivariate surrounding rock parameters (soft rock elastic modulus, Poisson’s ratio, cohesion, etc.) generated above and real-time extraction of key response data (vault settlement, horizontal convergence), and constructed a mapping relationship database covering “input parameters-spatial responses” to provide high-quality raw data support for subsequent proxy model modeling.

[0072] Step 5: Based on the multivariate response data training, a Kriging proxy model of the Gaussian kernel function is constructed to establish a nonlinear mapping relationship between parameter input and deformation output. Based on the Kriging proxy model, the inversion objective function is constructed with the goal of minimizing the root mean square error of the displacement of multiple measuring points. The inversion is solved using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of the surrounding rock parameters.

[0073] Specifically, the multivariate response data generated by numerical simulations is first smoothed using sliding window filtering to remove outliers caused by numerical fluctuations. Z-score normalization is then applied to generate standardized, low-redundancy, high-signal-to-noise ratio structured data. The Z-score (standard score) measures the position of a data point relative to the population mean, representing the difference between the sample value and the mean in units of standard deviation. In data normalization, the Z-score is often used to eliminate the dimensional and scale effects between different indicators to ensure analytical fairness.

[0074] Based on the Gaussian kernel function, the Kriging proxy model uses maximum likelihood estimation or Bayesian optimization to determine hyperparameters, constructs a nonlinear mapping relationship between surrounding rock structure data and tunnel deformation response, and adopts an active learning strategy to add new numerical simulation samples to construct the proxy model.

[0075] Secondly, the traditional inversion method reversely infers the surrounding rock mechanical parameters based on measured deformation data by constructing an inverse proxy model with tunnel response as input variable and surrounding rock parameters as output variable. However, this paradigm has inherent flaws in its mathematical nature: according to the forward mapping relationship of the geotechnical constitutive equation, a specific combination of surrounding rock parameters must correspond to a unique tunnel mechanical response, but the inverse mapping process has a non-injective characteristic - a single tunnel response may correspond to multiple feasible solution sets in the multidimensional parameter space, resulting in significant multi-solution inversion results. The root cause of this ill-posed problem is that the traditional method forcibly converts the forward system (surrounding rock parameters → tunnel response) with a clear physical causal relationship into an inverse proxy model (tunnel response → surrounding rock parameters), which violates the deterministic mapping law of the geotechnical mechanics system. Therefore, the present disclosure trains the proxy model based on the parameter-displacement dataset, innovatively adopts the forward modeling paradigm, constructs a Gaussian process proxy model with surrounding rock mechanical parameters as input and tunnel deformation response as output, and establishes a deterministic mapping relationship that strictly conforms to the physical causal law. Compared with the traditional inverse modeling strategy, this method improves the uniqueness of the inversion solution and significantly reduces the multi-solution of parameter inversion. The model accuracy is evaluated through cross-validation, and the model structure is dynamically optimized until it meets the engineering prediction requirements. Specifically, based on the Kriging proxy model of the Gaussian kernel function, the maximum likelihood estimation (MLE) or Bayesian optimization is used to determine the hyperparameters (length scale, variance), and to construct a nonlinear mapping relationship between the surrounding rock mechanical parameters and the tunnel deformation response. The prediction accuracy of the model is evaluated using 10-fold cross-validation, and the determination coefficient (R 2 ), root mean square error (RMSE) and mean absolute percentage error (MAPE), an active learning strategy is used to add numerical simulation samples and iteratively update the proxy model; the generalization ability of the model is verified by test data to ensure that its prediction error within the engineering parameter range meets the design requirements, providing an efficient and high-precision proxy model for subsequent parameter inversion and construction optimization.

[0076] Furthermore, an objective function is constructed with the goal of minimizing the root mean square error between the measured and predicted values. This includes defining an inversion mathematical model (objective function) with the goal of minimizing the root mean square error between the measured settlement and horizontal convergence and the corresponding predicted values, introducing a regularization term to suppress parameter ambiguity, and forming an inversion optimization framework that can be quantified and evaluated.

[0077] Specifically, an inversion optimization framework is established based on the difference between the measured displacement data of the tunnel and the predicted response of the numerical model. The root mean square error (RMSE) between the measured and predicted values ​​is used as the core objective function (Equation 1) to achieve collaborative optimization of multiple measurement points and multiple physical quantities; regularization or sparse constraints (L1 / L2 norm mixed penalty terms) are introduced to reduce the multi-solution nature of the model, and geological survey information is integrated through the prior distribution under the Bayesian framework (such as Gaussian prior or uniform distribution) to compress the non-physical feasible solution space; the Kriging proxy model is integrated to replace the high-cost calculation of the numerical model, and the proxy model is dynamically updated through an adaptive sampling strategy to achieve a balance between inversion accuracy and computing resources, forming an inversion framework that can quantify the evaluation of parameter inversion quality. The objective function is as follows:

[0078]

[0079] Where S and C are the tunnel convergence and settlement values, respectively, and n is the number of monitoring sections. The regularization term λ||θ||2 is used to suppress parameter ambiguity, and i represents the i-th monitoring section.

[0080] Furthermore, to address the problem of multi-parameter coupled inversion being prone to falling into local optimality, a dynamic adaptive particle swarm algorithm is designed for optimization and solution. The core of the algorithm includes:

[0081] 1) Dynamic inertia weight mechanism, which adjusts weights based on population fitness variance and particle dispersion;

[0082] 2) Parameter sensitivity guidance strategy, which gives highly sensitive parameters a larger step weight in speed update and optimizes the search direction;

[0083] 3) A hybrid mutation strategy introduces crossover perturbations and Cauchy jumps of differential evolution to particles that have not been optimized for a long time, and combines simulated annealing to probabilistically accept suboptimal solutions to break through the multi-peak trap.

[0084] As an embodiment, a multi-layer verification system is constructed for the obtained inversion parameters to ensure the credibility of the inversion parameters. The specific method is as follows:

[0085] 1) Quantify the goodness of fit between the back-substitution model and the measured data based on the root mean square error (RMSE ≤ 5%) and the coefficient of determination (R2 ≥ 0.95);

[0086] 2) At the geological matching level, the drilling core strength, elastic wave velocity test values ​​and ground stress measured data are compared, and the relative error of the physical quantities derived from the inversion parameters is required to be ≤10%, ensuring that the parameters have both numerical rationality and engineering rationality.

[0087] Finally, in one embodiment, a dynamic closed-loop platform based on data-model bidirectional drive is constructed to achieve real-time tracking of the time-varying characteristics of surrounding rock parameters, including: integrating multi-point displacement meters, stress sensors and seepage monitoring data to quantify the deviation rate between the predicted value and the measured value (the threshold is set at ±5%); when the cumulative deviation exceeds the limit, the newly added monitoring data is automatically integrated to iteratively update the proxy model to achieve adaptive matching of geological time variability and construction disturbance effects.

[0088] A graphical interface was developed based on PyQt5 to build an integrated intelligent platform for parameter input, calculation monitoring, and result visualization. It supports batch import of surrounding rock parameters and interactive display of 3D models, real-time monitoring of the inversion algorithm's running progress and resource consumption, dynamic generation of inversion results, and comparative analysis of predicted and measured data. The platform is compatible with mobile terminal operations and can be linked to the construction monitoring system to synchronize data in real time, forming a full-process visual closed-loop system from parameter input, intelligent inversion to dynamic verification.

[0089] Example 2

[0090] In one embodiment of the present disclosure, a data-driven dynamic identification system for surrounding rock parameters of soft rock tunnels is provided, comprising:

[0091] Parameter acquisition module, used to obtain multivariate tunnel surrounding rock parameters and perform preprocessing;

[0092] The sample space construction module is used to estimate the optimal marginal distribution of the preprocessed surrounding rock parameters, determine the optimal marginal distribution of each surrounding rock parameter, and construct a joint probability distribution model of multivariate surrounding rock parameters based on Copula theory. Monte Carlo simulation is performed based on the optimal Vine Copula structure, and efficient sample sampling is performed within the parameter constraint space based on the joint probability distribution model to generate a high-dimensional parameter sample library that meets physical constraints.

[0093] Numerical simulation module, used to establish a three-dimensional numerical model of the tunnel, perform automated numerical simulation on the three-dimensional numerical model of the tunnel, and generate multivariate response data;

[0094] The inversion optimization module is used to train and construct a Kriging proxy model of the Gaussian kernel function based on multivariate response data, establish a nonlinear mapping relationship between parameter input and deformation output, and construct an inversion objective function based on the Kriging proxy model with the goal of minimizing the root mean square error of the displacement of multiple measuring points. The inversion solution is performed using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of the surrounding rock parameters.

[0095] Example 3

[0096] In one embodiment of the present disclosure, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the data-driven dynamic identification method for surrounding rock parameters of a soft rock tunnel is implemented.

[0097] Example 4

[0098] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the data-driven dynamic identification method of surrounding rock parameters of a soft rock tunnel is implemented.

[0099] Example 5

[0100] In one embodiment of the present disclosure, an electronic device is provided, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device implements the data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels.

[0101] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0103] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels, characterized by: include: Obtain multivariate tunnel surrounding rock parameters and perform preprocessing; The optimal marginal distribution of the pre-processed surrounding rock parameters is estimated, the optimal marginal distribution of each surrounding rock parameter is determined, and a joint probability distribution model of multivariate surrounding rock parameters is constructed based on the Copula theory; Monte Carlo simulation is performed based on the optimal Vine Copula structure, and efficient sample sampling is performed based on the joint probability distribution model in the parameter constraint space to generate a high-dimensional parameter sample library that meets physical constraints; Establish a three-dimensional numerical model of the tunnel, perform automated numerical simulation on the three-dimensional numerical model of the tunnel, and generate multivariate response data; A Kriging proxy model of Gaussian kernel function is constructed based on multivariate response data training, and a nonlinear mapping relationship between parameter input and deformation output is established. Based on the Kriging proxy model, an inversion objective function is constructed with the goal of minimizing the root mean square error of multi-measurement point displacements. The inversion solution is performed using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of surrounding rock parameters.

2. The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels according to claim 1, characterized in that: Based on on-site geological surveys and indoor geotechnical test data, multivariate tunnel surrounding rock parameters and their value ranges were obtained, including key parameters such as the elastic modulus, Poisson's ratio, and cohesion of soft rock. Data cleaning, outlier removal, and standardization were performed to eliminate dimensional differences. Maximum likelihood estimation was used to estimate the optimal marginal distribution of each surrounding rock parameter. KS test fitting was performed to determine the optimal marginal distribution of each surrounding rock parameter. Finally, a joint probability distribution model for the multivariate surrounding rock parameters was constructed based on Copula theory.

3. The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels according to claim 1, characterized in that: The Vine Copula parameters are fitted using maximum likelihood estimation, and the optimal Vine Copula structure is determined using the maximum spanning tree rule. Monte Carlo simulation is performed based on the optimal Vine Copula structure to generate simulated data between 0 and 1 that satisfy the joint distribution and physical constraints. Then, an equal probability transformation is performed according to the optimal marginal distribution to obtain random samples of multivariate surrounding rock parameters, generating a high-dimensional parameter sample library that satisfies physical constraints and characterizes the spatial randomness of surrounding rock parameters.

4. The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels according to claim 1, characterized in that: Based on numerical software FLAC 3D A dynamically adjustable three-dimensional numerical model of the tunnel is established, the initial model is calibrated using field monitoring data, an automated calculation interface is developed, and parametric modeling, dynamic configuration of boundary conditions, and result post-processing are integrated to achieve batch assignment of multivariate surrounding rock parameters and real-time generation of multivariate response data, and to construct a mapping relationship database covering input parameters and spatial responses.

5. The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels according to claim 1, characterized in that: The multivariate response data generated by numerical simulation is normalized using Z-score, and the fluctuation of the time series response is smoothed by sliding window filtering to remove outliers caused by numerical oscillations, thus forming standardized, low-redundancy, and high signal-to-noise ratio structured data. Based on the Gaussian kernel function, the Kriging proxy model uses maximum likelihood estimation or Bayesian optimization to determine hyperparameters, constructs a nonlinear mapping relationship between surrounding rock structure data and tunnel deformation response, and adopts an active learning strategy to add new numerical simulation samples to construct the proxy model.

6. The data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels according to claim 1, characterized in that: The objective function is constructed with the goal of minimizing the root mean square error between the measured and predicted values ​​to achieve collaborative optimization of multiple measuring points and multiple physical quantities; regularization or sparse constraints are introduced, geological survey information is integrated through the prior distribution under the Bayesian framework, the non-physical feasible solution space is compressed, and the high-cost calculation of the numerical model is replaced by an integrated proxy model. The proxy model is dynamically updated through an adaptive sampling strategy, and a dynamic adaptive particle swarm algorithm is designed to avoid the multi-parameter coupled inversion from falling into the local optimum; a dynamic closed-loop platform based on data-model bidirectional drive is constructed to quantify the deviation rate between the predicted value and the measured value. When the cumulative deviation exceeds the limit, the newly added monitoring data is automatically integrated to iteratively update the proxy model to achieve real-time tracking of the time-varying characteristics of the surrounding rock parameters.

7. A data-driven dynamic identification system for surrounding rock parameters of soft rock tunnels, characterized by: include: Parameter acquisition module, used to obtain multivariate tunnel surrounding rock parameters and perform preprocessing; The sample space construction module is used to estimate the optimal marginal distribution of the preprocessed surrounding rock parameters, determine the optimal marginal distribution of each surrounding rock parameter, and construct a joint probability distribution model of multivariate surrounding rock parameters based on Copula theory. Monte Carlo simulation is performed based on the optimal Vine Copula structure, and efficient sample sampling is performed within the parameter constraint space based on the joint probability distribution model to generate a high-dimensional parameter sample library that meets physical constraints. Numerical simulation module, used to establish a three-dimensional numerical model of the tunnel, perform automated numerical simulation on the three-dimensional numerical model of the tunnel, and generate multivariate response data; The inversion optimization module is used to train and construct a Kriging proxy model of the Gaussian kernel function based on multivariate response data, establish a nonlinear mapping relationship between parameter input and deformation output, and construct an inversion objective function based on the Kriging proxy model with the goal of minimizing the root mean square error of the displacement of multiple measuring points. The inversion solution is performed using the adaptive particle swarm optimization algorithm to obtain the inversion identification parameters, and a dynamic feedback mechanism is constructed to achieve adaptive tracking and correction of the time-varying characteristics of the surrounding rock parameters.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the data-driven dynamic identification method for surrounding rock parameters of a soft rock tunnel according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels according to any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the data-driven dynamic identification method for surrounding rock parameters of soft rock tunnels as described in any one of claims 1 to 6.

Images