A tunnel geotechnical structure analysis method and system

Abstract

The application discloses a tunnel rock-soil structure analysis method and system, relates to the technical field of rock-soil engineering, and changes the tunnel rock-soil structure analysis method and system into a time sequence form for prediction, so that the change track of parameters in a preset period in the future can be directly obtained, and the change from post-event equivalent inversion to future-oriented prediction is realized; on the basis, the forward simulation element model trained by the high-fidelity numerical simulation result significantly reduces the cost of single calculation of the mechanical response of the tunnel structure, so that it is possible to quickly evaluate a plurality of supporting schemes and rock-soil evolution conditions within a reasonable time, and an analysis tool that can be iterated in real time is provided for on-site construction.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering technology, and in particular to a method and system for analyzing the geotechnical structure of tunnels. Background Technology

[0002] In tunnel construction projects involving deep burial, long distances, and complex geological conditions, the mechanical parameters and state of the surrounding rock are the foundation for tunnel geotechnical structure analysis and support design. Current technologies mostly utilize numerical inverse analysis and optimization calculations to invert the elastoplastic and softening parameters of the surrounding rock based on monitoring data such as surrounding rock deformation and support stress, in order to support surrounding rock classification and support parameter selection. In recent years, some solutions have introduced intelligent algorithms such as machine learning, which automatically learn the nonlinear mapping relationship between monitoring data and geotechnical parameters to improve inversion efficiency and accuracy, achieving intelligent inversion of surrounding rock mechanical parameters.

[0003] However, in working conditions such as soft rock and highly compressive strata, the surrounding rock after tunnel excavation is subjected to the combined effects of multiple factors such as excavation unloading, support installation, and groundwater. The stress environment in which it is located continues to evolve over a long period of time, and the strength and deformation parameters of the surrounding rock exhibit obvious time-varying characteristics and deterioration trends. This evolution process is continuous and phased. Existing inverse analysis or intelligent inverse methods generally use monitoring data within a certain time period to fit a set of equivalent mechanical parameters to represent the average or comprehensive state within that time period. It is assumed that the parameters remain unchanged within that time period, without explicitly characterizing the process of the evolution of the surrounding rock mechanical parameters over time.

[0004] Since the above parameter inversion results are essentially a retrospective fit of existing monitoring data, they reflect the mechanical behavior of the surrounding rock over a period of time. When such static equivalent parameters are directly used as the basis for the next stage of support design and safety assessment, they are prone to deviate from the continuous evolution of the surrounding rock in subsequent periods. Under the condition that the properties of the surrounding rock change rapidly over time, this method of extrapolating future working conditions based on historical equivalent parameters has the problems of delayed support design response and difficulty in timely reflecting the deterioration trend of the surrounding rock, which makes it difficult to meet the requirements for the adaptability and forward-looking control of the support system in projects such as deep-buried soft rock tunnels. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a method and system for analyzing tunnel geotechnical structures, which solves the problem that existing tunnel parameter back analysis is mostly based on fitting static equivalent parameters to monitoring data, making it difficult to reflect the continuous deterioration and stress redistribution of the surrounding rock over time, resulting in the support design being slow to respond to future working conditions and lacking foresight.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a method for analyzing the rock and soil structure of a tunnel, comprising:

[0009] Step S1: Obtain the time-series monitoring data and corresponding engineering time-series labels at the tunnel construction site. The time-series monitoring data includes at least displacement data and stress data of the surrounding rock and support structure. The engineering time-series labels include at least time information and process stage information related to the tunnel excavation cycle and support construction.

[0010] Step S2: Input the time-series monitoring data and engineering time-series labels into the pre-trained time-varying prediction model of surrounding rock mechanical parameters. The time-varying prediction model of surrounding rock mechanical parameters is constructed based on the modeling of the evolution of surrounding rock state coupled with excavation disturbance factors and time effects, and outputs the change curve of surrounding rock mechanical parameters in the future preset time period.

[0011] Step S3: Based on the change curve of the surrounding rock mechanical parameters and the current support scheme, input the surrounding rock mechanical parameters, support parameters and tunnel geometric parameters into the pre-trained forward simulation meta-model, and use the forward simulation meta-model to predict the mechanical response of the tunnel structure in the future preset time period;

[0012] Step S4: Using the support parameters as optimization variables, with the goal of keeping the predicted mechanical response within a preset safe range, the optimization algorithm is invoked to iteratively adjust the support parameters, and the predicted mechanical response of each iteration is evaluated using the forward simulation meta-model until the iteration termination condition is met, and the adjustment amount of the support parameters is output.

[0013] As a preferred embodiment of the tunnel geotechnical structure analysis method of the present invention, step S1 includes:

[0014] Step S11: Assign an engineering time sequence label to each monitoring data point. The engineering time sequence label includes a timestamp and a process stage identifier. The process stage identifier at least distinguishes between the tunneling cycle stage and the support construction stage.

[0015] Step S12: Based on the timestamp, the time-series monitoring data is aligned along the time axis to synchronize data from different measuring points and different monitoring types;

[0016] Step S13: Associate the aligned time-series monitoring data with the corresponding engineering time-series labels to form a structured monitoring data sequence with time-process labels.

[0017] As a preferred embodiment of the tunnel rock and soil structure analysis method of the present invention, the method for obtaining the time-varying prediction model of the surrounding rock mechanical parameters includes:

[0018] Using the structured monitoring data sequence as input features, and the sequence of surrounding rock mechanical parameters obtained through inversion analysis and / or field tests within the same engineering phase as supervised learning labels, the time series prediction model is trained so that it learns the mapping relationship from the monitoring data sequence to the sequence of changes in surrounding rock mechanical parameters.

[0019] The time-series prediction model extracts time-series features from the structured monitoring data sequence during training and performs coupled modeling of excavation disturbance factors and time effects.

[0020] As a preferred embodiment of the tunnel rock and soil structure analysis method of the present invention, the time-varying prediction model of the surrounding rock mechanical parameters is a time-series neural network model based on deep learning. The time-series neural network model includes at least a recurrent neural network layer or a self-attention mechanism layer for extracting time correlations, and a regression output layer for outputting the surrounding rock mechanical parameters at different time steps.

[0021] As a preferred embodiment of the tunnel geotechnical structure analysis method of the present invention, the method for obtaining the forward simulation meta-model includes:

[0022] High-fidelity tunnel geotechnical numerical simulation software was used to generate a simulation sample dataset covering different combinations of surrounding rock mechanical parameters, support parameters, and tunnel geometric parameters.

[0023] Using surrounding rock mechanical parameters, support parameters, and tunnel geometric parameters as input features, and the key indicators of displacement field and stress field in the corresponding simulation results as output labels, a deep neural network model is trained, enabling the deep neural network model to approximate the physical relationship between input and output at a calculation speed far exceeding that of the original numerical simulation software.

[0024] As a preferred embodiment of the tunnel geotechnical structure analysis method of the present invention, the method further includes:

[0025] Step S5: After outputting the support parameter adjustment amount, store the support scheme corresponding to the support parameter adjustment amount, as well as the characteristics of the surrounding rock mechanical parameter change curve that triggered the support parameter adjustment amount, into the historical case library.

[0026] Specifically, step S4, before invoking the optimization algorithm to iterate the support parameters, further includes:

[0027] Based on the characteristics of the current surrounding rock mechanical parameter variation curve, similarity matching is performed in the historical case library, and the adjustment amount of support parameters in the historical cases obtained by similarity matching is used as the initial iteration point of the optimization algorithm;

[0028] The similarity matching includes calculating the similarity between the target rock mechanical parameter variation curve features and the rock mechanical parameter variation curve features in historical cases, and selecting at least one historical case based on the similarity.

[0029] In a preferred embodiment of the tunnel geotechnical structure analysis method of the present invention, the optimization algorithm in step S4 is a heuristic global optimization algorithm and / or a gradient-based local optimization algorithm. The optimization algorithm searches for solutions that satisfy the following constraints within the preset search space of support parameters:

[0030] The predicted displacement and predicted stress-related indicators are all within the corresponding preset safety range.

[0031] Secondly, the present invention provides a tunnel geotechnical structure analysis system, comprising,

[0032] The data acquisition and processing module is used to acquire time-series monitoring data and engineering time-series labels at the tunnel construction site, and to align the time-series monitoring data with the time axis and associate it with the engineering time-series labels to generate a structured monitoring data sequence.

[0033] The parameter time-varying prediction module integrates a time-varying prediction model for surrounding rock mechanical parameters. It is used to receive the structured monitoring data sequence and predict the change curve of surrounding rock mechanical parameters within a future preset period based on the coupling modeling of excavation disturbance factors and time effects.

[0034] The support decision optimization module integrates a forward simulation meta-model and an optimization algorithm. It is used to calculate and predict the mechanical response by iteratively calling the forward simulation meta-model based on the change curve of the surrounding rock mechanical parameters and the support parameters corresponding to the current support scheme, and to perform optimization calculations within the support parameter search space, and output the adjustment amount of the support parameters.

[0035] As a preferred embodiment of the tunnel rock and soil structure analysis system of the present invention, the tunnel rock and soil structure analysis system further includes a case library management module for storing and managing a historical case library, wherein the historical case library includes at least the variation curve characteristics of different surrounding rock mechanical parameters and the corresponding adjustment amount of support parameters;

[0036] The case library management module is also used to perform similar matching in the historical case library based on the characteristics of the current surrounding rock mechanical parameter change curve before the support decision optimization module starts the optimization algorithm, so as to provide the support decision optimization module with the initial iteration point for the adjustment of support parameters.

[0037] As a preferred embodiment of the tunnel geotechnical structure analysis system of the present invention, the data acquisition and processing module is further used to detect outliers and handle missing values ​​in the original monitoring data, and to organize the processed monitoring data and engineering time series labels into multivariate time series data with a unified time step and input them into the parameter time-varying prediction module.

[0038] The beneficial effects of this invention are as follows: The tunnel geotechnical structure analysis method and system proposed in this invention introduces structured monitoring time-series data with time and process labels, and combines it with a surrounding rock state evolution model that couples excavation disturbance factors with time effects. This allows the surrounding rock mechanical parameters to no longer be regarded as fixed constants, but to be predicted in the form of time series. This enables the direct acquisition of the parameter change trajectory within a preset future time period, realizing a shift from ex-post equivalent inversion to future-oriented prediction. Based on this, the forward simulation meta-model trained with high-fidelity numerical simulation results significantly reduces the cost of calculating the mechanical response of the tunnel structure per calculation, making it possible to quickly evaluate various support schemes and surrounding rock evolution scenarios within a reasonable timeframe, providing a real-time iterative analysis tool for on-site construction. Furthermore, with the goal of ensuring that the predicted displacement and stress indices are within a safe range, the support parameters are used as optimization variables for iterative optimization. This allows the support design to proactively adjust to the future evolution trend of the surrounding rock, reducing the design lag risk caused by relying on historical equivalent parameters in traditional methods. Meanwhile, by storing the characteristics of the surrounding rock mechanical parameter change curves and the corresponding support parameter adjustment amounts in the historical case database, and performing similarity matching based on multi-index fusion similarity, the empirical support adjustment results of typical working conditions are used as the initial point for optimization. This not only accelerates the convergence speed of the optimization algorithm, but also realizes the systematic inheritance and reuse of engineering experience to a certain extent.

[0039] In summary, this invention establishes a close coupling relationship between time-varying parameter prediction, rapid simulation, and support optimization decision-making, which is beneficial to improving the adaptability, timeliness, and safety margin of tunnel support design under complex geological conditions. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0041] Figure 1 This is a flowchart illustrating the tunnel rock and soil structure analysis method in the embodiment.

[0042] Figure 2 This is a schematic diagram of the framework of the tunnel geotechnical structure analysis system in the embodiment. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0044] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0045] This application proposes a method for analyzing tunnel geotechnical structures, combined with... Figure 1 As shown, the method includes:

[0046] Step S1: Obtain the time-series monitoring data and corresponding engineering time-series labels at the tunnel construction site. The time-series monitoring data shall include at least the displacement data and stress data of the surrounding rock and support structure, and the engineering time-series labels shall include at least the time information and process stage information related to the tunnel excavation cycle and support construction.

[0047] In this embodiment, the time-series monitoring data can be provided by existing displacement and stress monitoring points at the tunnel construction site. Displacement monitoring points are deployed at the crown, waist, sidewalls, and invert of typical cross-sections, while stress monitoring points are deployed on support components such as shotcrete layers, anchor bolts, or steel arch frames. Each monitoring point is connected to a recording system with the same time reference through a data acquisition device to ensure the consistency of time information. Time information can be recorded in the form of year, month, day, hour, and minute, with a preferred time accuracy of no less than one minute to accurately depict the sequence and duration of tunneling cycles and support construction. Process stage information can be automatically generated by the construction management system or entered by on-site technicians based on construction logs. Each monitoring time point is labeled as a stage, such as during initial support construction or secondary lining construction, thus completing the process labeling of the monitoring data without changing the monitoring methods. For example, during routine implementation, the data acquisition interval for the displacement of the surrounding rock and support structure can be set to 10-30 minutes. This interval can be extended to one to two hours during periods of minimal disturbance or construction pauses. The stress data acquisition interval is typically consistent with or a multiple of the displacement interval; the specific value can be determined by the project's technical manager based on the number of monitoring points and the capacity of the data acquisition equipment. Similarly, to ensure the completeness of the tunneling cycle time and support construction time records, the construction team can record the event time points in the construction log each time blasting or tunneling machine starts, initial shotcreting begins, and steel frame installation is completed. These time points are then synchronized with the monitoring system's time reference, and manual correction is performed when the time deviation exceeds a preset threshold. Optionally, if a short-term failure of the monitoring equipment or communication interruption results in the loss of certain time points, a missing measurement mark can be added to that time period during the data preprocessing stage. In subsequent steps, interpolation or the use of the most recent valid value can be used to fill in the missing measurement, while retaining the missing measurement mark to avoid misjudgments in subsequent model training and prediction.

[0048] Step S2: Input the time-series monitoring data and engineering time-series labels into the pre-trained time-varying prediction model of surrounding rock mechanical parameters. The time-varying prediction model of surrounding rock mechanical parameters is constructed based on the modeling of the evolution of surrounding rock state coupled with excavation disturbance factors and time effects, and outputs the change curve of surrounding rock mechanical parameters in the future preset time period.

[0049] For example, the preset time period can be determined based on the tunnel construction progress and the update frequency of the support design. In deep-buried soft rock tunnels, the preset time period can be selected within a range of 3 to 30 days, or corresponding to several tunneling cycles, such as 5 to 20 cycles. This ensures that the prediction results cover the key stages of the stress evolution of the support structure without deviating too far from the construction plan. The time step size can be consistent with the time step size of the structured monitoring data so that the model uses a unified time grid during training and prediction. When the unified time step size is one hour, a prediction output length corresponding to a preset time period of seven days is 168 time steps. Optionally, in sections where the construction phase changes rapidly, the preset time period can be set relatively short, such as 3 to 7 days, to improve the timeliness of the prediction. In sections where the construction is relatively stable or the surrounding rock conditions are relatively good, the preset time period can be appropriately lengthened to give the support design and material preparation work a longer forward-looking perspective.

[0050] In this embodiment, the calling cycle of the time-varying prediction model of surrounding rock mechanical parameters can be coordinated with the construction organization. For example, a prediction can be automatically triggered every time a certain number of tunneling cycles are completed or every preset time interval, so that the prediction curve can be updated in real time during the construction process and provide the latest input for support decision optimization.

[0051] Step S3: Based on the change curve of the surrounding rock mechanical parameters and the current support scheme, input the surrounding rock mechanical parameters, support parameters and tunnel geometric parameters into the pre-trained forward simulation meta-model, and use the forward simulation meta-model to predict the mechanical response of the tunnel structure in the future preset time period.

[0052] Step S4: Using the support parameters as optimization variables, with the goal of keeping the predicted mechanical response within a preset safe range, the optimization algorithm is called to iteratively adjust the support parameters, and the predicted mechanical response of each iteration is evaluated using the forward simulation meta-model until the iteration termination condition is met, and the adjustment amount of the support parameters is output.

[0053] In one embodiment, step S1 includes:

[0054] Step S11: Assign an engineering time sequence label to each monitoring data point. The engineering time sequence label includes a timestamp and a process stage identifier. The process stage identifier can at least distinguish between the tunneling cycle stage and the support construction stage.

[0055] Step S12: Align the time-series monitoring data with the time axis based on the timestamp, and synchronize the data of different measuring points and different monitoring types;

[0056] Step S13: Associate the aligned time-series monitoring data with the corresponding engineering time-series labels to form a structured monitoring data sequence with time-series and process labels;

[0057] Specifically, structured monitoring data sequences can be stored in a computer as a two-dimensional array or a time-indexed data table. Rows correspond to discrete time steps on a unified time axis, while columns correspond to various monitoring variables and their process stage labels, including fields such as displacement, stress values, tunneling cycle number, and support construction status at different measuring points. The basic time step of the time axis can be selected based on the construction progress and monitoring density. Commonly used fixed step sizes include ten minutes, thirty minutes, or one hour, ensuring clear engineering significance between adjacent time steps and easy correlation with construction logs.

[0058] In this embodiment, to ensure the stability of the model input data, a one-hour time step can be preferentially used. When the original monitoring time interval is less than one hour, multiple observations can be averaged or the maximum value can be taken within a one-hour window and assigned to the corresponding time step through time aggregation. When the original monitoring time interval is greater than the unified time step, linear interpolation or keeping the data of the previous moment unchanged can be used to fill the values ​​of intermediate time steps. At the same time, internal markers are added to the data table to distinguish between measured values ​​and filled values. For example, in a continuous monitoring time window, if a monitoring point is missing data for more than 24 hours, the samples corresponding to that period can be removed or listed as a low-quality sample type when constructing the subsequent model training samples, to avoid the model learning unreliable time series patterns due to long-term missing data. Furthermore, after the structured data sequence is constructed, the data can be grouped according to tunnel mileage or zone, so that data of the same mileage segment or the same surrounding rock level can be input into the time series prediction model as a separate batch, thereby improving the model's adaptability to local geological features.

[0059] In one embodiment, the time-varying prediction model for surrounding rock mechanical parameters is obtained by:

[0060] Using structured monitoring data sequences as input features and rock mechanics parameter sequences obtained through inversion analysis and / or field tests within the same engineering phase as supervised learning labels, a time-series prediction model is trained to learn the mapping relationship from monitoring data sequences to rock mechanics parameter change sequences.

[0061] In this embodiment, the training samples for the time-series prediction model can be constructed from structured monitoring data sequences using a sliding time window. Each sample contains an input sequence of continuous time steps and a sequence of changes in surrounding rock mechanical parameters corresponding to several future time steps. The length of the input time window can be selected from 24 hours to 30 days, preferably 7-14 days, thus including sufficient information on the disturbance and recovery process while avoiding excessively long windows that could lead to model convergence difficulties. The sequence of surrounding rock mechanical parameters can include equivalent values ​​of parameters such as elastic modulus, cohesion, and internal friction angle. The label data is primarily derived from the inversion analysis results during construction. When inversion results are lacking for a certain time period, interpolation results from adjacent time periods or representative values ​​based on field tests can be used as supplementary labels, and the source type is indicated in the sample labeling to allow for setting different loss weights for different sources during training. Furthermore, the training data can be divided into training and validation sets according to different tunnel mileage sections or different surrounding rock grades. Common gradient descent optimization methods are used during training, with the number of training rounds controlled between tens and hundreds. Training is stopped early when the validation set error no longer decreases significantly within a few rounds to avoid overfitting. The value ranges of parameters such as elastic modulus, cohesion, and internal friction angle can be constrained by combining geological survey reports and laboratory test results. For example, the elastic modulus can be limited to the upper and lower limits corresponding to engineering geological conditions. During training, predicted outputs exceeding physically reasonable ranges are truncated or remapped to ensure that the model output results numerically conform to geotechnical engineering principles.

[0062] Among them, the time-series prediction model extracts time-series features from the structured monitoring data sequence during the training process and performs coupled modeling of excavation disturbance factors and time effects to characterize the evolution of surrounding rock mechanical parameters over time.

[0063] In one implementation, the excavation disturbance factor and the time effect are coupled and modeled in the following manner:

[0064] Step S21: In the time-series prediction model, the surrounding rock mechanical parameters are treated as discrete-time state quantities. The temporal evolution of the surrounding rock state is described through state update, using the following formula:

[0065] ,

[0066] in, Indicates time step The vector of surrounding rock mechanical parameters at a given location, wherein the vector components include at least one or more of the following surrounding rock mechanical parameters: elastic modulus, cohesion, and internal friction angle. Indicates time step The vector of surrounding rock mechanical parameters at the location. Indicates the first The time of each discrete time step Indicates the first The time of each discrete time step Indices representing discrete time steps. This represents the time interval between adjacent time steps, satisfying... , This represents the surrounding rock state evolution function, used to characterize the combined influence of current surrounding rock mechanical parameters, excavation disturbance factor, damage variables, stress path, and creep strain on the rate of change of surrounding rock mechanical parameters. Indicates time step Excavation disturbance factor at the location. Indicates time step The surrounding rock damage variable at the location, Indicates time step Vectorized representation of the equivalent stress vector or stress tensor along the stress path of the surrounding rock. Indicates time step Vectorized representation of the creep strain vector or creep strain tensor of the surrounding rock at the location;

[0067] This discrete evolution equation explicitly reflects the time effect in the state update increment. This provides a unified framework for the subsequent introduction of excavation disturbance factors, damage variables, and creep terms;

[0068] Step S22: Combining the tunneling cycle time information recorded in the project time sequence label, the excavation disturbance factor is constructed into a function form of the superposition and decay of multiple tunneling disturbances over time:

[0069] ,

[0070] in, This represents the excavation disturbance factor at continuous time intervals. Represents a continuous-time variable. This indicates that the tunneling cycle numbers are from 1 to... Summation operation, The index representing the tunneling cycle. This represents the total number of tunneling cycles included in the overlay calculation. Indicates the first The initial disturbance amplitude coefficient corresponding to each tunneling cycle can be calibrated based on information such as the advance, blasting parameters, and surrounding rock grade of that cycle. The decay coefficient represents the rate at which the excavation disturbance decays over time, and is used to characterize the rate at which the disturbance's influence decays over time. Indicates the first The start time of each tunneling cycle is obtained from the tunneling record in the project time sequence label. Represents the Heaviside step function, when When the value is 1, The value is 0, which indicates that the disturbance has no effect before tunneling begins;

[0071] This expression uses exponential decay superposition of the disturbance response of different tunneling cycles to enable the excavation disturbance factor to reflect the cumulative impact of multiple tunneling cycles on the surrounding rock condition, and corresponds one-to-one with the tunneling time information in the project time sequence label.

[0072] Step S23: To describe the impact of excavation disturbance and stress redistribution on the evolution of surrounding rock damage, the evolution differential equations of the surrounding rock damage variables are constructed as follows:

[0073] ,

[0074] in, Representing continuous time The time derivative of the surrounding rock damage variable. Indicates time The dimensionless damage variable of the surrounding rock, whose value range can be limited to between 0 and 1, represents the coefficient of influence of excavation disturbance factor on damage evolution, and is used to adjust... Contribution to damage growth, Indicates time The excavation disturbance factor at that location is consistent with the aforementioned definition. A coefficient representing the influence of stress path on damage evolution. Indicates time stress vector of surrounding rock The equivalent stress norm can be the Euclidean norm or the equivalent stress quantity commonly used in engineering. Indicates time The vectorized representation of the stress vector or stress tensor of the surrounding rock, and the state evolution equation The physical meaning is consistent. A coefficient representing the rate of damage recovery or damage evolution attenuation, used to characterize the slowing effect on the rate of damage growth under conditions of reduced disturbance or stress level;

[0075] Through this damage evolution equation, the excavation disturbance factor and stress path jointly drive the growth of damage variables, while the damage attenuation term makes the damage tend to stabilize, thus reflecting the coupling between time effect and excavation disturbance.

[0076] Step S24: After obtaining the damage variables and creep strain, convert them into time variations of surrounding rock mechanical parameters through the softening relationship;

[0077] For example, elastic modulus and cohesion are calculated using the following formula:

[0078] ,

[0079] in, Indicates time The equivalent elastic modulus of the surrounding rock. This represents the initial elastic modulus of the surrounding rock before damage occurs, corresponding to the state where the damage variable is 0. Indicates time The surrounding rock damage variables at the location are consistent with the definitions in the aforementioned damage evolution equation. Indicates time The equivalent cohesion of the surrounding rock. This represents the initial cohesion of the surrounding rock before significant creep and softening occur. This represents the cohesive softening coefficient corresponding to creep, used to measure the sensitivity of creep deformation to the degree of decrease in cohesive force. Indicates time The scalar invariants of creep strain in the surrounding rock can be calculated from the invariants of the creep strain tensor;

[0080] The aforementioned softening relationship allows damage variables and creep strain to be reflected in the surrounding rock state vector through specific mechanical parameters. This forms a time series of surrounding rock mechanical parameters that can be directly output by the time series prediction model;

[0081] Step S25: When the time-varying prediction model for surrounding rock mechanical parameters is implemented using a time-series neural network, the evolution function can be approximated within the network, for example, by using a linear combination form at discrete time steps.

[0082] ,

[0083] in, Represents the vector of mechanical parameters acting on the surrounding rock. The weight matrix, Indicates the disturbance factor acting on the excavation. The weight matrix or weight vector, Indicates the effect on the damage variable The weight matrix or weight vector, Indicates the stress vector acting on The weight matrix, Represents the strain vector acting on the creep. The weight matrix, Represents the bias vector;

[0084] The aforementioned weight matrix and bias vector are used as training parameters and are obtained through supervised learning of structured monitoring data and surrounding rock mechanical parameter sequences.

[0085] To embed the surrounding rock state evolution equation into the network during training, an evolution constraint loss term can be added to the supervised loss, for example:

[0086] ,

[0087] in, This represents the evolution constraint loss value. This indicates that the time step index ranges from 1 to... Summation operation, This represents the number of time steps considered in a single training sample. This indicates that the Euclidean norm operation is performed on the vector.

[0088] By combining the evolution constraint loss term with the supervision loss based on measured or inverted surrounding rock mechanical parameters in a weighted combination, the temporal neural network is constrained by the evolution equation while fitting the monitoring data, thus reflecting the coupled modeling of excavation disturbance factors and time effects in the model structure.

[0089] Specifically, the above implementation scheme provides a detailed description of the evolution scheme that can be embedded in a data-driven model;

[0090] By treating the surrounding rock mechanical indices as discrete-time states and introducing an incremental update form based on time steps, the structure structurally reflects the process of state change over time during construction. The excavation disturbance part adopts a multi-pulse superposition and exponential decay structure based on the tunneling cycle, transforming the cycle start and end times in the engineering time series label into a calculable disturbance curve. The damage evolution part considers both the disturbance effect and the stress path, describing their combined driving effect on deterioration as a time derivative term, and controlling the stability trend of damage over long time scales through the decay coefficient. Subsequently, the softening relationship maps the damage and creep behavior to specific parameters such as elastic modulus and cohesion, providing continuously changing inputs for subsequent simulation and prediction. Finally, the evolution function is explicitly embedded into the temporal neural network structure in the form of network weights, and a constraint loss based on the evolution equation is added, so that the training process takes into account both data fitting ability and physical rationality, thereby enhancing the model's ability to characterize the coupled behavior of excavation disturbance and time effects.

[0091] Similarly, to ensure that the excavation disturbance factor, damage variable, and creep strain have operable values ​​in actual engineering, this embodiment allows for on-site calibration of the correlation coefficients based on the typical geological conditions and construction techniques of the tunnel's location. The initial value of the surrounding rock state vector can be set before excavation based on laboratory rock tests and in-situ testing results. The initial value of the damage variable is typically zero, and the initial value of the creep strain can also be zero or a small initial value given based on pre-construction static observation results, ensuring that the evolution process begins from an undamaged or slightly damaged state. The number of superposition cycles for the excavation disturbance factor can be selected based on the construction organization and the sensitivity of the monitoring data. Typically, the most recent 10-30 tunneling cycles can be used. When the number of cycles is too high, resulting in a much smaller long-term disturbance contribution than the recent disturbance, cycles with longer time intervals can be ignored to simplify the calculation. The coefficients in the evolution equation, such as those characterizing the effects of disturbance, stress path, and damage recovery, can be numerically calibrated by minimizing the error between the model predictions and the actual inversion parameters. Reasonable value ranges can be set according to different surrounding rock grades; for example, coefficients can be controlled within a single order of magnitude at the same surrounding rock grade, and scaled empirically when changing across grades. Optionally, upper and lower limits can be set for damage variables and their derived equivalent elastic modulus, cohesion, and other parameters during numerical implementation. This prevents damage variables from exceeding the physical range of zero to one or equivalent parameters from becoming non-positive due to numerical oscillations or abnormal monitoring data. When parameters are detected to be out of limit, corrections can be made by truncating or reverting to the predicted value of the previous time step, and this time step is recorded as requiring manual verification.

[0092] In one embodiment, the time-varying prediction model for surrounding rock mechanical parameters is a time-series neural network model based on deep learning. The time-series neural network model includes at least a recurrent neural network layer or a self-attention mechanism layer for extracting time correlations, and a regression output layer for outputting surrounding rock mechanical parameters at different time steps.

[0093] In one embodiment, the forward simulation meta-model is obtained by:

[0094] High-fidelity tunnel geotechnical numerical simulation software was used to generate a simulation sample dataset covering different combinations of surrounding rock mechanical parameters, support parameters, and tunnel geometric parameters.

[0095] Using the surrounding rock mechanical parameters, support parameters, and tunnel geometric parameters as input features, and the key indicators of displacement field and stress field in the corresponding simulation results as output labels, a deep neural network model is trained. This enables the deep neural network model to approximate the physical relationship between input and output at a calculation speed far exceeding that of the original numerical simulation software, thus forming a positive simulation meta-model.

[0096] In one embodiment, the method further includes:

[0097] Step S5: After outputting the support parameter adjustment amount, store the support scheme corresponding to the support parameter adjustment amount, as well as the characteristics of the surrounding rock mechanical parameter change curve that triggered the support parameter adjustment amount, into the historical case library.

[0098] Step S4, before invoking the optimization algorithm to iterate the support parameters, also includes:

[0099] Based on the characteristics of the current surrounding rock mechanical parameter variation curve, similarity matching is performed in the historical case library, and the adjustment amount of support parameters in the historical cases obtained by similarity matching is used as the initial iteration point of the optimization algorithm.

[0100] Similarity matching involves calculating the similarity between the target rock mechanical parameter variation curve features and the rock mechanical parameter variation curve features in historical cases, and selecting at least one historical case based on the similarity.

[0101] Before calling the optimization algorithm in step S4, a set of more specific feature construction and similarity calculation sub-steps are introduced in the similarity matching process to calculate the similarity between the target surrounding rock mechanical parameter change curve features and historical case features.

[0102] In one implementation, the target surrounding rock mechanical parameter variation curve is discretized into a multi-time-step multi-parameter matrix and expanded into an eigenvector; for the current target working condition, in the time-step set... Upsampled surrounding rock mechanical parameter vector The parameter dimension is obtained as The number of time steps is Feature matrix and define:

[0103] ,

[0104] in, The characteristic vector representing the curve of change in the mechanical parameters of the target surrounding rock. This represents a vectorization operator used to rearrange a matrix into single-column vectors by columns. The matrix represents the composition of the curves showing the variation of the surrounding rock mechanical parameters under the target working condition, and its i-th Line number Column elements are , Indicates time step First The values ​​of the surrounding rock mechanical parameters, This represents the index of surrounding rock mechanical parameters, with values ​​ranging from 1 to... , Indicates the first The time of each discrete time step is consistent with the meaning of time step in the aforementioned evolution of the surrounding rock state. This indicates the number of surrounding rock mechanical parameters included in the characteristic structure. This indicates the number of discrete time steps considered within a single case;

[0105] For the first in the historical case database For each historical case, the matrix can be constructed using the same time-step discretization method. And obtain its eigenvectors:

[0106] ,

[0107] in, Indicates the first Feature vectors corresponding to each historical case Indicates the first A matrix composed of historical case rock mechanical parameter variation curves, the meanings of its rows and columns are as follows: Consistent Indicates a historical case index;

[0108] To eliminate differences in the dimensions and scales of different parameters, all feature dimensions are normalized; let the feature vector dimension be... , for the Standardize each feature dimension:

[0109] ,

[0110] in, Indicates the target feature vector at the th Normalization results across all dimensions Indicates the target feature vector at the th The original components in each dimension Indicates the first The first historical case in Normalized feature components in each dimension Indicates the first The first historical case in Original feature components in each dimension This indicates the first case in the historical case library. The sample mean of each feature dimension, This indicates the first case in the historical case library. The sample standard deviation of each feature dimension. This represents the feature dimension index, with values ​​ranging from 1 to... , This represents the total dimension of the feature vector;

[0111] In one implementation, a weighted Euclidean distance metric is used based on the normalized feature vector to measure the target case and the first... Differences between historical cases:

[0112] ,

[0113] in, Indicates the target case and the first The weighted Euclidean distance between historical cases Indicates the first The weight coefficients for each feature dimension are used to adjust the contribution of different surrounding rock mechanical parameters and different time periods to the overall similarity. and Representing the target case and the first The first historical case in Normalized components on each feature dimension This represents summation over all feature dimensions; the meanings of the remaining parameters are the same as described above; the weight coefficients can satisfy... and So as to form a convex combination in the parameter space;

[0114] To facilitate integration with other similarity metrics, the weighted Euclidean distance is converted into a similarity score ranging from 0 to 1, for example:

[0115] ,

[0116] in, The first term obtained based on weighted Euclidean distance is... The similarity score of each historical case, with a value ranging from 0 to 1. This represents a bandwidth parameter related to the feature space scale, used to control the rate of decay from distance to similarity. Consistent with the aforementioned definition;

[0117] In one implementation, a dynamic time regularization metric is introduced for the surrounding rock response with significant differences in time morphology, in order to capture the similarity of curve shape under time axis misalignment.

[0118] Taking the equivalent elastic modulus of the surrounding rock as an example, the construction target case and the first Scalar time series of historical cases:

[0119] ,

[0120] in, Indicates the target case at time step The scalar characteristic value at that point is taken here as the equivalent elastic modulus of the surrounding rock. Indicates the target case at time step The equivalent elastic modulus at that point is the same as that mentioned above. Consistent Indicates the first A historical case in time step scalar eigenvalue at that location Indicates the first A historical case in time step The equivalent elastic modulus at that point, This represents the time step index of historical cases. Indicates the first The time step Indicates the number of time steps for the target case. Indicates the first Number of time steps in each historical case;

[0121] Define the local cost function and the cumulative cost matrix:

[0122] ,

[0123] in, Indicates the target sequence at time step With the A historical sequence at time step Local cost, This represents the path from the starting point to the node on a dynamically time-warped path. The cumulative value of , and These represent arrivals from three adjacent predecessor nodes. The cumulative cost is used to construct the path with the minimum cumulative cost; the boundary conditions are initialized according to the conventional setting method of dynamic time warping.

[0124] After obtaining the cumulative cost matrix, the dynamic time warping distance is defined as:

[0125] ,

[0126] in, Indicates the target case and the first Dynamic time-normalized distance between historical cases Indicates the distance from the starting point to the end point. The minimum cumulative cost, This represents the path length of the corresponding optimal normalized path, used to normalize the cumulative cost; the dynamic time normalized distance can also be converted into a similarity score:

[0127] ,

[0128] in, This represents the similarity score obtained based on dynamic time warping. This represents the control parameters related to the dynamic time-warped distance scale. Consistent with the aforementioned definition;

[0129] In one implementation, the angular information of the normalized feature vectors can also be used to characterize the directional similarity of the parameter change patterns; the cosine similarity of the normalized feature vectors can be expressed as:

[0130] ,

[0131] in, Indicates the target case and the first The cosine similarity between historical cases ranges from -1 to 1. Indicates by The target normalized feature vector is formed. Indicates by The first of the composition Normalized feature vectors of historical cases, This represents the vector transpose operation. The Euclidean norm of a vector;

[0132] When comprehensively considering amplitude differences, temporal pattern differences, and directional information, a multi-indicator fusion similarity is constructed:

[0133] ,

[0134] in, Indicates the first The overall similarity score of each historical case. , and Let the weighted Euclidean similarity, dynamic time-warped similarity, and cosine similarity be the fusion weight coefficients, respectively, satisfying the following conditions: , , and , , and Each is consistent with the aforementioned definition;

[0135] In practical similarity matching, the overall similarity can be used as a basis. The historical cases are sorted, and the highest-scoring historical cases are selected. The corresponding support parameter adjustment amounts are used as the initial iteration points of the optimization algorithm.

[0136] Specifically, the similarity matching process is refined from three levels: feature composition, normalization, and similarity measurement.

[0137] A rock mechanics parameter matrix is ​​constructed using time steps and parameter dimensions, and then expanded into feature vectors. This allows for a unified representation of different rock mechanics parameters and information from different time periods within the same vector space. Standardization based on historical case statistics is then introduced to mitigate the impact of differences in dimensions and scales, providing a more stable numerical basis for subsequent distance measurements. In similarity calculation, weighted Euclidean distance reflects overall amplitude differences, dynamic time warping characterizes the morphological similarity of curves under time axis scaling or misalignment, and cosine similarity further characterizes the consistency of change pattern direction. Finally, a multi-index fusion method integrates the three similarities into a comprehensive score, which is used to rank and filter historical cases. The support adjustment information corresponding to the cases with the highest similarity is fed back to the optimization step to construct initial iteration points that are closer to engineering experience, thereby shortening the optimization calculation convergence process and improving the rationality of support decisions.

[0138] In this embodiment, the historical case library can be gradually established and expanded during tunnel construction. Initially, it can consist of data from similar projects or test sections. Subsequently, as actual support adjustment schemes and their effects accumulate, new cases are periodically entered into the library, and existing cases undergo necessary quality assessments and labeling updates. During similarity matching, all historical cases can be sorted from high to low based on their comprehensive similarity. Several cases with a comprehensive similarity not lower than a preset threshold are selected as candidates. The threshold can be set according to the size of the case library and actual engineering experience, for example, between 0.6 and 0.8. When the number of candidate cases is greater than one, a weighted average of the corresponding support parameter adjustments can be calculated based on the comprehensive similarity, or one to three schemes with the highest similarity can be selected as the initial point of the optimization algorithm, thereby ensuring diversity while also considering convergence speed. Furthermore, when the overall similarity of all cases in the case library is lower than the preset threshold, or when the size of the case library is insufficient to support reliable matching, it can degenerate into a mode that does not use historical cases. That is, it directly uses the parameters corresponding to the current support scheme as the initial iteration point of the optimization algorithm, and records the situation in the system where there are no reliable similar cases for this optimization, so as to prompt the designers to conduct a more careful review of the results.

[0139] In one embodiment, the optimization algorithm in step S4 is a heuristic global optimization algorithm and / or a gradient-based local optimization algorithm. The optimization algorithm searches for solutions that satisfy the following constraints within the preset search space of the support parameters:

[0140] The predicted displacement and predicted stress-related indicators are all within the corresponding preset safety range.

[0141] This application proposes a tunnel geotechnical structure analysis system, combined with... Figure 2 As shown, the system includes:

[0142] The data acquisition and processing module is used to acquire time-series monitoring data and engineering time-series labels at the tunnel construction site, and to align the time-series monitoring data with the time axis and associate it with the engineering time-series labels to generate a structured monitoring data sequence.

[0143] The parameter time-varying prediction module integrates a time-varying prediction model for surrounding rock mechanical parameters. It is used to receive structured monitoring data sequences and predict the change curve of surrounding rock mechanical parameters within a preset time period based on the coupling modeling of excavation disturbance factors and time effects.

[0144] The support decision optimization module integrates a forward simulation meta-model and an optimization algorithm. It is used to calculate and predict the mechanical response by iteratively calling the forward simulation meta-model based on the change curve of the surrounding rock mechanical parameters and the current support scheme, and to perform optimization calculations within the support parameter search space, and output the adjustment amount of the support parameters.

[0145] Optionally, during the support decision optimization process, upper and lower bounds for the search of support parameters can be set around the current design scheme. For example, the adjustment range of parameters such as shotcrete thickness, steel arch spacing, anchor length and spacing can be limited to 20% to 50% of the design value, while simultaneously meeting the requirements of the current tunnel design specifications regarding minimum component size and maximum spacing. The number of iterations of the optimization algorithm can be set according to the calculation speed of the forward simulation meta-model and the on-site requirements for response time. Typically, the maximum number of iterations can be controlled between 50 and 200. When the improvement of the objective function is lower than a preset threshold for several consecutive iterations, the iteration can be terminated early, and the current optimal support parameter adjustment amount is taken as the output result. In this embodiment, to ensure the feasibility of the obtained scheme, an engineering feasibility check is performed on the candidate support parameters in each iteration. When a candidate scheme is detected to cause the support parameters to exceed the allowable range of the specifications, it is directly judged as an infeasible solution and eliminated during the optimization process. Furthermore, if no support scheme that simultaneously makes all predicted displacements and predicted stresses fall within the safe range is found within the maximum number of iterations, the candidate scheme that minimizes the objective function value and its degree of exceeding the limit can be output. The scheme is marked as requiring manual review in the results, and the designer makes a comprehensive judgment based on field experience and safety reserve coefficient.

[0146] In one embodiment, the tunnel geotechnical structure analysis system further includes a case library management module for storing and managing a historical case library, which includes at least the characteristics of different surrounding rock mechanical parameter variation curves and their corresponding support parameter adjustment amounts.

[0147] The case library management module is also used to perform similar matching in the historical case library based on the characteristics of the current surrounding rock mechanical parameter change curve before the support decision optimization module starts the optimization algorithm, so as to provide the support decision optimization module with the initial iteration point for the adjustment of support parameters;

[0148] In one embodiment, the data acquisition and processing module is also used to detect outliers and handle missing values ​​in the raw monitoring data, and to organize the processed monitoring data and engineering time series labels into multivariate time series data with a uniform time step and input them into the parameter time-varying prediction module.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0150] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for analyzing the rock and soil structure of tunnels, characterized in that, include: Step S1: Obtain the time-series monitoring data and corresponding engineering time-series labels at the tunnel construction site. The time-series monitoring data includes at least displacement data and stress data of the surrounding rock and support structure. The engineering time-series labels include at least time information and process stage information related to the tunnel excavation cycle and support construction. Step S2: Input the time-series monitoring data and engineering time-series labels into the pre-trained time-varying prediction model of surrounding rock mechanical parameters. The time-varying prediction model of surrounding rock mechanical parameters is constructed based on the modeling of the evolution of surrounding rock state coupled with excavation disturbance factors and time effects, and outputs the change curve of surrounding rock mechanical parameters in the future preset time period. Step S3: Based on the change curve of the surrounding rock mechanical parameters and the current support scheme, input the surrounding rock mechanical parameters, support parameters and tunnel geometric parameters into the pre-trained forward simulation meta-model, and use the forward simulation meta-model to predict the mechanical response of the tunnel structure in the future preset time period; Step S4: Using the support parameters as optimization variables, with the goal of keeping the predicted mechanical response within a preset safe range, the optimization algorithm is invoked to iteratively adjust the support parameters, and the predicted mechanical response of each iteration is evaluated using the forward simulation meta-model until the iteration termination condition is met, and the adjustment amount of the support parameters is output.

2. The tunnel geotechnical structure analysis method as described in claim 1, characterized in that, Step S1 includes: Step S11: Assign an engineering time sequence label to each monitoring data point. The engineering time sequence label includes a timestamp and a process stage identifier. The process stage identifier at least distinguishes between the tunneling cycle stage and the support construction stage. Step S12: Based on the timestamp, the time-series monitoring data is aligned along the time axis to synchronize data from different measuring points and different monitoring types; Step S13: Associate the aligned time-series monitoring data with the corresponding engineering time-series labels to form a structured monitoring data sequence with time-process labels.

3. The tunnel rock and soil structure analysis method as described in claim 2, characterized in that, The methods for obtaining the time-varying prediction model of the surrounding rock mechanical parameters include: Using the structured monitoring data sequence as input features, and the sequence of surrounding rock mechanical parameters obtained through inversion analysis and / or field tests within the same engineering phase as supervised learning labels, the time series prediction model is trained so that it learns the mapping relationship from the monitoring data sequence to the sequence of changes in surrounding rock mechanical parameters. The time-series prediction model extracts time-series features from the structured monitoring data sequence during training and performs coupled modeling of excavation disturbance factors and time effects.

4. A method for analyzing the geotechnical structure of a tunnel as described in claim 1 or 3, characterized in that, The time-varying prediction model for surrounding rock mechanical parameters is a time-series neural network model based on deep learning. The time-series neural network model includes at least a recurrent neural network layer or a self-attention mechanism layer for extracting time correlations, and a regression output layer for outputting surrounding rock mechanical parameters at different time steps.

5. The tunnel geotechnical structure analysis method as described in claim 1, characterized in that, The methods for obtaining the forward simulation meta-model include: High-fidelity tunnel geotechnical numerical simulation software was used to generate a simulation sample dataset covering different combinations of surrounding rock mechanical parameters, support parameters, and tunnel geometric parameters. Using surrounding rock mechanical parameters, support parameters, and tunnel geometric parameters as input features, and the key indicators of displacement field and stress field in the corresponding simulation results as output labels, a deep neural network model is trained, enabling the deep neural network model to approximate the physical relationship between input and output at a calculation speed far exceeding that of the original numerical simulation software.

6. The tunnel geotechnical structure analysis method as described in claim 1, characterized in that, The method further includes: Step S5: After outputting the support parameter adjustment amount, store the support scheme corresponding to the support parameter adjustment amount, as well as the characteristics of the surrounding rock mechanical parameter change curve that triggered the support parameter adjustment amount, into the historical case library. Specifically, step S4, before invoking the optimization algorithm to iterate the support parameters, further includes: Based on the characteristics of the current surrounding rock mechanical parameter variation curve, similarity matching is performed in the historical case library, and the adjustment amount of support parameters in the historical cases obtained by similarity matching is used as the initial iteration point of the optimization algorithm; The similarity matching includes calculating the similarity between the target rock mechanical parameter variation curve features and the rock mechanical parameter variation curve features in historical cases, and selecting at least one historical case based on the similarity.

7. The tunnel geotechnical structure analysis method as described in claim 1, characterized in that, The optimization algorithm in step S4 is a heuristic global optimization algorithm and / or a gradient-based local optimization algorithm. The optimization algorithm searches for solutions that satisfy the following constraints within the preset search space of the support parameters: The predicted displacement and predicted stress-related indicators are all within the corresponding preset safety range.

8. A tunnel geotechnical structure analysis system, used to implement the tunnel geotechnical structure analysis method according to any one of claims 1-7, characterized in that, include: The data acquisition and processing module is used to acquire time-series monitoring data and engineering time-series labels at the tunnel construction site, and to align the time-series monitoring data with the time axis and associate it with the engineering time-series labels to generate a structured monitoring data sequence. The parameter time-varying prediction module integrates a time-varying prediction model for surrounding rock mechanical parameters. It is used to receive the structured monitoring data sequence and predict the change curve of surrounding rock mechanical parameters within a future preset period based on the coupling modeling of excavation disturbance factors and time effects. The support decision optimization module integrates a forward simulation meta-model and an optimization algorithm. It is used to calculate and predict the mechanical response by iteratively calling the forward simulation meta-model based on the change curve of the surrounding rock mechanical parameters and the support parameters corresponding to the current support scheme, and to perform optimization calculations within the support parameter search space, and output the adjustment amount of the support parameters.

9. The tunnel geotechnical structure analysis system as described in claim 8, characterized in that, The tunnel geotechnical structure analysis system also includes a case library management module for storing and managing historical case libraries. The historical case library includes at least the variation curve characteristics of different surrounding rock mechanical parameters and their corresponding adjustment amounts of support parameters. The case library management module is also used to perform similar matching in the historical case library based on the characteristics of the current surrounding rock mechanical parameter change curve before the support decision optimization module starts the optimization algorithm, so as to provide the support decision optimization module with the initial iteration point for the adjustment of support parameters.

10. A tunnel geotechnical structure analysis system as described in claim 8 or 9, characterized in that, The data acquisition and processing module is also used to detect outliers and handle missing values ​​in the raw monitoring data, and to organize the processed monitoring data and engineering time series labels into multivariate time series data with a unified time step and input them into the parameter time-varying prediction module.

Images