Tunnel surrounding rock blasting excavation parameter setting method and system based on neural network

Abstract

The invention provides a tunnel surrounding rock blasting excavation parameter setting method and system based on a neural network, and relates to the technical field of tunnel construction, and the method comprises the steps: obtaining geological characteristic data, blasting excavation historical records and excavation contour data of surrounding rock; sparse subspace clustering processing is carried out according to blasting excavation historical records to obtain a feature data set; performing stress state analysis according to the geological characteristic data, the excavation contour data and the characteristic data set to obtain a dynamic stress state and topological change data of the surrounding rock; performing association rule mining according to the blasting excavation historical record and the feature data set to obtain an optimized decision rule; performing nonlinear finite element analysis according to the dynamic stress state and the topological change data to obtain an equivalent material model of the surrounding rock; and according to the equivalent material model and the optimization decision rule, a blasting excavation parameter setting scheme is obtained. According to the invention, dynamic response analysis is carried out by using the neural network, and real-time monitoring of the surrounding rock under different blasting conditions is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel construction, and in particular to a method and system for setting tunnel surrounding rock blasting excavation parameters based on a neural network. Background Art

[0002] In recent years, with the continuous upgrading of transportation network construction, there are more and more tunnel projects across complex geological conditions such as mountainous areas and seabeds, and the role of tunnel construction in various infrastructure construction has become increasingly prominent. Faced with increasingly complex geological conditions and stringent engineering requirements, the safety and efficiency of tunnel excavation have become key issues that need to be urgently addressed in engineering design and construction. Blasting excavation of surrounding rock is one of the core links in tunnel construction, and the reasonable setting of its parameters is of great significance to ensure the stability of surrounding rock, reduce construction deformation, and control construction costs.

[0003] In the existing technology, the setting of surrounding rock blasting excavation parameters mainly relies on traditional manual analysis and static mechanical models. Specifically, engineers collect geological property data of the surrounding rock and previous blasting history records, use simple static stress models to calculate the blasting effect, and then set blasting parameters such as blasthole arrangement, explosive dosage, and detonation sequence. However, the existing technology still has significant shortcomings. First, the traditional static model cannot effectively capture the nonlinear characteristics of the surrounding rock under the action of blasting loads, resulting in inaccurate predictions of the dynamic behavior of the surrounding rock. Secondly, the existing technology relies on empirical methods and lacks scientific quantitative analysis. It usually only considers the initial state of the surrounding rock, but fails to monitor and adjust the dynamic changes during the construction process in real time.

[0004] In view of the above problems, the present invention proposes a tunnel surrounding rock blasting excavation parameter setting method and system based on neural network, aiming to optimize the blasting excavation parameter setting by real-time analysis of the surrounding rock dynamic response, thereby improving the safety and efficiency of tunnel construction. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for setting tunnel surrounding rock blasting excavation parameters based on neural network to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present application provides a method for setting tunnel surrounding rock blasting excavation parameters based on a neural network, comprising:

[0007] Obtain geological characteristics data of surrounding rock, blasting excavation history and excavation profile data;

[0008] Performing sparse subspace clustering processing according to the blasting excavation historical records, and obtaining a feature data set by performing sparse subspace decomposition on surrounding rock characteristics and blasting effects in the historical data;

[0009] Performing stress state analysis according to the geological characteristic data, the excavation profile data and the characteristic data set, modeling the stress state and spatial variation of the surrounding rock using a time-dependent topological analysis method by simulating the stress variation process of the surrounding rock at different time points, and obtaining dynamic stress state and topological variation data of the surrounding rock;

[0010] Performing association rule mining based on the blasting excavation history records and the feature data set, and obtaining an optimization decision rule by mining the intrinsic association between the excavation sequence and the blasthole layout plan;

[0011] Performing nonlinear finite element analysis according to the dynamic stress state and the topological change data, and obtaining an equivalent material model of the surrounding rock by establishing a mechanical behavior model of the surrounding rock and performing numerical calculations;

[0012] According to the equivalent material model and the optimization decision rule, after being processed by a preset viscoelastic fluid model, a blasting and excavation parameter setting scheme is obtained by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting. The blasting and excavation parameter setting scheme includes excavation steps, blasthole arrangement scheme and detonation sequence.

[0013] In a second aspect, the present application also provides a tunnel surrounding rock blasting excavation parameter setting system based on a neural network, comprising:

[0014] An acquisition module is used to acquire geological property data of surrounding rocks, blasting excavation history records and excavation profile data;

[0015] A clustering module is used to perform sparse subspace clustering processing according to the blasting excavation historical records, and obtain a feature data set by performing sparse subspace decomposition on surrounding rock characteristics and blasting effects in the historical data;

[0016] An analysis module is used to perform stress state analysis based on the geological characteristic data, the excavation profile data and the characteristic data set, and to model the stress state and spatial variation of the surrounding rock using a time-dependent topological analysis method by simulating the stress variation process of the surrounding rock at different time points, so as to obtain dynamic stress state and topological variation data of the surrounding rock;

[0017] A mining module is used to mine association rules based on the blasting excavation history records and the feature data set, and obtain an optimization decision rule by mining the intrinsic association between the excavation sequence and the blasthole layout plan;

[0018] A construction module is used to perform nonlinear finite element analysis according to the dynamic stress state and the topological change data, and obtain an equivalent material model of the surrounding rock by establishing a mechanical behavior model of the surrounding rock and performing numerical calculations;

[0019] The output module is used to obtain a blasting and excavation parameter setting scheme based on the equivalent material model and the optimization decision rule, after being processed by a preset viscoelastic fluid model, by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting. The blasting and excavation parameter setting scheme includes an excavation sequence, a blasthole arrangement scheme, and a detonation sequence.

[0020] The beneficial effects of the present invention are:

[0021] The present invention acquires the geological characteristic data of the surrounding rock, the blasting excavation history records and the excavation profile data in real time, and uses a neural network to perform dynamic response analysis to achieve real-time monitoring of the surrounding rock under different blasting conditions. By performing nonlinear finite element analysis on the dynamic stress state of the surrounding rock, a more realistic surrounding rock mechanical behavior model is established, thereby being able to effectively evaluate and predict the stress changes and potential risks of the surrounding rock during the excavation process.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 It is a schematic flow chart of a method for setting tunnel surrounding rock blasting excavation parameters based on a neural network according to an embodiment of the present invention;

[0025] Figure 2 Flow chart of the density-based clustering algorithm described in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0028] Embodiment 1:

[0029] This embodiment provides a method for setting tunnel surrounding rock blasting excavation parameters based on a neural network.

[0030] See also Figure 1 , the figure shows that the method includes steps S100 to S600.

[0031] Step S100, obtaining geological property data of surrounding rocks, blasting excavation history records and excavation profile data;

[0032] It is understandable that geological property data include information such as rock type, strength, crack distribution, porosity, and groundwater level of the surrounding rock. Detailed surrounding rock information can be obtained by using geological exploration techniques (such as drilling, geological radar, and seismic reflection method) to provide support for subsequent dynamic analysis. At the same time, blasting excavation history records include blasting parameters and effect feedback in past construction, which can provide valuable reference for current projects and help identify past successes and failures. The integration of these data can provide a historical basis for the dynamic response of the surrounding rock, making subsequent analysis more targeted and scientific. In addition, the excavation profile data describes the geometry and size of the tunnel design, providing guidance for actual operations during the construction process.

[0033] Step S200, performing sparse subspace clustering processing according to the blasting excavation historical records, and obtaining a feature data set by performing sparse subspace decomposition on the surrounding rock characteristics and blasting effects in the historical data;

[0034] It should be explained that the basic idea of ​​sparse subspace clustering is to find the intrinsic relationship between surrounding rock characteristics and blasting effect by mapping high-dimensional data to low-dimensional space. In this process, the algorithm models the relationship between surrounding rock parameters (such as rock type, strength, porosity, etc.) in historical records and corresponding blasting effects (such as excavation quality, surrounding rock stability, excavation efficiency, etc.). Specifically, sparse subspace decomposition can effectively separate historical records with similar characteristics and cluster them into several subspaces, each of which reflects the changes in surrounding rock characteristics under specific conditions and their impact on blasting effect.

[0035] Step S300, performing stress state analysis based on geological characteristic data, excavation profile data and characteristic data set, simulating the stress change process of surrounding rock at different time points, using time-dependent topological analysis method to model the stress state and spatial change of surrounding rock, and obtaining dynamic stress state and topological change data of surrounding rock;

[0036] It can be understood that this step uses a time-dependent topological analysis method, which can effectively capture the stress state of the surrounding rock under dynamic loading and its spatial variation. This method first uses the finite volume method or the finite element method to spatially discretize the surrounding rock to generate a high-fidelity numerical model. This model can reflect the stress distribution and deformation of the surrounding rock at various time points. Then, through the time-stepping method, the blasting load is applied to the model, and the simulation of stress redistribution is gradually promoted to obtain the stress state of the surrounding rock at various times during the excavation process. Using time-dependent topological analysis, especially persistent homology technology, stress concentration areas and newly formed fractures that change with time in the stress field can be identified. This process not only reveals the dynamic stress state of the surrounding rock at different time points, but also provides information on the overall stability of the surrounding rock and potential unstable areas.

[0037] Step S400: mining association rules based on blasting excavation history records and feature data sets, and obtaining optimization decision rules by mining the intrinsic association between excavation sequence and blasthole layout scheme;

[0038] It can be understood that, firstly, the stress evolution process of different sequences is converted into a time series matrix through characteristic sequence analysis, so as to compare and identify patterns. Dynamic time warping can handle time series of different lengths and asynchronous time series, match similar stress change processes, and thus reveal the potential similarities between the steps. After obtaining the stress evolution sequence, the clustering algorithm is used to classify the steps with similar stress change characteristics to identify the typical stress patterns under different excavation conditions. Then, the clustering results are quantified through Bayesian network modeling to reveal the conditional dependency and probability distribution between the excavation sequence and the blasthole layout, so as to describe the influence and success probability of each blasthole layout scheme under specific conditions. This network model represents the interaction between complex variables through the relationship between nodes and edges, so that the system can quantify the effect of each layout scheme on stress evolution and help formulate a scientific and reasonable combination of excavation sequence and blasthole layout scheme.

[0039] Step S500, performing nonlinear finite element analysis according to the dynamic stress state and topological change data, establishing a mechanical behavior model of the surrounding rock and performing numerical calculations to obtain an equivalent material model of the surrounding rock;

[0040] It should be noted that this process can capture the dynamic changes of the surrounding rock gradually releasing stress and redistributing it after blasting and excavation, and reflect the nonlinear behavior of the material. Finally, the equivalent material model of the surrounding rock is constructed through the inversion of the numerical analysis results. This equivalent model can not only simplify subsequent calculations, but also comprehensively express the overall mechanical properties of the surrounding rock, providing a more concise and reliable basis for the parameter setting of subsequent construction.

[0041] Step S600, according to the equivalent material model and the optimization decision rule, after being processed by the preset viscoelastic fluid model, by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting, a blasting excavation parameter setting scheme is obtained, and the blasting excavation parameter setting scheme includes an excavation sequence, a blasthole arrangement scheme, and a detonation sequence.

[0042] It can be understood that the viscoelastic fluid model is used to capture the stress release and deformation process of the surrounding rock in different blasting rounds. This model can simultaneously consider the elastic recovery capacity and plastic deformation accumulation of the surrounding rock, and simulate the time-dependent behavior in the actual excavation process by abstracting the deformation characteristics of the rock mass into a viscoelastic body. Based on this model, the stress wave propagation, rock mass rupture and stress release process generated by each round of blasting are simulated step by step, and the dynamic response of the surrounding rock under different hole layout and detonation sequences is calculated. Finally, the blasting excavation parameter setting scheme generated by the simulation process includes a detailed excavation sequence, blasthole layout scheme and detonation sequence. The selection of each parameter combination has been simulated, evaluated and optimized to ensure that the scheme can maximize the control of surrounding rock deformation, ensure construction safety, and reduce the number of blasting and construction costs. This setting scheme provides specific guidance for on-site construction, making stress management during construction more scientific and efficient.

[0043] Further, step S200 includes step S210 to step S240.

[0044] Step S210: Perform data dimensionality reduction processing according to the blasting excavation history records, extract surrounding rock characteristic data and corresponding blasting effect data through principal component analysis, and obtain a low-dimensional feature set;

[0045] It is understandable that in the process of blasting excavation, historical data contains a large amount of multidimensional information, such as geological characteristics of surrounding rocks (lithology, strength, porosity, etc.), blasting parameters (blasthole arrangement, explosive quantity, detonation sequence, etc.) and construction effects (surrounding rock stability, deformation, excavation efficiency, etc.). These data are multi-dimensional and have certain redundancy. Direct analysis and processing will lead to excessive computational complexity and difficulty in identifying effective information. Therefore, it is necessary to simplify the data structure through dimensionality reduction. Principal component analysis (PCA) is a dimensionality reduction method that maps the original high-dimensional data to a low-dimensional space through linear transformation and ensures that the variance information of the data is retained to the greatest extent. In the specific implementation, the covariance matrix of the original data is first calculated to analyze the correlation between different dimensions. Then, through eigenvalue decomposition, several principal components are extracted from the data. Each principal component is a linear combination of the dimensions of the original data and is sorted according to the size of the variance explained. The principal components with larger variance retain the main characteristics of the data, while the components with smaller variance represent redundant information and can be discarded. Through principal component analysis, a low-dimensional feature set can be extracted from complex historical records, which covers the main variation patterns between surrounding rock properties and blasting effects.

[0046] Step S220, performing sparse coding processing according to the low-dimensional feature set, performing sparse representation on the features through Lasso regression, and obtaining a sparse feature matrix;

[0047] It should be noted that Lasso regression is a regression model with L1 regularization, which selects features while performing regression analysis. Specifically, Lasso regression introduces L1 norm constraints in the loss function to make some regression coefficients tend to zero, thereby automatically screening out important features and ignoring features that have little impact on the results. In this step, Lasso regression is used to identify and retain variables that have a key impact on the blasting effect. The process of sparse coding processing includes: first, a regression model is constructed based on the data in the low-dimensional feature set to predict the contribution of each feature in the blasting effect; then, by introducing L1 regularization restrictions, the model compresses the coefficients of certain features to zero during the optimization process, retaining only the most influential features. Finally, these screened features are organized into a sparse feature matrix, in which each column in the matrix represents an important feature, and most irrelevant features are eliminated or set to zero.

[0048] Step S230, performing similarity measurement according to the sparse feature matrix, calculating the similarity between different historical blasting records by cosine similarity, and obtaining a similarity matrix;

[0049] In this step, each pair of burst records in the sparse feature matrix is ​​compared as a vector and the cosine similarity between them is calculated. Since these features are sparse after Lasso regression, the calculation process is more efficient and the results are more interpretable. The similarity calculation results between all pairs of records will be organized into a similarity matrix, where each element of the matrix represents the cosine similarity between a pair of records.

[0050] Step S240: Perform sparse subspace clustering processing according to the similarity matrix, and divide the blasting records into at least two related subspaces by using a sparse subspace clustering algorithm to obtain a feature data set.

[0051] In the specific implementation, the goal of sparse subspace clustering is to divide records into different subspaces so that the records in each subspace have high correlation in certain specific dimensions. The algorithm first constructs an adjacency graph between data based on the similarity matrix generated in the previous step. In this graph, records with high similarity are connected by edges with large weights. Then, the sparse subspace clustering algorithm performs a sparse representation on these records, that is, it assumes that each record can be represented by a linear combination of a small number of other records in the same subspace. By optimizing the sparsity of this linear combination, the algorithm can automatically discover the different subspaces to which these records belong.

[0052] The key to this process is to capture the local characteristics in multidimensional data: even if some records are not very similar in all dimensions, they are strongly correlated in specific dimensions or conditions (such as specific geological structures, blasthole layouts, or blasting effects). Sparse subspace clustering algorithms can identify these local patterns and divide records into subspaces related to these features. Finally, a feature dataset is generated based on the clustering results, where each subspace represents a class of blasting records with similar characteristics.

[0053] Further, step S300 includes step S310 to step S340.

[0054] Step S310: Perform spatial discretization processing according to the geological characteristic data and the excavation profile data, and grid-discretize the tunnel surrounding rock using the finite volume method to obtain a spatial discretization model;

[0055] Preferably, the finite volume method (FVM) is used in this embodiment to grid and discretize the surrounding rock. The core idea of ​​the finite volume method is to divide the solution area into multiple finite control volumes (i.e., grid cells) and perform calculations through the conservation laws within these cells. In the scenario of tunnel construction, geological property data (such as rock type, density, porosity, etc.) and excavation profile data (geometric size and shape of the tunnel) jointly determine the grid division method. Important structural details and stress concentration areas in the region will use finer grid division to ensure the accuracy of the calculation, while coarser grids can be used in relatively stable areas to improve calculation efficiency.

[0056] The advantage of the finite volume method is that it ensures the conservation of mass, momentum and energy by performing flux integration on the boundaries of each control volume. This method shows high adaptability when simulating complex geometric structures (such as irregular tunnel sections) and inhomogeneous materials (such as combinations of different types of surrounding rocks). In addition, the finite volume method can better capture the changes in local stress concentrations or crack areas, and is particularly effective for dynamic stress redistribution problems that occur during tunnel construction.

[0057] Step S320, calculating the initial stress field according to the spatial discrete model, calculating the surrounding rock deadweight and formation pressure and using Gaussian numerical integration to calculate stress, and obtaining the initial stress distribution state of the surrounding rock;

[0058] It should be noted that this step calculates the initial stress field of the tunnel surrounding rock based on the spatial discrete model. The stress field is formed by the combined effect of the surrounding rock's deadweight and formation pressure, reflecting the natural equilibrium state of the surrounding rock before construction. The deadweight stress of the surrounding rock increases linearly with increasing depth. The specific calculation depends on the density, gravity acceleration and depth of the rock mass, while the formation pressure takes into account factors such as geological structure and lateral pressure coefficient, especially in fault zones or high stress areas. Through the Gaussian numerical integration method, the stress value is calculated at the Gaussian integration point of each grid unit, and the sum is performed according to the preset weights, so as to accurately approximate the stress distribution of each unit and generate a complete initial stress distribution state. The stress field represents the stress conditions of each unit in different directions in the form of a tensor, providing reliable initial conditions for stress redistribution simulation in subsequent construction.

[0059] Step S330, performing a dynamic stress change simulation according to the initial stress distribution state, by gradually promoting the stress redistribution of the surrounding rock during the excavation process, and simulating the stress release and redistribution of the surrounding rock due to the removal of the rock mass at different time points during the excavation process, to obtain time series stress distribution data;

[0060] It should be noted that as the tunnel is gradually excavated, the stress state of the surrounding rock will be redistributed. Therefore, it is necessary to simulate the evolution of stress step by step to predict the stress release and redistribution behavior of the surrounding rock at each stage. In the specific implementation, the simulation is carried out in stages according to the excavation sequence, that is, each excavation will remove a part of the rock mass. This removal behavior leads to the rapid release of local stress and promotes the redistribution of stress in the remaining rock mass.

[0061] Furthermore, the simulation process uses a step-by-step loading and iterative solution method, imposing new boundary conditions in each time step to reflect the area of ​​rock mass that has been removed. At this point, the new mechanical equilibrium state is calculated and the corresponding stress changes are recorded in each grid cell. This process is nonlinear because as the rock mass is continuously removed, stress may be concentrated in certain specific areas, leading to the generation of new cracks or the emergence of potential unstable areas. Therefore, each step of the simulation needs to be iterated until the mechanical equilibrium of each stage converges. In addition, this simulation of dynamic stress changes generates time series stress distribution data, which records the stress state of the surrounding rock at different time points. The time series stress distribution data not only shows how stress changes as construction progresses, but also reveals the stress peak area and potential risk area at critical time points.

[0062] Step S340: Perform time-dependent topological analysis based on the time series stress distribution data, identify changes in stress concentration areas and newly formed fractures in the stress field that occur over time through a persistent homology method, and obtain dynamic stress state and topological change data.

[0063] Furthermore, as excavation progresses, stress is continuously released and redistributed, and stress concentration or new fractures may occur in local areas. These changes directly affect the stability of the tunnel. To identify these dynamic changes, the persistent homology method is introduced, which is a topological data analysis tool that can quantify the changes in topological structure at different time scales.

[0064] In the specific analysis process, the persistent coherence method tracks stress concentration areas in time series data and regards them as key features in the topological structure. For example, as excavation progresses, certain areas may experience stress concentration, crack expansion, or the formation of new fractures. Persistent coherence can record when these phenomena appear and disappear, and quantify their "persistence." The higher the persistence, the greater the impact of the stress concentration area or fracture on the stability of the surrounding rock during the entire construction process. By identifying these topological features, not only can obvious fractures and concentration areas be discovered, but also some hidden but subtle changes that may have an impact in subsequent construction can be captured. The dynamic stress state and topological change data finally generated reveal the key evolution patterns and potential unstable areas in the surrounding rock stress field.

[0065] Further, step S400 includes step S410 to step S440.

[0066] Step S410: Perform feature sequence analysis based on the blasting excavation history records and feature data sets, compare the surrounding rock stress evolution patterns under different step sequence conditions through dynamic time warping, and obtain the surrounding rock stress sequence matrix under different excavation step sequence conditions;

[0067] It should be noted that dynamic time warping is an algorithm for nonlinear matching of time series data, which can effectively process stress evolution sequences of different lengths and time steps. In blasting excavation, different excavation sequence will cause the surrounding rock stress to be released or concentrated at different time points, and the patterns of these stress changes may not be completely synchronized in time. Dynamic time warping aligns these time series and finds the optimal matching path between them. Even if there is a lag or asynchrony in time, it can accurately compare the similarity of stress evolution.

[0068] Specifically, the stress evolution data under each step condition is first extracted from the blasting history records and combined with the relevant features in the feature data set to form an initial time series data set. Then, dynamic time warping compares these sequences one by one and calculates the similarity of stress evolution between different sequences. This process not only identifies which step condition has the most similar stress change pattern, but also reveals which steps may cause abnormal stress concentration under specific conditions. Finally, the comparison results are integrated into a surrounding rock stress sequence matrix, in which each row represents an excavation step and each column corresponds to the stress state at a different time step. This matrix clearly shows the impact of different sequences on the evolution of surrounding rock stress, providing a quantitative basis for subsequent analysis.

[0069] Step S420, clustering stress patterns according to the surrounding rock stress sequence matrix, and obtaining clustering results of excavation sequence and stress evolution by clustering step combinations with similar stress evolution characteristics;

[0070] Preferably, this step adopts an improved density-based clustering algorithm. Specifically, first, the stress sequence matrix generated by the historical blasting data and the characteristic data set is directly used to analyze the changes in stress data under different step conditions. The improved density clustering method does not need to preset the number of clusters, but automatically discovers clusters based on the density of data points. The stress evolution characteristics corresponding to each excavation step are regarded as a data point, and whether it belongs to a high-density area is judged based on the number of its neighboring points. When a data point of a step is located in a high-density area, it will be regarded as a core point and classified into a cluster together with other points in its neighborhood. On the contrary, if there are not enough neighboring points near a point (that is, the density is not high enough), the point is marked as a noise point and removed from the cluster. This feature is particularly suitable for processing complex stress evolution data during excavation, where some steps may show abnormal stress concentration patterns, while other steps show smoother stress distributions.

[0071] like Figure 2 As shown, Figure 2 This is a flowchart of the density-based clustering algorithm. When applying this method, a neighborhood radius parameter (∈) and a minimum number of neighborhood points (MinPts) are first defined for each data point to control the density requirement of clustering. By traversing the data points step by step, high-density areas are found, and points that meet the density requirements are classified into the same cluster. The density-based clustering algorithm can not only identify those steps with similar stress characteristics, but also filter out the steps with abnormal stress evolution behavior, thereby improving the stability and accuracy of the clustering results. Through this density-driven clustering, the clustering results of each excavation step and its corresponding stress evolution pattern are finally generated.

[0072] Step S430, analyzing the association rules of blasthole layout according to the clustering results, modeling the probability dependency relationship between the excavation sequence and the blasthole layout through the Bayesian network, and quantifying the influence of each blasthole layout on the stress evolution of a specific sequence, to obtain the probability association rules;

[0073] It can be understood that in the Bayesian network, each node represents a random variable, the step node describes different stress evolution characteristics, and the blasthole layout node covers the design parameters such as the location, spacing, depth and detonation sequence of the blasthole. Specifically, in this step, the structure learning and parameter learning are first performed through historical data, the network structure is automatically generated and the conditional probability table (CPT) is calculated to quantify the probability distribution of different layout schemes under a specific step sequence. Then, the probability inference ability of the Bayesian network is used to infer the optimal layout scheme and its corresponding stress evolution effect according to the given step sequence, thereby generating practical probabilistic association rules. These rules not only reveal the dependency between the step sequence and the blasthole layout, but also quantify the stress control effect of different schemes under each step sequence and their comprehensive impact on blasting efficiency, stability and cost.

[0074] Step S440, multi-objective optimization processing is performed according to the probability association rule, and the stress evolution and surrounding rock stability are simulated and evaluated by simulating the combination of different blasthole schemes and step sequences, and the stability and economy of the tunnel surrounding rock are used as the reward function for screening to obtain the optimization decision rule.

[0075] It is understandable that, first of all, according to the probability association rules, a variety of combinations of blasthole layouts and sequence are constructed, and a detailed simulation evaluation is performed for each combination. The simulation process covers the stress change pattern under different geological conditions, the evolution of stress concentration areas, and the expansion of cracks to ensure the adaptability of the scheme in various scenarios. The key evaluation contents include the stress release efficiency of the surrounding rock, the local stress concentration control effect, and the stability change to quantify the mechanical performance of each scheme.

[0076] In the multi-objective optimization process, the stability of the surrounding rock and the economic efficiency of construction are used as the main reward functions. The stability index focuses on evaluating the control ability of the scheme on the deformation and stress redistribution of the surrounding rock to avoid instability and sudden fracture; while the economic index measures the construction cost of each combination, including factors such as the amount of explosives, construction progress, manpower and equipment costs. In actual optimization, these two goals often conflict, so the multi-objective optimization algorithm is used to find a balance between the two to ensure that the construction plan is both safe and stable and has cost advantages.

[0077] The final output optimization decision rules do not directly give the parameters of each specific blasthole, but provide a set of guiding design frameworks. The final generated optimization decision rules are expressed in mathematical form to provide guidance for construction under different geological conditions. These rules form a scientific and adaptable design framework, which provides theoretical support and optimization paths for specific parameter setting schemes, thereby ensuring that construction achieves cost control and improves construction efficiency while meeting safety requirements.

[0078] Further, step S500 includes step S510 to step S540.

[0079] Step S510: fitting nonlinear material property parameters according to the dynamic stress state and topological change data, optimizing and fitting the nonlinear constitutive relationship of the surrounding rock material by means of a genetic algorithm, and obtaining nonlinear material parameters of the surrounding rock;

[0080] It is understandable that the nonlinear mechanical behavior of the surrounding rock, including plastic deformation, stress relaxation, and crack expansion, will change with stress release and redistribution during the construction process. Therefore, in order to accurately describe these complex behaviors, it is necessary to find the most realistic material parameters through a fitting process. These parameters cannot be obtained through direct measurement, but need to be fitted by combining actual stress data with theoretical models through numerical optimization.

[0081] Specifically, the genetic algorithm simulates the process of natural selection and evolution. It first generates multiple possible parameter combinations as initial solutions and evaluates their fitness by calculating the error between each combination and the actual stress data. The higher the fitness, the more accurately the parameter combination can reflect the actual stress response of the surrounding rock. Then, the algorithm selects the parameter combinations with better performance for crossover and mutation to generate new parameter combinations. Crossover simulates the information exchange between different schemes, while mutation introduces random disturbances to help the algorithm escape the trap of local optimal solutions. Through continuous iteration and step-by-step optimization, the algorithm finds a parameter combination with the minimum error and stability. The resulting nonlinear material parameters can truly reflect the deformation and mechanical properties of the surrounding rock under various stress conditions.

[0082] Step S520, dividing the nonlinear finite element unit grid according to the nonlinear material parameters, optimizing the grid of the spatial discrete model of the surrounding rock through an adaptive grid generation algorithm, and obtaining a finite element grid model containing nonlinear characteristics;

[0083] It should be noted that since the surrounding rock undergoes complex stress redistribution and nonlinear deformation during the blasting process, ordinary uniform meshing cannot accurately capture the details of the key areas. Therefore, this step uses an adaptive mesh generation algorithm to dynamically optimize the mesh according to the characteristics of nonlinear material parameters and the stress concentration of the surrounding rock.

[0084] Specifically, the nonlinear material parameters obtained by fitting are first used to define the mechanical behavior characteristics of each area. Based on these parameters, it is possible to determine which areas may experience stress concentration, plastic deformation, or crack expansion, and to divide finer mesh units in these areas to ensure the calculation accuracy. At the same time, for areas with small stress changes or more stable behaviors, larger mesh units are used to reduce the amount of calculation and improve the simulation efficiency. This meshing method enables the model to ensure accuracy while optimizing computing time and resources. The working mechanism of the adaptive mesh generation algorithm is to optimize the mesh through iterative analysis. After the initial meshing is completed, a preliminary mechanical calculation is performed, and the mesh structure is adjusted according to the calculation results. For areas with high stress concentration or drastic changes, the mesh is automatically refined; for areas with insignificant stress changes, the mesh units are merged or expanded. This adaptive adjustment process continues until the mesh structure meets the requirements of calculation accuracy and efficiency. The final generated finite element mesh model not only contains the geometric structure and material properties of the surrounding rock, but also reflects its nonlinear mechanical behavior during blasting excavation.

[0085] Step S530, performing nonlinear mechanical behavior simulation according to the finite element mesh model, based on the initial stress state of the surrounding rock and the mesh model, gradually and iteratively solving the stress and strain of each unit until a mechanical equilibrium state is reached, and obtaining the stress-strain distribution result of the surrounding rock;

[0086] Specifically, the initial stress state of the surrounding rock is first imported as the initial condition of the model. In the early stage of construction, the stress distribution in each grid cell is set according to the deadweight of the surrounding rock and the formation pressure. As the construction progresses, the stress in the surrounding rock will be gradually released and redistributed after part of the rock mass is removed by blasting. This nonlinear process needs to be simulated by step-by-step iterative solution: in each iteration, the stress and strain of all grid cells are calculated, and these results are matched with the nonlinear constitutive relationship of the material.

[0087] The iterative solution process uses a nonlinear finite element solver and uses a progressive loading method to gradually apply the external force boundary conditions of blasting and excavation. In each round of iteration, the equilibrium state of all units is checked and it is determined whether it meets the convergence conditions. If the mechanical equilibrium between units has not been achieved, the mesh will be adjusted according to the calculated stress and strain results, and the next round of iteration will be carried out. This process will continue until the internal and external forces of all units reach a state of equilibrium, that is, the mechanical convergence criteria are met. The final stress-strain distribution results not only include the stress and strain magnitude of each unit, but also reveal the stress concentration areas, potential instability areas and deformation modes in the entire surrounding rock structure. This step-by-step iterative simulation method can dynamically capture the response behavior of the surrounding rock at different stages of the excavation process, providing reliable support for subsequent blasting and excavation design.

[0088] Step S540: construct an equivalent material model according to the stress-strain distribution result, determine the equivalent material properties of the surrounding rock through an inversion analysis algorithm, and obtain an equivalent material model.

[0089] It should be noted that since the actual material properties of the surrounding rock often show strong nonlinearity, such as local cracks, stress concentration and bedding structure, the direct use of a complete nonlinear model will lead to complex and time-consuming calculations. Therefore, it is necessary to map the complex mechanical properties into a simplified equivalent model through an inversion algorithm. First, the stress-strain data obtained from the nonlinear finite element simulation are imported. These data reflect the stress distribution and deformation mode of the surrounding rock in different regions. Then, the parameters of the equivalent model are adjusted using the inversion analysis algorithm to make its stress-strain response as consistent as possible with the actual results. In each round of iteration, by comparing the simulation results with the actual data, the error is calculated and the material parameters are updated until the error converges to an acceptable range. The final equivalent material model contains simplified key parameters, such as equivalent elastic modulus, Poisson's ratio and shear strength, which can comprehensively describe the average response behavior of the surrounding rock while reducing the computational complexity, making it suitable for subsequent construction analysis and optimization.

[0090] Further, step S600 includes step S610 to step S640.

[0091] Step S610, performing dynamic response analysis according to the equivalent material model, modeling the viscoelastic characteristics of the surrounding rock by applying a piecewise linearization method, and obtaining a dynamic response function;

[0092] It is understandable that since the stress and deformation behavior of the surrounding rock material in different time periods will show complex viscoelastic properties, that is, the stress response of the material is not only related to the current strain state, but also to the past stress history, an efficient method is needed to describe its dynamic changes. Directly using a nonlinear model to simulate the entire process is computationally complex and time-consuming, while the piecewise linearization method simplifies the calculation by dividing the nonlinear viscoelastic behavior into multiple small intervals and approximating it as a linear response in each interval.

[0093] Step S620, using the time domain integration method to simulate the dynamic response of the surrounding rock, by applying the dynamic response function to the time history of each round of blasting, and gradually integrating the time to calculate the stress and deformation at each moment, to obtain a stress-deformation curve;

[0094] Specifically, the dynamic response function obtained in the previous stage is first imported, which describes the stress and strain evolution characteristics of the surrounding rock in different time periods. In order to simulate the dynamic response in the actual construction process, the external force generated by each round of blasting is used as an input signal and applied to the dynamic response model. After each round of blasting, the surrounding rock will experience stress release, redistribution, and deformation accumulation, which change over time. To capture these changes, the time domain integration method decomposes the entire time history into several small time steps, and calculates the current stress and deformation in each time step.

[0095] Furthermore, based on the dynamic response function in the viscoelastic model, the stress state in each time step is integrated, the deformation and stress changes of the previous time step are accumulated, and the results are used as the initial conditions of the next time step. This step-by-step integration method can dynamically capture the propagation path of stress, the deformation mode of the surrounding rock, and the potential risk of instability. As time goes on, the calculation results are continuously superimposed, and finally a comprehensive stress-deformation curve is obtained, which reflects the mechanical evolution of the surrounding rock during the entire construction process. The stress-deformation curve not only shows the deformation amplitude and stress concentration of the surrounding rock at different time points, but also provides a basis for engineers to identify key time points in the blasting process, such as the time when the stress peak occurs and the possible deformation inflection point. The calculation formulas involved include:

[0096]

[0097]

[0098]

[0099]

[0100] (σ,ε)={(σ0,ε0),(σ1,ε1),…,(σ n ,εn )};

[0101] Where t represents the current time point; τ represents the past time point; n represents the sequence number of the discrete time step; σ(t) represents the stress value of the surrounding rock at time t; ε(t) represents the strain of the surrounding rock at time t; E(t-τ) represents the viscoelastic modulus, which describes the elastic and viscous properties of the surrounding rock material within the time difference t-τ; σ0 represents the initial stress state; ε0 represents the initial strain state; C(τ) is the energy dissipation function, which represents the energy lost by the surrounding rock at time τ due to internal friction, plastic deformation or material damage; Δt represents the time step; F ext,n is the external blasting load, which represents the blasting force applied in the nth time step; σ n represents the stress value at the nth time step; σ n+1 represents the stress value at the n+1th time step; ε n is the strain variable at the nth time step; ε n+1 is the strain at the n+1th time step; E n is the dynamic modulus at the nth time step, indicating the elastic recovery capacity of the surrounding rock in this time step, and is used to reflect the transient response of the material; C n is the energy dissipation of the nth time step, which is used to quantify the energy dissipated by the surrounding rock due to plastic deformation or material damage in this time step; (σ, ε) is the coordinate pair of the stress-deformation curve, which records the stress and strain at each time step and shows the evolution of the mechanical state of the surrounding rock during the entire construction process; (σ0, ε0) is the initial stress and strain, indicating the initial state of the surrounding rock at the beginning of the simulation; (σ1, ε1) is the stress and strain of the first time step, reflecting the initial response after the blasting load is applied; (σ n ,ε n ) is the stress and strain at the nth time step, recording the stress and deformation evolution of the surrounding rock during the advancement of each time step.

[0102] Step S630, performing critical state analysis according to the stress-deformation history curve, comparing the maximum stress in the curve with the shear strength of the surrounding rock through the limit equilibrium analysis method, and obtaining the safety factor of the surrounding rock;

[0103] It can be understood that this step first extracts the maximum stress value of each key time point from the stress-deformation history curve, that is, the peak stress of the surrounding rock in each blasting stage and time step. These maximum stress values ​​reflect the most severe mechanical state that the surrounding rock is subjected to during the construction process. Next, these stress values ​​are compared with the shear strength of the surrounding rock. Shear strength is the maximum stress that the surrounding rock material can withstand before reaching shear failure, which depends on material properties such as the internal friction angle, cohesion and pore pressure of the surrounding rock. By comparing the maximum stress with the shear strength, it can be determined whether the surrounding rock is within a safe range. Next, the safety factor of the surrounding rock is calculated using the limit equilibrium analysis method. The calculation of the safety factor is used to quantify the stability of the surrounding rock under the current stress state, which is defined as the ratio of shear strength to maximum stress. By analyzing the maximum stress and safety factor at multiple time points, not only can the high-risk areas of the surrounding rock be identified, but also the possible instability time period of the surrounding rock can be predicted. The calculation formula is:

[0104]

[0105] τ c =c+σ0·tan(φ)-u;

[0106] Among them, SF n represents the safety factor of the nth time step; τ c Represents the shear strength of surrounding rock; σ max,n represents the maximum stress value at the nth time step; c represents the bonding strength between surrounding rock particles; σ0 represents the normal stress; φ represents the internal friction angle between surrounding rock particles; u represents the fluid pressure in the pores inside the rock mass.

[0107] Step S640, perform parameter optimization according to the safety factor and the optimization decision rule, establish a fuzzy comprehensive evaluation model based on the preset fuzzy comprehensive evaluation algorithm, take the safety factor, construction cost, and explosive consumption as input, set fuzzy rules and evaluate through the weighted average method to obtain the blasting excavation parameter setting plan.

[0108] In this step, the safety factor, construction cost and explosive dosage are used as input variables to establish a fuzzy comprehensive evaluation model to comprehensively evaluate the mutual influence and trade-off of these factors under different conditions. Since the relationship between these factors has certain uncertainty and complexity, fuzzy language descriptions such as "high", "medium" and "low" are used to construct fuzzy rules based on expert experience or historical data. These rules provide a logical basis for the selection of schemes in different scenarios. A comprehensive evaluation is performed by weighted average method, and weights are set according to the importance of each factor in different scenarios. For example, the safety factor is given priority in high-risk areas, while the weight of cost is increased when the budget is limited. After the results of each rule are weighted averaged, a comprehensive score is generated to evaluate the safety, economy and rationality of the current scheme. According to the comprehensive score, the blasting parameters are further optimized, including step design, blasthole layout scheme and detonation sequence. If the score shows that the safety factor is insufficient, the blasthole density will be increased or the detonation sequence will be adjusted; if the cost is too high, the explosive dosage will be reduced or the excavation sequence will be simplified. After multiple rounds of iterative optimization, a specific blasting excavation parameter setting plan was generated to clarify the location, depth, spacing, explosive dosage and detonation sequence of the blastholes. This plan not only ensures a balance between construction safety and economy, but also has flexibility and adaptability. It can be dynamically adjusted according to on-site feedback, providing scientific and efficient operation guidance for the construction process.

[0109] Further, step S610 includes step S611 to step S614.

[0110] Step S611, defining viscoelastic properties according to the equivalent material model, deriving the viscoelastic constitutive equation of the surrounding rock by analytical method, and obtaining a dynamic material model of the surrounding rock;

[0111] Specifically, the parameters in the equivalent material model are first used as input to capture the viscous and elastic characteristics of the surrounding rock and establish the mathematical form of the viscoelastic constitutive relationship. In the derivation process, the generalized Maxwell model is preferably used to describe the viscous damping effect and elastic recovery capacity of the material. The Maxwell model represents the stress response of the material as a combination of viscous elements and elastic elements, and describes it through time-dependent differential equations. The key to the analytical derivation is to map the combination relationship of viscoelastic elements into a standardized equation to ensure that the model can accurately reflect the dynamic response of the surrounding rock under different stress states and loading times. In the derivation process, the stress and strain are solved in stages, and the loading rate and time effects are incorporated into the equation to generate a dynamic material model covering elastic, viscous and stress relaxation characteristics. The final viscoelastic constitutive equation can describe the evolution of stress and deformation of the surrounding rock during dynamic construction, and predict the mechanical performance of the surrounding rock under different construction conditions. The viscoelastic constitutive equation is:

[0112]

[0113] Among them, σ(t) represents the stress value at time t, describing the internal force state of the surrounding rock at that time; ε(t) represents the strain at time t, describing the deformation degree of the surrounding rock; E1 is the elastic modulus, which represents the elastic recovery capacity in the Maxwell model; η1 is the viscous damping coefficient, which represents the viscous effect of the material; λ1 is the time constant.

[0114] Step S612, performing transient response calculation according to the dynamic material model, solving the dynamic response of the surrounding rock under the blasting load by explicit finite difference method, and obtaining the dynamic response data of the surrounding rock at each time step;

[0115] It should be noted that the load applied to the surrounding rock during the blasting process is a short-term, high-intensity impact load, which will produce complex stress wave propagation and stress redistribution in the surrounding rock. Therefore, it is necessary to gradually solve its dynamic response through efficient numerical methods.

[0116] Specifically, this step first applies the blasting load as a boundary condition to the surrounding rock model, and divides the entire calculation process into multiple small time steps to capture the stress and deformation state of the surrounding rock at every moment during the blasting process. The advantage of the explicit finite difference method is that its calculation steps are clear, and the results in each time step only depend on the data of the previous time step. This step-by-step solution method is particularly suitable for dealing with short-term dynamic impact loads. In each time step, the stress and strain of each unit are calculated according to the dynamic material model, and the results are passed to the next time step to ensure that the stress wave propagation and stress redistribution process can be accurately simulated.

[0117] During the calculation process, the mechanical state of each unit, including stress, strain and deformation rate, is continuously updated to track the impact of blasting loads on the surrounding rock. This time-step calculation can not only capture the deformation trajectory of the surrounding rock, but also identify stress concentration areas and potential instability points, providing engineers with key construction safety information. Finally, the calculation results in all time steps are integrated to form a dynamic response data set, which records in detail the stress and deformation of the surrounding rock at each time point during the blasting process.

[0118] Step S613: Perform energy analysis based on the dynamic response data, and obtain the energy dissipation characteristics of the surrounding rock by calculating the product of stress and strain;

[0119] It should be noted that during blasting construction, the transmission and dissipation of energy have an important influence on the deformation, crack expansion and stability of the surrounding rock. Therefore, it is necessary to evaluate the response and instability risk of the surrounding rock through energy analysis. Specifically, this step first extracts the stress and strain state of each unit from the dynamic response data of each time step. For each time step, the instantaneous energy density of each unit is calculated, that is, the product of stress and strain per unit volume, reflecting the elastic energy absorbed and released by the unit. These instantaneous energy data are gradually accumulated to form a complete energy evolution process, showing the energy dissipation path of the surrounding rock during the entire blasting process. In the analysis process, energy is mainly divided into elastic energy and dissipated energy. Elastic energy reflects the recovery ability of the surrounding rock, and this part of energy can be partially recovered when the stress is released; while dissipated energy reflects the energy consumed by permanent deformation and internal friction of the material, which cannot be recovered. By calculating the dissipated energy of each unit, the areas where plastic deformation or crack expansion may occur in the surrounding rock can be identified. These areas are often high dissipated energy concentration points and are prone to become potential risk areas of instability. Finally, the energy data of each unit are summarized into an energy dissipation characteristic data set, which shows the energy distribution and evolution law of the surrounding rock at different stages.

[0120] Step S614: construct a function based on the dynamic response data and the energy dissipation characteristics, establish a mathematical model using regression analysis, and use stress, strain and energy as input to obtain a dynamic response function through fitting.

[0121] It is understandable that this step uses the dynamic response data obtained from the previous step, including stress, strain and energy dissipation data in each time step, and uses these variables as the input of the model. Since the relationship between stress and strain shows nonlinear characteristics, and energy dissipation reflects the complex deformation and damage behavior inside the material, it is necessary to find the best matching relationship between the variables through regression analysis. Preferably, a regression algorithm in machine learning (such as ridge regression or support vector regression) is used to capture these complex mechanical relationships. In the specific fitting process, training is performed based on dynamic data, and the error between the predicted value and the actual observed data is gradually reduced by adjusting the model parameters to ensure that the final dynamic response function can accurately reflect the true mechanical behavior of the surrounding rock. The regression model can not only describe the evolution of stress and strain over time, but also reveal the relationship between energy dissipation and surrounding rock response under different construction conditions. The dynamic response function finally fitted is a simplified but efficient mathematical expression that can quickly predict the dynamic response behavior of the surrounding rock under given stress, strain and energy conditions.

[0122] Embodiment 2:

[0123] This embodiment provides a tunnel surrounding rock blasting excavation parameter setting system based on a neural network, the system comprising:

[0124] An acquisition module is used to acquire geological property data of surrounding rocks, blasting excavation history records and excavation profile data;

[0125] Clustering module, used to perform sparse subspace clustering processing based on the blasting excavation history records, and obtain feature data sets by performing sparse subspace decomposition on the surrounding rock characteristics and blasting effects in the historical data;

[0126] The analysis module is used to analyze the stress state according to the geological characteristic data, excavation profile data and characteristic data set. By simulating the stress change process of the surrounding rock at different time points, the stress state and spatial change of the surrounding rock are modeled using the time-dependent topological analysis method to obtain the dynamic stress state and topological change data of the surrounding rock;

[0127] The mining module is used to mine association rules based on the blasting excavation history records and feature data sets, and obtain the optimization decision rules by mining the intrinsic association between the excavation sequence and the blasthole layout plan;

[0128] The construction module is used to perform nonlinear finite element analysis based on dynamic stress state and topological change data, and obtain the equivalent material model of the surrounding rock by establishing a mechanical behavior model of the surrounding rock and performing numerical calculations;

[0129] The output module is used to obtain the blasting excavation parameter setting scheme based on the equivalent material model and optimization decision rules, after being processed by the preset viscoelastic fluid model, by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting. The blasting excavation parameter setting scheme includes the excavation sequence, blasthole arrangement scheme and detonation sequence.

[0130] In a specific implementation disclosed in the present application, the clustering module includes:

[0131] The first dimension reduction unit is used to perform data dimension reduction processing according to the blasting excavation historical records, extract surrounding rock characteristic data and corresponding blasting effect data through principal component analysis, and obtain a low-dimensional feature set;

[0132] The first encoding unit is used to perform sparse coding processing according to the low-dimensional feature set, and sparsely represent the features through Lasso regression to obtain a sparse feature matrix;

[0133] The first calculation unit is used to perform similarity measurement according to the sparse feature matrix, calculate the similarity between different historical blasting records by cosine similarity, and obtain a similarity matrix;

[0134] The first clustering unit is used to perform sparse subspace clustering processing according to the similarity matrix, and divide the blasting records into at least two related subspaces through a sparse subspace clustering algorithm to obtain a feature data set.

[0135] In a specific embodiment disclosed in the present application, the analysis module includes:

[0136] The first processing unit is used to perform spatial discretization processing according to the geological characteristic data and the excavation profile data, and to perform grid discretization on the tunnel surrounding rock through the finite volume method to obtain a spatial discretization model;

[0137] The second calculation unit is used to calculate the initial stress field according to the spatial discrete model, calculate the surrounding rock deadweight and formation pressure and use Gaussian numerical integration to calculate the stress, and obtain the initial stress distribution state of the surrounding rock;

[0138] The first simulation unit is used to simulate the dynamic stress change according to the initial stress distribution state, and obtain the time series stress distribution data by gradually promoting the stress redistribution of the surrounding rock during the excavation process and simulating the stress release and redistribution of the surrounding rock due to the removal of rock mass at different time points during the excavation process;

[0139] The first analysis unit is used to perform time-dependent topological analysis based on time series stress distribution data, identify stress concentration area changes and newly formed fractures in the stress field that occur over time through a persistent homology method, and obtain dynamic stress state and topological change data.

[0140] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for setting tunnel surrounding rock blasting excavation parameters based on neural network, characterized in that: include: Obtain geological characteristics data of surrounding rock, blasting excavation history and excavation profile data; Performing sparse subspace clustering processing according to the blasting excavation historical records, and obtaining a feature data set by performing sparse subspace decomposition on surrounding rock characteristics and blasting effects in the historical data; Performing stress state analysis according to the geological characteristic data, the excavation profile data and the characteristic data set, modeling the stress state and spatial variation of the surrounding rock using a time-dependent topological analysis method by simulating the stress variation process of the surrounding rock at different time points, and obtaining dynamic stress state and topological variation data of the surrounding rock; Performing association rule mining based on the blasting excavation history records and the feature data set, and obtaining an optimization decision rule by mining the intrinsic association between the excavation sequence and the blasthole layout plan; Performing nonlinear finite element analysis according to the dynamic stress state and the topological change data, and obtaining an equivalent material model of the surrounding rock by establishing a mechanical behavior model of the surrounding rock and performing numerical calculations; According to the equivalent material model and the optimization decision rule, after being processed by a preset viscoelastic fluid model, a blasting and excavation parameter setting scheme is obtained by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting. The blasting and excavation parameter setting scheme includes excavation steps, blasthole arrangement scheme and detonation sequence.

2. The method for setting tunnel surrounding rock blasting excavation parameters based on neural network according to claim 1, characterized in that: According to the blasting excavation history records, sparse subspace clustering is performed, and by performing sparse subspace decomposition on the surrounding rock characteristics and blasting effects in the historical data, a feature data set is obtained, including: Performing data dimensionality reduction processing according to the blasting excavation historical records, extracting surrounding rock characteristic data and corresponding blasting effect data through principal component analysis to obtain a low-dimensional feature set; Performing sparse coding processing on the low-dimensional feature set, sparsely representing the features through Lasso regression, and obtaining a sparse feature matrix; Performing similarity measurement based on the sparse feature matrix, calculating the similarity between different historical blasting records by cosine similarity, and obtaining a similarity matrix; Sparse subspace clustering processing is performed according to the similarity matrix, and the blasting records are divided into at least two related subspaces by a sparse subspace clustering algorithm to obtain a feature data set.

3. The method for setting tunnel surrounding rock blasting excavation parameters based on neural network according to claim 1, characterized in that: The stress state analysis is performed according to the geological characteristic data, the excavation profile data and the characteristic data set. By simulating the stress change process of the surrounding rock at different time points, the stress state and spatial change of the surrounding rock are modeled using a time-dependent topological analysis method to obtain the dynamic stress state and topological change data of the surrounding rock, including: Performing spatial discretization processing according to the geological characteristic data and the excavation profile data, gridding and discretizing the tunnel surrounding rock through the finite volume method, and obtaining a spatial discretization model; The initial stress field is calculated according to the spatial discrete model, and the initial stress distribution state of the surrounding rock is obtained by calculating the deadweight of the surrounding rock and the formation pressure and using Gaussian numerical integration to calculate the stress; Performing a dynamic stress change simulation according to the initial stress distribution state, by gradually promoting the stress redistribution of the surrounding rock during the excavation process, and simulating the stress release and redistribution of the surrounding rock due to the removal of the rock mass at different time points during the excavation process, to obtain time series stress distribution data; A time-dependent topological analysis is performed based on the time series stress distribution data, and the stress concentration area changes and newly formed fractures generated in the stress field over time are identified by a persistent homology method to obtain dynamic stress state and topological change data.

4. The method for setting tunnel surrounding rock blasting excavation parameters based on neural network according to claim 1, characterized in that: Association rule mining is performed based on the blasting excavation history records and the feature data set, and optimization decision rules are obtained by mining the intrinsic association between the excavation sequence and the blasthole layout plan, including: Performing feature sequence analysis based on the blasting excavation history records and feature data sets, comparing surrounding rock stress evolution patterns under different step sequence conditions through dynamic time warping, and obtaining surrounding rock stress sequence matrices under different excavation step sequence conditions; Performing stress pattern clustering according to the surrounding rock stress sequence matrix, and obtaining clustering results of excavation step sequence and stress evolution by clustering step combinations with similar stress evolution characteristics; According to the clustering results, an association rule analysis of blasthole layout is performed, and the probability dependency relationship between the excavation sequence and the blasthole layout is modeled by a Bayesian network, and the influence of each blasthole layout on the stress evolution of a specific sequence is quantified to obtain a probability association rule; Multi-objective optimization processing is performed according to the probabilistic association rules. The stress evolution and surrounding rock stability are simulated and evaluated by simulating the combination of different blasthole schemes and step sequences. The stability and economy of the tunnel surrounding rock are used as reward functions for screening to obtain the optimization decision rules.

5. The method for setting tunnel surrounding rock blasting excavation parameters based on neural network according to claim 1, characterized in that: According to the dynamic stress state and the topological change data, nonlinear finite element analysis is performed, and an equivalent material model of the surrounding rock is obtained by establishing a mechanical behavior model of the surrounding rock and performing numerical calculations, including: Fitting nonlinear material property parameters according to the dynamic stress state and the topological change data, optimizing and fitting the nonlinear constitutive relationship of the surrounding rock material through a genetic algorithm, and obtaining nonlinear material parameters of the surrounding rock; Dividing the nonlinear finite element unit grid according to the nonlinear material parameters, optimizing the grid of the spatial discrete model of the surrounding rock through an adaptive grid generation algorithm, and obtaining a finite element grid model containing nonlinear characteristics; The nonlinear mechanical behavior simulation is performed according to the finite element mesh model, and based on the initial stress state of the surrounding rock and the mesh model, the stress and strain of each unit are gradually and iteratively solved until a mechanical equilibrium state is reached, thereby obtaining the stress-strain distribution result of the surrounding rock; An equivalent material model is constructed according to the stress-strain distribution result, and the equivalent material properties of the surrounding rock are determined by an inversion analysis algorithm to obtain an equivalent material model.

6. The method for setting tunnel surrounding rock blasting excavation parameters based on neural network according to claim 1, characterized in that: According to the equivalent material model and the optimization decision rule, after being processed by the preset viscoelastic fluid model, by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting, a blasting excavation parameter setting scheme is obtained, including: Performing dynamic response analysis based on the equivalent material model, modeling the viscoelastic characteristics of the surrounding rock by applying a piecewise linearization method, and obtaining a dynamic response function; The dynamic response of the surrounding rock is simulated using the time domain integration method, by applying the dynamic response function to the time history of each round of blasting, and calculating the stress and deformation at each moment by step-by-step time integration to obtain the stress-deformation curve; Perform critical state analysis based on the stress-deformation history curve, compare the maximum stress in the curve with the shear strength of the surrounding rock through the limit equilibrium analysis method, and obtain the safety factor of the surrounding rock; Parameters are optimized according to the safety factor and the optimization decision rules, a fuzzy comprehensive evaluation model is established based on a preset fuzzy comprehensive evaluation algorithm, the safety factor, construction cost, and explosive dosage are taken as inputs, fuzzy rules are set, and evaluation is performed through the weighted average method to obtain a blasting excavation parameter setting plan.

7. The method for setting tunnel surrounding rock blasting excavation parameters based on neural network according to claim 6, characterized in that: According to the equivalent material model, dynamic response analysis is performed, and the viscoelastic characteristics of the surrounding rock are modeled by applying the piecewise linearization method to obtain the dynamic response function, including: The viscoelastic properties are defined according to the equivalent material model, and the viscoelastic constitutive equation of the surrounding rock is derived by an analytical method to obtain a dynamic material model of the surrounding rock; Perform transient response calculation according to the dynamic material model, solve the dynamic response of the surrounding rock under blasting load by explicit finite difference method, and obtain the dynamic response data of the surrounding rock at each time step; Performing energy analysis based on the dynamic response data, and obtaining energy dissipation characteristics of the surrounding rock by calculating the product of stress and strain; Function construction processing is performed according to the dynamic response data and energy dissipation characteristics, a mathematical model is established using regression analysis, and stress, strain and energy are used as inputs to obtain a dynamic response function through fitting.

8. A tunnel surrounding rock blasting excavation parameter setting system based on neural network, characterized in that: include: An acquisition module is used to acquire geological property data of surrounding rocks, blasting excavation history records and excavation profile data; A clustering module is used to perform sparse subspace clustering processing according to the blasting excavation historical records, and obtain a feature data set by performing sparse subspace decomposition on surrounding rock characteristics and blasting effects in the historical data; An analysis module is used to perform stress state analysis based on the geological characteristic data, the excavation profile data and the characteristic data set, and to model the stress state and spatial variation of the surrounding rock using a time-dependent topological analysis method by simulating the stress variation process of the surrounding rock at different time points, so as to obtain dynamic stress state and topological variation data of the surrounding rock; A mining module is used to mine association rules based on the blasting excavation history records and the feature data set, and obtain an optimization decision rule by mining the intrinsic association between the excavation sequence and the blasthole layout plan; A construction module is used to perform nonlinear finite element analysis according to the dynamic stress state and the topological change data, and obtain an equivalent material model of the surrounding rock by establishing a mechanical behavior model of the surrounding rock and performing numerical calculations; The output module is used to obtain a blasting and excavation parameter setting scheme based on the equivalent material model and the optimization decision rule, after being processed by a preset viscoelastic fluid model, by simulating the stress release and deformation process of the surrounding rock under all rounds of blasting. The blasting and excavation parameter setting scheme includes an excavation sequence, a blasthole arrangement scheme, and a detonation sequence.

9. The neural network-based tunnel surrounding rock blasting excavation parameter setting system according to claim 8, characterized in that: The clustering module comprises: A first dimension reduction unit is used to perform data dimension reduction processing according to the blasting excavation historical records, extract surrounding rock characteristic data and corresponding blasting effect data through principal component analysis to obtain a low-dimensional feature set; A first encoding unit is used to perform sparse encoding processing according to the low-dimensional feature set, and sparsely represent the features through Lasso regression to obtain a sparse feature matrix; A first calculation unit is used to perform similarity measurement according to the sparse feature matrix, calculate the similarity between different historical blasting records by cosine similarity, and obtain a similarity matrix; The first clustering unit is used to perform sparse subspace clustering processing according to the similarity matrix, and divide the blasting records into at least two related subspaces by a sparse subspace clustering algorithm to obtain a feature data set.

10. The neural network-based tunnel surrounding rock blasting excavation parameter setting system according to claim 8, characterized in that: The analysis module comprises: A first processing unit is used to perform spatial discretization processing according to the geological characteristic data and the excavation profile data, and to perform grid discretization on the tunnel surrounding rock through a finite volume method to obtain a spatial discretization model; A second calculation unit is used to calculate the initial stress field according to the spatial discrete model, calculate the surrounding rock deadweight and formation pressure and use Gaussian numerical integration to calculate stress, and obtain the initial stress distribution state of the surrounding rock; A first simulation unit is used to simulate dynamic stress changes according to the initial stress distribution state, by gradually promoting stress redistribution of the surrounding rock during the excavation process, and simulating stress release and redistribution of the surrounding rock due to removal of rock mass at different time points during the excavation process, to obtain time series stress distribution data; The first analysis unit is used to perform time-dependent topological analysis based on the time series stress distribution data, identify the stress concentration area changes and newly formed fractures in the stress field that occur over time through a persistent homology method, and obtain dynamic stress state and topological change data.

Images