Automatic modeling analysis method for highway tunnel drainage pipe network combined with natural language

Abstract

The present application belongs to the technical field of tunnel engineering, and is used for solving the model distortion defect problem caused by ignoring the coupling relationship between semantic ambiguity and geological risk in the prior art, and specifically is an automatic modeling analysis method for highway tunnel drainage pipe network combined with natural language, comprising the following steps: obtaining first interactive instructions containing fuzzy semantic description of arrangement parameters of target tunnel drainage components and geological exploration data of the target tunnel along its axis; extracting the tolerance interval implied in the fuzzy semantic description and the linkage rules of the tolerance interval and the geological conditions, generating a component spacing parameter sequence that changes dynamically along the axis of the target tunnel, and driving the parameterized modeling program to generate a three-dimensional model of the drainage pipe network of the target tunnel according to the component spacing parameter sequence; the present application can recognize the trigger condition, response rule expression and sensitivity coefficient linked with the geological conditions by guiding the large language model to deeply interpret the fuzzy semantic description through system prompt words.

Description

Technical Field

[0001] This invention belongs to the field of tunnel engineering technology, specifically a method for automated modeling and analysis of highway tunnel drainage networks using natural language processing. Background Technology

[0002] As a key node in transportation infrastructure, the reliability of the drainage system of highway tunnels is directly related to structural safety and operational stability, especially in extreme rainfall weather and water-rich mountainous geological conditions. As tunnel construction extends to complex geological areas, the geological conditions faced by drainage design are becoming increasingly diverse, which puts forward higher requirements for design accuracy.

[0003] To improve the level of design automation, existing technologies attempt to introduce large language models to assist in the modeling of drainage pipe networks. The typical processing method is as follows: the large language model is guided by preset prompts to parse the natural language descriptions input by the designers. For example, a qualitative statement such as "circumferential duct spacing of 5 to 15 meters" is directly converted into a set of specific, discrete spacing values. Subsequently, the parametric modeling program reads the list and arranges the drainage components sequentially along the tunnel axis to generate a three-dimensional pipe network model. This linear processing flow of "natural language - discrete parameters - fixed model" has initially realized the automated conversion from semantics to model.

[0004] However, there is a fundamental logical gap in the above transformation process. When designers use interval descriptions such as "5 to 15 meters," their true intention is often not to take a uniform value within the interval, but rather to imply an expectation that the geological conditions along the tunnel will change: for example, in high-risk areas such as fault fracture zones and water-rich areas, a smaller spacing value is expected to be used to enhance drainage capacity; in areas with good surrounding rock, the spacing can be appropriately widened to save costs. In other words, the fuzzy semantic description itself carries engineering experience that is dynamically coupled with the geological environment; but existing technology treats it only as noise and forces a "formally correct" discrete array to be generated through the model, which is completely decoupled from the geological risk distribution along the tunnel.

[0005] In sections where geological conditions change drastically, such as when a tunnel needs to traverse a water-rich fault fracture zone, this decoupling can lead to serious consequences. The drainage network generated based on a randomly or uniformly distributed discrete spacing array may have a component density in critical sections that deviates significantly from the designers' desired density scheme. Hydraulic analysis based on this model will severely underestimate the actual drainage needs of the section, leading to overly optimistic conclusions, while potential safety hazards such as excessive lining water pressure and water inrush are masked during the design process. This means that while existing technologies achieve formal automation, their core element—converting natural language into design parameters—loses the dynamic intent inherent in the semantics and related to geological risks, causing the model's accuracy to fail under critical conditions. Therefore, effectively decoupling the environmentally changing engineering intent from the fuzzy descriptions of natural language and integrating it with the objective risk distribution revealed by geological exploration data to generate network parameters that match real-world conditions has become a core problem that must be overcome to improve the level of automated tunnel drainage design. Summary of the Invention

[0006] The purpose of this invention is to provide an automated modeling and analysis method for highway tunnel drainage networks that incorporates natural language processing.

[0007] The technical problem to be solved by this invention is: how to effectively decouple the engineering intent that changes with geological conditions from the fuzzy description of natural language, and integrate it with the objective risk distribution revealed by discrete geological exploration data to generate dynamic pipeline parameters that match the actual working conditions along the tunnel, so as to overcome the model distortion defects caused by ignoring the coupling relationship between semantic fuzziness and geological risk in the existing technology.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] An automated modeling and analysis method for highway tunnel drainage networks using natural language processing includes the following steps:

[0010] Acquire a first interactive command containing a fuzzy semantic description of the arrangement parameters of the drainage components of the target tunnel, as well as geological exploration data of the target tunnel along its axis.

[0011] Semantic interpretation is performed on the fuzzy semantic description to extract the tolerance range and the linkage rules between the tolerance range and geological conditions. The tolerance range includes the basic spacing and the allowable fluctuation range of the spacing. The linkage rules include at least one type of geological condition used to trigger the linkage rules and a response rule expression describing how the spacing changes with geological risk.

[0012] Based on geological exploration data, a normalized geological risk index function is constructed that is continuously distributed along the axis of the target tunnel. The normalized geological risk index function is used to characterize the degree of geological risk at each location on the tunnel axis.

[0013] The normalized geological risk index function is substituted into the response rule expression as an input variable, and combined with the tolerance interval, a sequence of component spacing parameters that dynamically changes along the target tunnel axis is generated. The spacing value at each location in the component spacing parameter sequence is calculated by the normalized geological risk index at that location according to the response rule expression, and the spacing value is constrained within the allowable fluctuation range of the spacing.

[0014] Based on the component spacing parameter sequence, the parametric modeling program is driven to generate a three-dimensional model of the drainage pipe network of the target tunnel.

[0015] The present invention has the following beneficial effects:

[0016] 1. By guiding a large language model through system prompts, this method performs in-depth interpretation of fuzzy semantic descriptions. It not only extracts the basic spacing and allowable fluctuation range of the spacing, but more importantly, identifies the triggering conditions, response rule expressions, and sensitivity coefficients linked to geological conditions. Unlike traditional methods that simply convert interval descriptions into a list of discrete values, this method retains the conditional intent from engineering experience, such as "densification is required in fault fracture zones," transforming fuzzy qualitative descriptions into calculable, rule-based parameters. This process ensures that the natural language input by designers is no longer noise requiring "precision," but rather a knowledge carrier carrying dynamic engineering intent, laying the information foundation for subsequent deep integration with geological data.

[0017] 2. Based on discrete borehole data, a normalized geological risk index function continuously distributed along the tunnel axis is constructed. This function is used as an input variable in the response rule expression to calculate the component spacing matching the geological risk at each mileage location in real time. When the tunnel passes through high-risk sections such as fault fracture zones, the system automatically reduces the spacing to enhance drainage capacity. In sections with good surrounding rock, the spacing is appropriately increased to save costs. This nonlinear coupling mechanism of "field modulation rule" ensures that the generated component spacing parameter sequence is strictly aligned with the objective distribution of geological risk, fundamentally solving the model distortion problem caused by the decoupling of the fixed spacing array from the physical environment in traditional methods, and ensuring the pertinence and effectiveness of drainage design in different geological sections.

[0018] 3. After performing hydraulic analysis on the generated 3D model, attribution analysis is conducted based on the distribution characteristics of hydraulic anomaly sections: for local anomalies, the geological risk field is corrected by supplementing virtual borehole points; for systematic anomalies, the sensitivity coefficient is optimized using the least squares method. This feedback mechanism transforms the hydraulic analysis results into a basis for correcting preceding parameters, forming a complete closed loop of "design-analysis-correction-redesign". Compared with the traditional one-time design method, this method can automatically identify and correct deviations caused by sparse geological data, inappropriate sensitivity coefficients, etc., so that the hydraulic performance of the final model meets the preset safety threshold, significantly improving the reliability and safety of the drainage system under extreme conditions.

[0019] 4. Before generating the spacing, the overlap index between the default applicable range of the geological condition type in the linkage rule and the actual risk section is pre-calculated and compared with the preset matching degree threshold. When the overlap is lower than the threshold, the system automatically corrects the rule's effective range to the intersection of the two, ensuring that the linkage rule only works in the section where the geological condition actually exists. This pre-verification mechanism aligns the potential overgeneralization in the fuzzy semantic description (such as the default rule being applicable to the entire tunnel) with the actual geological distribution, avoiding the incorrect application of encryption rules in sections that do not have the corresponding geological conditions. At the same time, matching degree prompts are generated for designers to refer to, providing a quantitative basis for engineering judgment. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the method according to Embodiment 1 of the present invention;

[0022] Figure 2 This is a flowchart of the method in Embodiment 2 of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] As a critical node in transportation infrastructure, the reliability of the drainage system in highway tunnels directly affects structural safety and operational stability, especially under conditions of extreme rainfall and in water-rich mountainous areas. As tunnel construction extends into more complex geological regions, the geological conditions faced by drainage design are becoming increasingly diverse, placing higher demands on design accuracy and adaptability. While the industry has begun to explore the use of digital tools to improve design efficiency, tunnel drainage design still heavily relies on designers manually adjusting specifications and drawings, making it difficult to achieve a rapid and accurate mapping from geological conditions to design parameters.

[0025] To improve the level of design automation, existing technologies attempt to introduce large language models to assist in the modeling of drainage pipe networks. A typical approach is as follows: Pre-set prompts guide the large language model to parse the natural language descriptions input by the designer. For example, a qualitative statement such as "circumferential duct spacing 5 to 15 meters" is directly converted into a list of specific, discrete spacing values. Subsequently, the parametric modeling program reads this list and sequentially arranges drainage components along the tunnel axis, thereby generating a three-dimensional pipe network model. This linear processing flow of "natural language – discrete parameters – fixed model" initially achieves automated conversion from semantics to model.

[0026] However, a fundamental logical gap exists in the aforementioned transformation process. When designers use interval descriptions like "5 to 15 meters," their true intention is often not to take uniform values ​​within the interval, but rather to implicitly anticipate changes in geological conditions along the tunnel route: for example, in high-risk areas such as fault fracture zones and water-rich areas, a smaller spacing value is expected to enhance drainage capacity; in areas with good surrounding rock, the spacing can be appropriately widened to save costs. In other words, the fuzzy semantic description itself carries engineering experience dynamically coupled with the geological environment. However, existing technologies treat this merely as noise, forcibly generating a "formally correct" discrete array through the model, which is completely decoupled from the geological risk distribution along the tunnel route.

[0027] In sections where geological conditions change drastically, such as when a tunnel needs to traverse a water-rich fault fracture zone, this decoupling can lead to serious consequences. Drainage networks generated based on randomly or uniformly distributed discrete spacing arrays may have component densities in critical sections that deviate significantly from the designers' desired density. Hydraulic analyses based on this model will severely underestimate the actual drainage needs of the section, leading to overly optimistic conclusions, while potential safety hazards such as excessive lining water pressure and water inrush are masked during the design process. This means that while existing technologies achieve formal automation, their core element—translating natural language into design parameters—loses the dynamic intent inherent in the semantics and related to geological risks, causing the model's accuracy to fail under critical conditions. Therefore, effectively decoupling the environmentally changing engineering intent from the fuzzy descriptions of natural language and integrating it with the objective risk distribution revealed by geological exploration data to generate network parameters that match actual working conditions has become a core problem that must be overcome to improve the level of automated tunnel drainage design.

[0028] Example 1: As Figure 1 As shown, the automated modeling and analysis method for highway tunnel drainage networks, which incorporates natural language processing, includes the following steps:

[0029] Step S1: Obtain a first interactive command containing a fuzzy semantic description of the arrangement parameters of the drainage components of the target tunnel and geological exploration data of the target tunnel along its axis.

[0030] Step S1 first acquires two types of input data. The first interactive instruction refers to the natural language description entered by the designer through the interactive interface. For example, in the modeling system dialog box, the instruction "The spacing of the circumferential guide pipes is generally 5-15 meters, and it needs to be densified to about 5 meters in fault fracture zones, and appropriately densified in water-rich areas" is entered. This instruction is transmitted to the back-end processing module in plain text form. The geological exploration data consists of discrete borehole records distributed along the tunnel axis. Each borehole record contains the mileage location of the borehole and four parameters at that location: the quantified value of the surrounding rock grade, rock quality index, groundwater pressure, and fracture density. The surrounding rock grade is divided into grades I to VI according to national standards. In this embodiment, it is quantified as a numerical value: I is 1, II is 2, and so on up to grade VI is 6. The rock quality index is a percentage, ranging from 0 to 100. The groundwater pressure unit is MPa, and the fracture density unit is fractures per meter. All borehole data is stored in a structured table format, with each row corresponding to one borehole. Columns include borehole mileage, surrounding rock grade quantification, rock quality indicators, groundwater pressure, and fracture density, arranged in ascending order of mileage to form a geological exploration dataset. These two types of data will be used for subsequent semantic interpretation and geological risk function construction, respectively.

[0031] Step S2: Perform semantic interpretation on the fuzzy semantic description, extract the tolerance range and the linkage rules between the tolerance range and geological conditions implied in the fuzzy semantic description. The tolerance range includes the basic spacing and the allowable fluctuation range of the spacing. The linkage rules include at least one type of geological condition used to trigger the linkage rules and a response rule expression describing how the spacing changes with geological risk.

[0032] Step S2 performs semantic interpretation on the first interactive instruction obtained in step S1, extracting the implicit tolerance range and linkage rules. To achieve this function, this embodiment pre-deploys a lightweight large language model with 7 billion parameters. This model is built on the Transformer decoder architecture and has 12 decoding layers and a hidden layer dimension of 4096. The lightweight model was chosen instead of a large model with hundreds of billions of parameters because this step has high requirements for response speed and output format stability: the lightweight model can guarantee a single inference time of less than 500 milliseconds while maintaining sufficient semantic understanding capabilities, and it is easier to constrain the output format through prompt words. The model is deployed on a local server and provides service calls through a RESTful API.

[0033] The core of semantic interpretation lies in the design of system prompts. System prompts consist of three parts: The first part is the task description, explicitly informing the model that it needs to extract "basic spacing," "lower limit of the allowable spacing fluctuation range," "upper limit of the allowable spacing fluctuation range," "linkage triggering condition," "response rule expression," and "sensitivity coefficient" from the user input and return them in JSON format. The second part is the format constraints, stipulating that the JSON key names must use the full Chinese names, such as "basic spacing," "lower limit of the tolerance range," "upper limit of the tolerance range," "linkage triggering condition," "response rule expression," and "sensitivity coefficient," and specifying the value type for each key: basic spacing is a floating-point number, the lower and upper limits of the tolerance range are floating-point numbers, the linkage triggering condition is a list of strings, the response rule expression is a string, and the sensitivity coefficient is a floating-point number within the range [0,1]. The third part is an example, providing a complete input-output pair: assuming the input is "lateral guide spacing 8 to 12 meters, using the lower limit value in soft rock deformation sections," the expected output is {

[0034] "Basic Spacing": 10.0

[0035] "Lower limit of the tolerance range": 8.0

[0036] "Upper limit of tolerance range": 12.0,

[0037] "Triggering Conditions for Linkage": ["Soft Rock Deformation Section"],

[0038] "Response rule expression": "

[0039] Sensitivity coefficient: 0.4

[0040] };

[0041] Where "base" represents the base spacing, "k" represents the sensitivity coefficient, and "R" represents the normalized geological risk index. The expression in the example is in linear form, which can be varied according to the semantic description, but this embodiment uses this linear form to simplify subsequent calculations.

[0042] The system concatenates the first interactive instruction obtained in step S1 with the aforementioned system prompts to form the complete model input text, which is then sent to the large language model via API. The text data returned by the model is first processed by the format cleaning module. This module uses regular expressions to search for JSON blocks in the text; if no JSON block matching the format is found, an error message is returned and the user is prompted to re-enter the text. The extracted JSON string is then fed into Python's json library for decoding, resulting in a data object containing six key-value pairs.

[0043] The meanings of the specific fields obtained after decoding are as follows:

[0044] The value of the “Basic Spacing” field is a floating-point number, representing the center reference value intended by the designer, such as 10.0;

[0045] The value of the "Tolerance Range Lower Limit" field is a floating-point number, representing the minimum allowable spacing, such as 5.0;

[0046] The value of the "Tolerance Range Upper Limit" field is a floating-point number, representing the maximum allowable spacing, such as 15.0;

[0047] The value of the “Linkage Trigger Condition” field is a list of strings, such as [“Fault Fracture Zone”, “Water-rich Area”]. Each element in the list is a geological condition type, indicating that when the tunnel passes through this type of geological section, the spacing should be adjusted according to the response rules.

[0048] The value of the "Response Rule Expression" field is a string, such as " This expression will be parsed into a computable mathematical function in subsequent steps, where base is the base spacing, k is the sensitivity coefficient, and R is the normalized geological risk index.

[0049] The value of the "Sensitivity Coefficient" field is a floating-point number in the range [0,1]. For example, 0.5 indicates the intensity of the geological risk's control over the spacing: the closer the coefficient is to 1, the greater the impact of risk changes on the spacing; the closer the coefficient is to 0, the closer the spacing is to the base value.

[0050] If the initial parsing fails, the system will automatically perform the following fault-tolerant operations:

[0051] Try using regular expressions to extract all possible JSON fragments, and then decode each fragment in turn.

[0052] If it still fails, the model returns text to a secondary validation model (a specially trained small-parameter classifier), which determines whether the text contains all the necessary fields and attempts to extract them;

[0053] If all of the above fail, the system will log the interaction command, the original text returned by the model, and the error message, and return a prompt to the user that "semantic interpretation failed, please describe again or check the input". At the same time, a manual correction interface is provided, allowing users to directly edit the structured data in JSON format.

[0054] If a key is missing or a data type mismatch occurs during decoding (e.g., "linkage trigger condition" is not a list), the system will also return an error message and require re-entry. Upon successful parsing, the system encapsulates the values ​​of the six fields into a structured object, which is then passed as the output of step S2 to subsequent steps. At this point, the fuzzy semantic description has completed the transformation from natural language to structured parameters, laying the foundation for the subsequent generation of dynamic spacing for the linkage geological risk function.

[0055] The large language model used in this embodiment is based on an open-source foundation model and undergoes domain-adaptive fine-tuning. The fine-tuning dataset is constructed as follows:

[0056] We collected 500 sets of typical natural language descriptions in highway tunnel drainage design, covering various component types such as circumferential ducts, transverse ducts, and longitudinal blind drains;

[0057] Each description is labeled with a structured JSON output by a senior tunnel engineer, including the foundation spacing, tolerance range, linkage triggering conditions, response rule expression, and sensitivity coefficient.

[0058] The LoRA parameter efficient fine-tuning method is adopted, an adaptation layer is added on the basis of the base model, only the parameters of the adaptation layer are trained, and the weights of the main body of the base model are frozen;

[0059] The training objective is to minimize the cross-entropy loss between the model output and the labeled JSON;

[0060] After training, the adaptation layer parameters are merged with the base model and deployed on a local server. This fine-tuning improves the model's parsing accuracy for specific semantics in the tunnel drainage domain to over 92%, and the output format stability reaches 99%.

[0061] Step S3: Based on geological exploration data, construct a normalized geological risk index function that is continuously distributed along the target tunnel axis. The normalized geological risk index function is used to characterize the degree of geological risk at each location on the tunnel axis.

[0062] Step S3, based on the geological exploration data obtained in step S1, constructs a normalized geological risk index function that is continuously distributed along the tunnel axis. This function expands the geological information of discrete borehole points into a continuous representation of geological risk at any location along the entire axis, providing input variables for subsequent dynamic spacing generation.

[0063] First, the tunnel's central axis needs to be parameterized. In this embodiment, the tunnel's central axis is a three-dimensional spatial curve, which can be represented by the parametric equation L(t), where t is a normalized mileage parameter, ranging from [0,1], with t=0 corresponding to the tunnel's starting point and t=1 corresponding to the tunnel's ending point. Let the total length of the tunnel axis be... Then the normalized parameter corresponding to any mileage position s .

[0064] The geological exploration data obtained in step S1 includes n discrete boreholes, each recording its mileage location. And the geological parameters at that location. For the i-th borehole, its mileage is converted into normalized parameters. The set of geological parameters for each borehole location includes at least four items: quantified value of the surrounding rock grade. Rock quality indicators Groundwater pressure and fracture density The surrounding rock grade is classified into grades I to VI according to the national standard GB / T50218-2014 "Engineering Rock Mass Classification Standard". In this embodiment, it is quantified into numerical values: Grade I = 1, Grade II = 2, Grade III = 3, Grade IV = 4, Grade V = 5, Grade VI = 6. Rock quality indicators Groundwater pressure is represented as a percentage (0~100). Fracture density is expressed in megapascals (MPa). The unit is per meter.

[0065] Because the physical meanings and dimensions of various geological parameters differ, they need to be normalized to unify the indicators within the [0,1] interval. Normalization employs extreme value linear transformation, requiring the minimum and maximum values ​​of each parameter to be pre-determined across the entire tunnel. Let the minimum value of the quantified surrounding rock grade in all boreholes be... The maximum value is The dimensionless value of the surrounding rock grade of the i-th borehole is... ;

[0066] For rock quality indicators, since a larger Q value indicates better rock quality and lower geological risk, a complementary form is used to represent the risk. ;

[0067] in and For all drill holes The minimum and maximum values. Groundwater pressure and fracture density are positively correlated with geological risk, therefore positive normalization is directly used: , ;in , These represent the minimum and maximum water pressure, respectively. , These represent the minimum and maximum values ​​of the fracture density, respectively.

[0068] Taking a certain tunnel as an example, assuming the total length is... =2000 meters, with a total of 5 boreholes. For borehole 1 (t=0.1), the surrounding rock grade is IV, i.e., G=4. The minimum surrounding rock grade for the entire tunnel is 3 (Grade III) and the maximum is 6 (Grade VI). The rock quality index Q=65, and the minimum Q for the entire tunnel is 30, while the maximum Q is 95. The groundwater pressure W = 0.35, and the minimum water pressure in the entire tunnel is 0.1, while the maximum is 0.8. Therefore... The fracture density F=2.5, with a minimum of 0.5 and a maximum of 5.0 throughout the tunnel. .

[0069] After obtaining the dimensionless values ​​of each parameter, they need to be combined into a single-point geological risk score. This embodiment uses a weighted summation method, with weighting coefficients preset based on engineering experience, reflecting the relative importance of each parameter's contribution to geological risk. According to statistics from numerous engineering cases, the surrounding rock grade has the most significant impact on drainage demand, followed by rock quality indicators, while groundwater pressure and fracture density are relatively localized. Therefore, the weighting coefficient is set as follows: Surrounding Rock Grade Rock quality indicators Groundwater pressure Crack density Note that the sum of the four weights is 1. Therefore, the single-point geological risk score for the i-th borehole is... Substituting the example data above, we get Repeat this calculation for all boreholes to obtain a set of discrete data points. .

[0070] Since geological risks should change continuously along the tunnel axis, and borehole data only provides information on limited locations, a continuous function r(t) needs to be constructed using interpolation methods. This embodiment uses cubic spline interpolation because cubic splines ensure the interpolation curve itself is continuous, the first derivative is continuous, and the second derivative is continuous, thus ensuring the smoothness of the geological risk function, conforming to the physical laws of gradual geological changes, and avoiding large oscillations caused by the Runge phenomenon. The specific interpolation process is as follows: For a node, the corresponding function value is A cubic polynomial is constructed between every two adjacent nodes to ensure that the function r(t) and its first and second derivatives are continuous across the entire interval. Natural spline conditions are used as boundary conditions, meaning the second derivative at both endpoints is zero. Cubic spline interpolation algorithms are well-established and can be implemented using numerical libraries (such as the `interp1d` function in Python's SciPy, with `kind='cubic'`). The internal formulas are not detailed here. When the number of boreholes is less than four, cubic spline interpolation may produce unreasonable oscillations due to insufficient data points. In this case, the system automatically downgrades to linear interpolation, connecting adjacent boreholes with straight lines. Simultaneously, the system outputs a prompt informing the user that the current borehole density is low, limiting the reliability of the interpolation results, and suggesting supplementing geological exploration data to improve model accuracy.

[0071] Through cubic spline interpolation, a continuous function r(t) defined on t∈[0,1] is obtained, representing the distribution of the original geological risk along the axis. However, the numerical range of r(t) may not be uniform. To facilitate the subsequent calculation of the response rule expression, it needs to be normalized to the [0,1] interval. First, the maximum value of the risk score for all borehole points is found. and minimum value ,Right now , Then, r(t) is linearly normalized: The normalized geological risk index function R(t) is obtained, with a range of [0, 1]. R(t) reflects the relative degree of geological risk at any location on the tunnel axis: R(t) = 0 indicates the lowest geological risk at that location (relative to the tunnel itself), and R(t) = 1 indicates the highest risk. This function will be used as an input variable in subsequent steps to calculate the dynamic spacing. It should be noted that if the single-point geological risk scores at all borehole locations are equal, i.e. If the value is zero, it indicates that the geological conditions of the entire tunnel are completely uniform. In this case, effective normalization is not possible, and the system directly sets R(t) = 0.5, indicating that the geological risk along the entire line is at a moderate level. Subsequent dynamic spacing will be equal to the foundation spacing. Simultaneously, the system outputs a prompt message reminding designers to verify the validity of the geological data.

[0072] Thus, step S3 completes the construction of a continuous normalized risk function from discrete geological exploration data. The output R(t) is stored in function form, and can also be discretized into a list of sampling points as needed, but this embodiment retains its function form for flexible use. The entire processing flow is clear: the input is the borehole location and its geological parameters, and the output is the continuous risk function R(t), achieved through normalization, weighted synthesis, spline interpolation, and global normalization. The mathematical definitions of each step are clear and there are no ambiguous concepts.

[0073] Step S4: Substitute the normalized geological risk index function as an input variable into the response rule expression, and combine it with the tolerance interval to generate a sequence of component spacing parameters that dynamically change along the target tunnel axis. The spacing value at each location in the component spacing parameter sequence is calculated by the normalized geological risk index at that location according to the response rule expression, and the spacing value is constrained within the allowable fluctuation range of the spacing.

[0074] Step S4, based on the tolerance range and linkage rules extracted in step S2 and the normalized geological risk index function constructed in step S3, generates a sequence of dynamically changing component spacing parameters along the tunnel axis. This sequence will serve as the direct input for subsequent parametric modeling, controlling the actual arrangement spacing of drainage components at different mileage locations.

[0075] First, the system extracts the base spacing (base), lower tolerance limit (low), upper tolerance limit (high), response rule expression string, and sensitivity coefficient (k) from the structured data obtained in step S2. In this embodiment, the response rule expression uniformly adopts a linear form, i.e., the one given in the example of step S2. Where base and k are known values, and R is the normalized geological risk index. The system parses this string into a computable mathematical function. , used for subsequent calculations.

[0076] Next, the continuous geological risk function needs to be discretized. The system divides the tunnel axis into a series of sampling points according to a preset step size Δt. The selection of the step size should take into account both calculation accuracy and efficiency. In this embodiment, Δt=0.01 is used, that is, sampling is performed once for every 1% of the tunnel length (equivalent to 10 meters for a 1000-meter tunnel). Since the normalized mileage parameter t ranges from [0,1], the number of sampling points... The specific calculation is as follows: The normalized mileage of each sampling point is: Where j = 0, 1, ..., N−1, corresponding to the actual mileage , This represents the total length of the tunnel axis.

[0077] Before performing the spacing calculation, the system first checks the reasonableness of the lower limit (low) and upper limit (high) of the tolerance interval. If low ≤ 0, it is automatically corrected to the minimum positive spacing allowed by the project (0.5 meters in this embodiment), and a warning message is output. If high is greater than 10% of the total tunnel length, it is automatically corrected to 10% of the total tunnel length, and a warning message is output. This is to prevent the generated spacing from exceeding the physically achievable range due to abnormal values ​​input by the user.

[0078] For each sampling point First, the normalized geological risk index function R(t) constructed in step S3 is called to obtain the risk value at that point. Then, substitute the values ​​into the response rule expression to calculate the initial spacing. The physical meaning of this expression is: when Rj=0.5 (geological risk is at a medium level), the initial spacing is equal to the foundation spacing; when... (High-risk area) ) is positive. That is, the spacing is reduced and the components are densified; when (in low-risk areas) ) is negative, This means that the spacing increases and the components become sparser. The sensitivity coefficient k determines the adjustment range of the spacing when the risk deviates from the moderate level; the larger k is, the more drastic the adjustment.

[0079] After the initial spacing calculation is completed, it needs to be limited to the tolerance range [low, high] to ensure that the results do not exceed the basic range described by the designer. The clamping processing rules are as follows: After clamping That is, the sampling point The final spacing value at that location.

[0080] The calculation process is illustrated using a specific tunnel as an example. Assume the total length of the tunnel is... Meters, base spacing Meters, lower limit of the tolerance range meters, upper limit Meters, sensitivity coefficient Step S3 has constructed the normalized geological risk function R(t). Now, regarding... The calculation is performed at a sampling point (corresponding to a mileage of 400 meters). Let this point be... The initial spacing Meters. Since 9.0 meters is within the interval [5, 15], the final distance d = 9.0 meters. If at another point t = 0.8, R(0.8) = 0.9, then Meters, still within the interval, are taken as 8.0 meters. If R is very low at a certain point, for example, R=0.1, then The distance is still within the range. If R = 1.0, then d′ = 10.0 × (1 − 0.5 × 0.5) = 7.5 meters; if R = 0, then d′ = 10.0 × (1 − 0.5 × (− 0.5)) = 12.5 meters. These are all within the limits. Assume the calculated... If the value is less than 5 or greater than 15, it will be clamped to 5 or 15.

[0081] Repeat the above calculation for all j = 0, 1, ..., 100 to obtain a series Data pairs. Arranged in ascending order of mileage, they form a sequence of component spacing parameters. .

[0082] Each element in this sequence explicitly specifies the component spacing value to be used at the corresponding mileage location. Since the sampling step size is fixed, if the spacing value for any intermediate mileage is needed during actual modeling, it can be obtained through linear interpolation in subsequent steps. Thus, step S4 completes the transformation from geological risk functions and semantic rules to specific spacing parameters, and the output data stream is an ordered spacing sequence, providing a precise layout basis for the 3D modeling of the drainage network.

[0083] Step S5: Based on the component spacing parameter sequence, drive the parametric modeling program to generate a 3D model of the drainage pipe network of the target tunnel.

[0084] Step S5, based on the component spacing parameter sequence generated in step S4, drives the parametric modeling program to generate a 3D model of the drainage network of the target tunnel. This step transforms the abstract spacing data into concrete spatial geometric objects, providing a model foundation for subsequent hydraulic analysis.

[0085] First, the system needs to obtain the geometric parameters of the tunnel's central axis and inner contour line. The tunnel's central axis is a three-dimensional spatial curve, determined in step S1 or S3, and can be represented by the parametric equation C(t), where t∈[0,1] is the normalized mileage parameter, t=0 corresponds to the tunnel's starting point, and t=1 corresponds to the tunnel's ending point. The tunnel's inner contour line is a closed curve representing the tunnel's cross-sectional shape, typically stored as a set of points or a parametric curve on a two-dimensional plane, located in a local coordinate system perpendicular to the central axis. In this embodiment, the inner contour line adopts a standardized design, for example, a combination of circular arcs and straight line segments with radius R; specific data is provided by the design drawings.

[0086] Component spacing parameter sequence It has been generated in step S4, where This represents the actual mileage (in meters) of the j-th sampling point. This represents the expected spacing at that point (in meters). Note that... With normalized parameters The relationship is , This represents the total length of the tunnel axis.

[0087] The generation of the 3D model of the drainage pipe network is a process of gradually arranging components along the tunnel axis. The system initializes the current mileage position. This indicates that the first component is placed starting from the tunnel's origin, and the component counter m is initialized to 0 to record the number of components already generated. Then, a loop is entered: when... Repeat the following sub-steps:

[0088] Sub-step 1: Determine the expected spacing corresponding to the current mileage position. Since the component spacing parameter sequence D only stores the spacing values ​​of discrete sampling points, and the current mileage... It may not coincide with any sampling point, therefore the distance at the current position needs to be calculated through interpolation. The system searches for the corresponding point in sequence D. Two adjacent sampling points and ,satisfy .like If the distance is exactly equal to the mileage of a certain sampling point, then the distance value of that point is directly taken; otherwise, the linear interpolation formula is used: This formula ensures that the spacing changes continuously with mileage and is consistent with the values ​​of adjacent sampling points. For example, assuming... meters, adjacent sampling points are and The interpolation result is Meters; when searching for adjacent sampling points, if... If an anomaly occurs (usually due to data duplication), skip the duplicate point and continue searching for the next valid sampling point, ensuring that the denominator of the interpolation calculation is not zero. If there is only one sampling point for the entire tunnel, directly take the spacing value of that point as the expected spacing for the entire line.

[0089] Sub-step 2: Determine the pose matrix of the tunnel cross-section at the current mileage position. The normalized parameter corresponding to the tunnel centerline C(t) at mileage pos is... .

[0090] The position vector of this point can be calculated using the parametric equations. The coordinates of the point and the tangent vector T (i.e., the axis direction) are also considered. To construct the local coordinate system of the cross section, the normal vector N and the binormal vector B also need to be determined. Typically, the cross product of the global vertical upward direction (e.g., the z-axis) and the tangent vector can be chosen as the normal vector, and the binormal vector is obtained through the cross product of the tangent vector and the normal vector, thus forming an orthogonal frame. The pose matrix M is a 4×4 homogeneous transformation matrix. Its upper left 3×3 sub-block consists of the rotation part composed of the three column vectors T, N, and B, and its upper right 3×1 sub-block is the position vector P. The last row takes the values ​​[0, 0, 0, 1]. This matrix is ​​used to transform the geometry from the local coordinate system to the global coordinate system.

[0091] Sub-step 3: Generate the 3D geometric object of the drainage component at the current mileage location. The system predefines templates for various drainage components; for example, a circumferential duct is an arc along its inner contour, and a transverse duct is a straight line segment. The templates are stored in a local coordinate system, and their position and orientation are related to the cross-section. Taking the circumferential duct as an example, the template is an arc curve located on the yz plane (assuming the axis direction is x). The system multiplies the coordinates of each point on the template by the pose matrix M and transforms it to the global coordinate system to obtain the actual position and orientation of the component in the tunnel. When generating the geometric object, metadata such as the component's unique identifier, type, and spatial coordinate range are recorded and stored in the component list.

[0092] Sub-step 4: Update the current mileage position. Add the current mileage to the expected spacing to obtain the mileage at which the next component should be placed: Meanwhile, the component counter m = m + 1. The loop returns to sub-step one to continue placing subsequent components.

[0093] when Reaching or exceeding the tunnel end The loop terminates at this point. At this point, all drainage components covering the entire tunnel have been generated. Finally, the system combines all recorded 3D geometric objects of the components to form a complete 3D model data file of the drainage network, typically stored in a common 3D format (such as STEP, IGES, or a custom BIM format) for subsequent hydraulic analysis modules to use.

[0094] It is worth noting that in actual engineering projects, drainage pipe networks contain various types of components (circumferential ducts, longitudinal ducts, transverse ducts, etc.), and their arrangement rules may differ. In this embodiment, it is assumed that all components are arranged using the same spacing sequence D, or that component types are distinguished by other parameters. If it is necessary to distinguish different types, a preset component type mapping table can be queried in the loop based on the current mileage position, and the corresponding template can be called to generate the component type. For the sake of simplicity, this embodiment uses a single type of component as an example; multiple types can be implemented through extension without affecting the core logic.

[0095] At this point, step S5 completes the transformation from the spacing parameter sequence to the three-dimensional model. The output three-dimensional model of the drainage network has clear geometric information and topological relationships, and can be directly used for subsequent hydraulic analysis and visualization rendering.

[0096] After generating the three-dimensional model of the drainage network in step S5, this embodiment further performs hydraulic analysis on the model and corrects the model parameters based on the analysis results to improve the reliability and safety of the drainage design.

[0097] First, the system converts the 3D model of the drainage network output in step S5 into an input format recognizable by the hydraulic calculation software. This embodiment uses EPANET as the hydraulic analysis kernel; therefore, the model's topology (nodes, pipe segments) and geometric parameters (pipe length, pipe diameter, roughness) need to be organized according to the EPANET INP file format. Specifically, the connection points of drainage components are defined as nodes, and the components themselves are defined as pipe segments. Node coordinates and pipe segment lengths are extracted from the component geometric information recorded in step S5. Furthermore, the equivalent diameter and roughness coefficient of each pipe segment need to be set according to design specifications. After format conversion, the system calls the EPANET solver to perform hydraulic calculations, obtaining the water pressure value for each node and the flow rate and velocity value for each pipe segment. For ease of subsequent processing, the system stores the calculation results in a data frame format by node and pipe segment index, where each node corresponds to a water pressure value (unit: MPa), and each pipe segment corresponds to a flow rate value (unit: liters per second) and a velocity value (unit: meters per second).

[0098] When converting a 3D model of a drainage network into an EPANET input file, the following mapping rules should be followed:

[0099] The endpoints of each drainage component are defined as nodes, and are assigned globally unique numbers in ascending order of mileage (starting from 1).

[0100] The components between adjacent nodes are defined as pipe segments, and the pipe segment numbers correspond one-to-one with the components.

[0101] The pipe section length is extracted directly from the three-dimensional geometry, the pipe diameter is read from the design parameter table according to the component type, and the roughness coefficient is preset to 0.014 according to the specification.

[0102] Boundary condition settings: Nodes at the tunnel entrance and exit are set as free outflow boundaries, and other nodes have no special boundaries;

[0103] If the spacing parameter sequence generated in step S4 results in a local pipe segment length of less than 0.5 meters, then the pipe segment is merged with the adjacent pipe segment to avoid generating excessively short pipe segments that could affect the stability of the calculation.

[0104] After obtaining the hydraulic analysis results, the system executes the abnormal section identification process. First, three safety thresholds are preset: the first safety threshold corresponds to the minimum allowable flow rate, denoted as... In this embodiment, 0.5 liters per second is used; the second safety threshold corresponds to the minimum allowable flow rate, denoted as... Take 0.6 meters per second (self-cleaning flow rate requirement); the third safety threshold corresponds to the maximum allowable water pressure, denoted as... Take 0.5 MPa (upper limit of lining bearing capacity). For each pipe section, if its flow rate is lower than... Or its flow rate value is lower than If so, the mileage interval corresponding to that pipe segment is marked as an "abnormal section with insufficient drainage capacity". The mileage interval corresponding to a pipe segment can be determined based on the mileage positions of its two endpoints: Let the mileages of the two endpoints of the pipe segment be respectively... and Then the mileage range covered by this pipeline segment is For each node, if its water pressure value is higher than... Then, the mileage location corresponding to the node is marked as a "water pressure over-limit abnormal point", and the distance is extended to both sides of the node by a preset length. (In this embodiment, a 5-meter section is used) to form an abnormal section with excessive water pressure. All marked abnormal sections are merged; that is, if multiple abnormal sections overlap or are adjacent, they are merged into a single continuous abnormal section, ultimately resulting in a set of hydraulically abnormal sections. ,in and These are the starting and ending mileages of the l-th abnormal segment, respectively.

[0105] Next, the feedback and correction process is executed. The system iterates through each segment in the abnormal segment set E. Calculate the length of this section. Set a first length threshold. Second length threshold In this embodiment, 10 meters and 50 meters are used respectively. If... If the anomaly is identified, the section is determined to be a local anomaly, possibly caused by a sudden geological change or missing borehole data. For this type of anomaly, the center mileage of the section is determined. This serves as a virtual borehole point, and a correction risk score for the virtual borehole point is set based on the degree of deviation of hydraulic indicators within that section. The degree of deviation can be measured by the average deviation of the flow rate or velocity in each pipe section within that section from the safety threshold, for example, by defining a deviation coefficient. (Only for sections with insufficient traffic); then adjust the risk score. ;in This represents the original value of the current geological risk field at this point. An adjustment factor is used (0.5 in this embodiment). The result is... Then, it is merged with the original geological exploration data, and the interpolation and normalization process in step S3 is repeated to obtain the corrected normalized geological risk index function. .

[0106] If multiple consecutive abnormal segments exist and their cumulative coverage length exceeds the second length threshold. If the result is negative, it is determined to be a systematic anomaly, possibly caused by an improper sensitivity coefficient setting. In this case, it is necessary to extract the set T of all sampling point locations within the continuous anomaly region. The method for determining the continuous anomaly region is as follows: merge adjacent anomaly segments (if the interval is less than a preset merging threshold, such as 10 meters), resulting in several continuous anomaly regions. Select the region whose cumulative length exceeds [a certain threshold]. The region. For each sampling point t∈T, the ideal spacing value required to satisfy the safety threshold at that point needs to be calculated. The reverse calculation method is based on the hydraulic model: assuming other parameters remain constant, only the pipe spacing near the point is adjusted, and the minimum spacing required to restore the hydraulic parameters to the safe threshold is solved iteratively using a bisection method or empirical formula. This embodiment uses a simplified method, approximating the spacing based on the inverse relationship between flow rate and pipe length: Let the current spacing at this point be... (This can be obtained from the spacing sequence generated in step S4), the current flow rate is... The target traffic is Then the ideal spacing (If the flow rate is insufficient); the simplified formula above is based on the assumption of constant flow rate and is suitable for local adjustments where the hydraulic state of the pipe section does not change significantly. If a higher precision back-calculation is required, the system can call the hydraulic solver for iterative calculation: while keeping other parameters constant, gradually adjust the length of the pipe section near the sampling point (by changing the position of the components), and repeatedly perform hydraulic analysis until the result meets the safety threshold. For computational efficiency, this embodiment uses the simplified formula by default; users can select the high-precision mode and start iterative calculation as needed. At the same time, obtain the normalized geological risk index Rt at this point (obtained from R(t) in step S3). Then, with the goal of minimizing the sum of squared deviations between the current spacing and the ideal spacing of all sampling points, the sensitivity coefficient k is optimized using the least squares method. The specific optimization model is as follows: This optimization problem can be solved analytically by solving a system of linear equations, or by using numerical optimization algorithms (such as gradient descent). The corrected sensitivity coefficients are then obtained. Then, substitute it into the response rule expression; set the maximum number of iterations to 3, and set a convergence condition: if the total length of the abnormal segment after the current iteration is less than 5% of the previous iteration, then it is considered to have converged and the iteration stops. If the maximum number of iterations is reached but the convergence condition is still not met, the system outputs the current model and adds a warning message, prompting the designer to manually review it.

[0107] After completing the above corrections, the system will use the corrected normalized geological risk index function. Or the corrected sensitivity coefficient As input, step S4 is re-executed to generate a new sequence of component spacing parameters, which drives step S5 to update the 3D model of the drainage network. If both local and systemic anomalies exist simultaneously, local anomalies can be processed first (by supplementing virtual borehole points) to obtain a new risk function, and then the sensitivity coefficient can be optimized based on the new risk function.

[0108] Example 2: Figure 2 As shown, the automated modeling and analysis method for highway tunnel drainage networks, which incorporates natural language processing, also includes:

[0109] After constructing the normalized geological risk index function in step S3 and before generating the component spacing parameter sequence in step S4, this embodiment also performs a pre-verification process. This process aims to evaluate the degree of matching between the linkage rules extracted from natural language and the actual risk distribution revealed by geological exploration data, and to dynamically correct the applicable mileage range of the linkage rules based on the matching results, so as to avoid model distortion caused by the decoupling of semantic rules from actual geological conditions.

[0110] Specifically, the system first obtains at least one geological condition type from the linkage rules extracted in step S2. The linkage trigger condition field is stored as a list of strings, such as ["fault fracture zone", "water-rich area"]. Each element in the list represents a geological condition type. For each geological condition type, the system needs to preset a corresponding risk threshold. The risk threshold is set based on engineering experience and relevant standards: for example, for "fault fracture zone", sections with a normalized geological risk index greater than or equal to 0.7 are generally considered high-risk areas, therefore a threshold is set. For "water-rich areas," sections where groundwater pressure contributes significantly, a [specific action / design] can be set... It should be noted that the normalized geological risk index is a comprehensive indicator, and the impact of a single parameter (such as water pressure) on the final value is limited. If the user explicitly requires the identification of water-rich areas based on specific geological parameters (such as water pressure alone), a 'dedicated threshold mode' can be specified in the system configuration. In this mode, the system no longer uses the comprehensive risk index, but directly performs threshold judgments on the normalized individual parameters. This embodiment uses the comprehensive risk index by default to ensure consistent processing with other geological conditions. These thresholds can be pre-stored in the system configuration file, or manually entered by the user before modeling according to the specific engineering situation. This embodiment uses the former, i.e., preset fixed thresholds.

[0111] Based on the normalized geological risk index function R(t) constructed in step S3, the system scans along the tunnel axis and marks continuous mileage intervals where R(t) is greater than or equal to a preset risk threshold as actual risk sections corresponding to that geological condition type. The specific determination method is as follows: the normalized mileage t is discretized with a small step size (e.g., Δs=0.001), and R(t) is calculated for each discrete point. If R(t) ≥ the threshold at a certain point, then that point is marked as a risk point; consecutive adjacent risk points are merged into continuous intervals to obtain a list of actual risk sections. Taking a fault fracture zone as an example, assuming a threshold of 0.7, the scan finds that R(t) in the intervals t∈[0.2,0.35] and t∈[0.6,0.65] is not less than 0.7. These two intervals are the actual risk sections corresponding to the fault fracture zone.

[0112] Meanwhile, the system needs to obtain the default applicable mileage range of the linkage rules. This default applicable mileage range has two possible sources: one is the entire tunnel axis, meaning the linkage rules are assumed to apply to the entire tunnel by default; the other is determined by the mileage range implicit in the fuzzy semantic description. For the latter, for example, if the user input mentions "near the fault fracture zone in the middle section of the tunnel," then the specific mileage range corresponding to "middle section" needs to be interpreted from the semantics, such as K10+200 to K10+800. In this embodiment, if the linkage rules extracted in step S2 do not include explicit mileage limitations, then the default applicable mileage range is the entire tunnel axis, i.e., t∈[0,1].

[0113] For each geological condition type, the system calculates the overlap index between its default applicable mileage interval and the actual risk section. The overlap index is quantified by the ratio of the overlap length to the total union length. Let the default applicable mileage interval be represented on the normalized mileage as an interval. The actual risk zone consists of a series of sub-intervals. The union of the default interval and each actual risk segment. First, calculate the total length of the overlap between the default interval and each actual risk segment: Then calculate the total length of the union of the default interval and the actual risk interval: The overlap index is defined as: The value of M ranges from [0,1]. M=1 indicates that the default interval completely covers the actual risk area and does not exceed it; M=0 indicates that the default interval has no overlap with the actual risk area.

[0114] Taking a fault fracture zone as an example, assuming the default applicable mileage interval is the entire tunnel t∈[0,1], i.e., A=[0,1]. The actual risk sections are t∈[0.2,0.35] and t∈[0.6,0.65], then the overlap length... Default interval length The actual total length of the risk zone is 0.15 + 0.05 = 0.2, therefore... The overlap index M = 0.2 / 1.0 = 0.2.

[0115] The system presets a matching threshold, for example The calculated overlap ratio is compared with this threshold. If... If so, it is considered that the linkage rule matches the actual geological situation well, and the original rule will be followed subsequently. If so, the matching degree prompt information generation process will be triggered.

[0116] The matching degree prompt includes the following: the name of the geological condition type, the default applicable mileage range (in actual mileage, such as K10+000 to K12+000), the list of actual risk sections (in actual mileage), and the overlap index value (such as 0.2). This prompt is generated in text form and stored as an additional output along with the subsequent component spacing parameter sequence. It can be reviewed by designers after modeling is completed to determine the applicability of semantic rules.

[0117] Furthermore, when the overlap index is less than a preset matching threshold, the system can automatically execute a correction process. The goal of the correction is to trim the applicable mileage range of the linkage rule according to the actual risk area, so that the rule only takes effect in areas where there is actual geological risk. The corrected applicable mileage range is determined as the intersection of the default applicable mileage range and the actual risk area: This means that only the portion of the default interval that overlaps with each actual risk zone is retained. If the intersection is empty, it means that the linkage rule is completely inapplicable, and the system can output a warning and suggest re-examining the semantic description or geological data.

[0118] Taking the above example as an example, the corrected applicable mileage range is: After being converted to actual mileage, for example, if the total length of the tunnel is 2000 meters and the starting point is K10+000, then it corresponds to K12+000 to K12+300 (assuming 0.2 corresponds to K12+000 and 0.35 corresponds to K12+300) and K12+800 to K12+900 (assuming 0.6 corresponds to K12+800 and 0.65 corresponds to K12+900).

[0119] When generating the component spacing parameter sequence in subsequent step S4, the system will perform differentiated calculations based on the modified applicable mileage interval: within the modified applicable mileage interval, the spacing value is dynamically generated according to the response rule expression combined with the geological risk index; outside the modified applicable mileage interval, i.e., in the default interval but in sections where the geological conditions do not actually exist, the component spacing parameter directly adopts the foundation spacing base, without being controlled by the linkage rule. For example, in the sections t∈[0, 0.2), t∈(0.35, 0.6), and t∈(0.65, 1] ​​in the above example, the spacing is always equal to the foundation spacing of 10 meters, and does not fluctuate with geological risk.

[0120] This correction mechanism ensures that the scope of the linkage rules is strictly aligned with the actual geological risk distribution, avoiding design deviations caused by ambiguity or overgeneralization in semantic description. The corrected applicable mileage interval will be passed to step S4 as a constraint for generating the spacing parameters. At this point, the entire pre-verification and correction process is complete, and the system continues to execute step S4 and subsequent steps.

[0121] In this embodiment, the above process needs to be inserted after step S3 and before step S4. Specifically, after the normalized geological risk index function is constructed and R(t) is output in step S3, the above matching degree calculation and correction process is executed immediately. The corrected linkage rule information (including the corrected applicable mileage interval, basic spacing, tolerance interval, response rule expression, etc.) is used as the input of step S4 to ensure that the subsequent spacing generation is based on the corrected rule execution.

[0122] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0123] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0124] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An automated modeling and analysis method for highway tunnel drainage networks using natural language processing, characterized in that: Includes the following steps: Acquire a first interactive command containing a fuzzy semantic description of the arrangement parameters of the drainage components of the target tunnel, as well as geological exploration data of the target tunnel distributed along its axis. The fuzzy semantic description is semantically interpreted to extract the tolerance range and the linkage rules between the tolerance range and geological conditions. The tolerance range includes the basic spacing and the allowable fluctuation range of the spacing. The linkage rules include at least one type of geological condition for triggering the linkage rules and a response rule expression describing how the spacing changes with geological risk. Based on the geological exploration data, a normalized geological risk index function is constructed that is continuously distributed along the axis of the target tunnel. The normalized geological risk index function is used to characterize the degree of geological risk at each location on the tunnel axis. The normalized geological risk index function is substituted into the response rule expression as an input variable, and combined with the tolerance interval, to generate a sequence of component spacing parameters that dynamically change along the target tunnel axis. The spacing value at each location in the component spacing parameter sequence is calculated by the normalized geological risk index at that location according to the response rule expression, and the spacing value is constrained within the allowable fluctuation range of the spacing. Based on the sequence of component spacing parameters, the parametric modeling program is driven to generate a three-dimensional model of the drainage network of the target tunnel.

2. The automated modeling and analysis method for highway tunnel drainage pipe networks combining natural language as described in claim 1, characterized in that, The extraction process of the tolerance interval implicit in the fuzzy semantic description and the linkage rules between the tolerance interval and geological conditions includes: The first interactive instruction is concatenated with a preset system prompt and then input into the large language model. The system prompt includes constraints on the output data format and example descriptions of the tolerance range and linkage rules. Receive text data returned by the large language model, and extract a structured string conforming to JSON format from the text data; The structured string is decoded into JSON to obtain structured data containing a basic spacing field, a lower limit tolerance range field, an upper limit tolerance range field, a linkage trigger condition field, a response rule expression field, and a sensitivity coefficient field. The value of the basic spacing field is the basic spacing, the value of the lower limit tolerance range field is the lower limit of the allowable fluctuation range of the spacing, the value of the upper limit tolerance range field is the upper limit of the allowable fluctuation range of the spacing, the value of the linkage trigger condition field is a list of one or more geological condition types, the value of the response rule expression field is a mathematical expression with the basic spacing, the sensitivity coefficient, and the normalized geological risk index as independent variables, and the value of the sensitivity coefficient field is a value in the interval [0,1].

3. The automated modeling and analysis method for highway tunnel drainage pipe networks combining natural language as described in claim 1, characterized in that, The process of constructing the normalized geological risk index function continuously distributed along the target tunnel axis includes: obtaining the parameterized representation L(t) of the tunnel center axis, where t∈[0,1] is the normalized mileage parameter, t=0 corresponds to the tunnel start point, and t=1 corresponds to the tunnel end point; Obtain the mileage parameter values ​​corresponding to n discrete borehole locations. and a set of geological parameters at each borehole location, the set of geological parameters including at least the quantified value of the surrounding rock grade. Rock quality indicators Groundwater pressure and fracture density ; Normalize the geological parameters at each borehole location to obtain dimensionless values ​​of each parameter in the interval [0,1]. ; According to the preset weighting coefficients , , , The dimensionless values ​​of various parameters at the same borehole location are weighted and summed to obtain the single-point geological risk score for that location. ; Based on the location of each borehole ( , Using the known point as an example, a cubic spline interpolation method is used to construct the original risk function r(t) that is continuous along the tunnel axis, where t∈[0,1]. Obtain the maximum value of the single-point geological risk score at all borehole locations. and minimum value The original risk function is linearly normalized to obtain the normalized geological risk index function. The range of the normalized geological risk index function is [0,1].

4. The automated modeling and analysis method for highway tunnel drainage pipe networks combining natural language as described in claim 3, characterized in that, The weighting coefficient , , , The values ​​are 0.4, 0.3, 0.2, and 0.

1.

5. The automated modeling and analysis method for highway tunnel drainage networks combining natural language as described in claim 1, characterized in that, The process of generating the sequence of component spacing parameters that dynamically change along the target tunnel axis includes: discretizing the tunnel axis into N sampling points according to a preset step size Δt. j = 0, 1, ..., N−1, where ; For each sampling point Obtain the normalized geological risk index at this point. ; Will Substituting the values ​​into the response rule expression, the preliminary spacing value of the sampling point is calculated. The response rule expression is: , where base is the base spacing and k is the sensitivity coefficient; Obtain the lower limit of the allowable fluctuation range of the spacing. and The initial spacing value is clamped to obtain the final spacing value. : ; The mileage location corresponding to each sampling point Compared with the final spacing value Related, among which The total length of the tunnel axis is arranged in ascending order of mileage to form the sequence of component spacing parameters. .

6. The automated modeling and analysis method for highway tunnel drainage pipe networks combining natural language as described in claim 5, characterized in that, The process of generating the three-dimensional model of the drainage pipe network of the target tunnel includes: Obtain the geometric parameters of the tunnel's central axis and the tunnel's inner contour line. The tunnel's central axis is a three-dimensional spatial curve, and the tunnel's inner contour line is a closed curve representing the tunnel's cross-sectional shape. Initialize current mileage location And component counter m=0; when Repeat the following steps: Search for the component spacing parameter sequence D. Two adjacent sampling points and ,satisfy The expected distance value corresponding to the current mileage position is calculated by linear interpolation. : Based on current mileage location Using the parametric equations of the tunnel's central axis, calculate the pose matrix of the tunnel cross section at that location, where the pose matrix includes rotational and translational components. In the local coordinate system determined by the pose matrix, based on the geometric parameters of the tunnel contour and the component type corresponding to the current mileage position, a three-dimensional geometric object of the drainage component at that position is generated, and the spatial coordinates of the component are recorded. Update the position of the next component The component counter m = m + 1; After the loop ends, all the generated three-dimensional geometric objects of the drainage components are combined to form the three-dimensional model of the drainage network.

7. The automated modeling and analysis method for highway tunnel drainage pipe networks combining natural language as described in claim 1, characterized in that, Also includes: Hydraulic analysis is performed on the generated three-dimensional model of the drainage network to obtain hydraulic analysis result data. The hydraulic analysis result data includes the first hydraulic index value, the second hydraulic index value, and the third hydraulic index value of each pipe section and each node. The first hydraulic index value is the flow rate, the second hydraulic index value is the flow velocity, and the third hydraulic index value is the water pressure. Obtain a preset first safety threshold, a second safety threshold, and a third safety threshold; compare the first hydraulic index value with the first safety threshold, compare the second hydraulic index value with the second safety threshold, and compare the third hydraulic index value with the third safety threshold. For each pipe section, if the first hydraulic index value of the pipe section is lower than the first safety threshold or the second hydraulic index value of the pipe section is lower than the second safety threshold, then the mileage interval corresponding to the pipe section is marked as an abnormal section with insufficient drainage capacity. For each node, if the third hydraulic index value of the node is higher than the third safety threshold, the mileage position corresponding to the node is marked as a water pressure over-limit abnormal point, and a water pressure over-limit abnormal section is formed by extending a preset length to both sides of the node as the center. All marked anomalous sections are merged to obtain a set of hydraulic anomaly sections. Each abnormal segment is defined by a starting mileage and an ending mileage.

8. The automated modeling and analysis method for highway tunnel drainage pipe networks combining natural language as described in claim 7, characterized in that, Also includes: Obtain the set E of hydraulically abnormal sections, and for each abnormal section Calculate the length of this segment. ; like If the length is less than the first length threshold, the abnormal segment is determined to be a local anomaly, and the center mileage of the abnormal segment is recorded. As a virtual borehole point, a correction risk score is set for the virtual borehole point based on the degree of deviation of hydraulic indicators within that section. The virtual drilling point After merging with the existing geological exploration data, the normalized geological risk index function is recalculated; If multiple consecutive anomalous segments exist and their cumulative coverage length exceeds the second length threshold, then extract the set T of all sampling point locations within the consecutive anomalous area. For each sampling point t∈T, the ideal spacing value required to satisfy the safety threshold at that sampling point is deduced based on the hydraulic analysis results. Get the current generated spacing value at the sampling point. and the normalized geological risk index at the sampling point To minimize all sampling points The optimization problem is solved using the least squares method with the sum of the possible values ​​as the objective. The corrected sensitivity coefficient is obtained. Substitute the modified normalized geological risk index function or the modified sensitivity coefficient into the response expression to regenerate the component spacing parameter sequence and drive the parametric modeling program to update the three-dimensional model of the drainage network.

Images