A railway culvert design drawing system and method based on a hybrid intelligent algorithm

Abstract

The application discloses a railway culvert design drawing system and method based on a hybrid intelligent algorithm. The prior art lacks the ability to meet the design requirements of railway culverts, such as rapidness, standardization and high reliability. The application comprises a data input layer, a pre-trained model layer, an engineering quantity calculation layer and a CAD drawing layer. First, an M1 unsupervised anomaly detection model and an M2 reinforcement prediction model are constructed. Then, parameter input and preliminary verification are performed. The M1 unsupervised anomaly detection model is used for anomaly detection and correction, and the rationality of the parameter combination is identified by the pre-trained M1 model. The M2 reinforcement prediction model is used to generate a culvert reinforcement scheme. Based on the modified and confirmed design parameters, the automatic accounting of the engineering quantity is realized. Finally, the structure parameters, load parameters, reinforcement parameters and the temporarily stored engineering quantity of the railway culvert are taken as inputs to generate the output of the railway culvert standard drawing and the engineering quantity. The application improves the design efficiency and reduces the design and drawing difficulty of the railway culvert.

Description

Technical Field

[0001] This invention belongs to the field of intelligent design technology for railway culverts, specifically relating to a railway culvert design and drawing system and method based on a hybrid intelligent algorithm. Background Technology

[0002] As a core ancillary structure in railway engineering, railway box culverts must be designed in strict accordance with the "Railway Culvert Design Code" (TB 10003-2016). In the traditional design process, designers need to manually determine the structural dimensions and stress parameters based on experience and specifications, calculate the reinforcement scheme, and finally use CAD to draw the drawings. This process has the following technical defects:

[0003] 1. Parameter verification process: The parameters such as aperture, net height, backfill height, and load input by the designers need to be manually verified against the specifications one by one. It is easy to miss abnormal parameters due to insufficient experience (such as backfill height exceeding 10m or top plate thickness less than 0.2m). In addition, there may be inconsistencies between multiple parameters, leading to parameter imbalance. This is especially difficult for junior designers.

[0004] 2. Reinforcement calculation process: The area of ​​main and auxiliary reinforcement bars in the top slab / web / bottom slab of the box culvert needs to be manually verified using finite element software (such as Midas / GTS) or empirical formulas. The calculation time for a single scheme is ≥8 hours, and the deviation rate of calculation results from different designers can reach 15%-20%.

[0005] 3. CAD drawing process: After the dimensions and reinforcement scheme are determined, it is necessary to manually draw the structural dimension drawings, cross-sectional drawings, reinforcement layout drawings, dimension annotations, notes, and quantity calculations in CAD. The drawing time for a single drawing is ≥4 hours, and the entire process needs to be redrawn after the parameters are modified, resulting in extremely low iteration efficiency.

[0006] 4. Insufficient adaptation to intelligent technologies: Existing deep learning-based culvert parameter optimization systems rely heavily on a large amount of labeled data, resulting in severe black-box modeling. They also focus only on parameter optimization and fail to achieve a closed-loop process of "parameter verification - reinforcement prediction - drawing". In addition, such systems require GPU computing power, making them difficult to adapt to the lightweight deployment requirements of railway engineering site desktops.

[0007] In summary, existing technologies lack an integrated system that takes into account parameter compliance verification, accurate reinforcement prediction, automated CAD drawing, and engineering quantity calculation, and therefore cannot meet the design requirements of railway culverts for speed, standardization, and high reliability. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention provides a railway culvert design and drawing system and method based on a hybrid intelligent algorithm, which realizes a closed-loop process of intelligent verification of culvert design parameters, automatic prediction of reinforcement, and automated drawing. This not only improves design efficiency and quality but also reduces the difficulty of railway culvert design and drawing.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] A railway culvert design and drawing method based on a hybrid intelligent algorithm, characterized by the following steps:

[0011] Step 1: Construct an unsupervised anomaly detection model for M1 and a reinforcement prediction model for M2;

[0012] Step 2: Parameter input and preliminary verification;

[0013] Step 3: Anomaly detection and correction using the M1 unsupervised anomaly detection model, identifying the rationality of parameter combinations through a pre-trained M1 model;

[0014] Step 4: Generate the culvert reinforcement scheme using the M2 reinforcement prediction model;

[0015] Step 5: Based on the modified and confirmed design parameters, automate the calculation of project quantities;

[0016] Step 6: Based on the structural parameters, load parameters, reinforcement parameters, and temporarily stored quantities of the railway culvert obtained in Steps 2, 3, 4, and 5, generate standard drawings of the railway culvert and output the quantities of the culvert.

[0017] Furthermore, step 1 includes the following steps:

[0018] S1.1: Acquisition and processing of training data, generating training and test sets;

[0019] S1.2: Model Training and Saving: The Isolation Forest algorithm is used to train the M1 unsupervised anomaly detection model; the XGBoost multi-output regression algorithm is used to train the M2 reinforcement prediction model. A labeled dataset is constructed with reinforcement parameters as output labels, and the two trained models are saved.

[0020] Furthermore, step 2 specifically includes:

[0021] S2.1: Designers can input structural parameters and stress parameters one by one, or input them in batches through a file.

[0022] S2.2: Perform preliminary validation on the input parameters.

[0023] Furthermore, step 3 specifically includes:

[0024] S3.1: After standardizing the culvert design parameters obtained in the parameter input stage, input them into the M1 unsupervised anomaly detection model; the model uses a weighted sampling algorithm to calculate the anomaly score of the parameter combination;

[0025] S3.2: Based on the processing result of step S3.1, if the anomaly score is greater than the preset engineering safety threshold, calculate the SHAP value of each input parameter and sort them in descending order of anomaly degree, and output a clear correction prompt to the user so that subsequent modifications can be made;

[0026] S3.3: After correcting the abnormal parameters based on the prompts, repeat S3.1-S3.2 until the abnormal scores of all culvert parameters are less than or equal to the preset engineering safety threshold.

[0027] Furthermore, step 4 specifically includes:

[0028] S4.1: Import the culvert design parameters output after anomaly detection and correction in step 3 into the pre-trained M2 reinforcement prediction model; the M2 reinforcement prediction model calls the pre-stored railway culvert reinforcement training weights, which are trained based on the labeled dataset.

[0029] S4.2: The M2 reinforcement prediction model outputs 9 types of reinforcement prediction parameters.

[0030] Furthermore, step 6 specifically includes:

[0031] S6.1: CAD Environment Initialization and Parameter Parsing;

[0032] S6.2: Batch drawing of standard drawings;

[0033] S6.3: Drawing output;

[0034] S6.4: Drawing optimization and saving.

[0035] A railway culvert design and drawing system based on hybrid intelligent algorithms, characterized in that:

[0036] It includes a data input layer, a pre-trained model layer, an engineering quantity calculation layer, and a CAD drawing layer.

[0037] The data input layer allows designers to input railway culvert design parameters and performs preliminary format verification on the input parameters;

[0038] The pre-trained model layer calls the pre-trained hybrid intelligent model, which includes the M1 unsupervised anomaly detection model and the M2 reinforcement prediction model.

[0039] The engineering quantity calculation layer constructs an engineering quantity calculation module to calculate the engineering quantities related to the culvert and outputs the relevant quantities to the CAD drawing layer.

[0040] The beneficial effects of this invention are:

[0041] 1) This invention shortens the entire process of traditional culvert design, which involves "parameter verification (0.5 days) + reinforcement calculation (1 day) + CAD drawing (1 day)," to within 60 minutes. Parameter verification and reinforcement prediction take only a few seconds, and CAD drawing takes only 1-2 minutes, thus reducing the design cycle and improving design efficiency.

[0042] 2) The anomaly detection model of this invention filters out non-compliant parameters in advance to avoid structural safety risks caused by human oversight; the reinforcement prediction model is based on pre-trained data constrained by specifications, and the output results naturally conform to the railway culvert design specifications, reducing reinforcement deviations caused by human experience and ensuring design quality.

[0043] 3) Designers of this invention do not need to master machine learning algorithms or advanced CAD drawing skills. They only need to input basic parameters to complete the entire design process, which is suitable for designers with different experience levels and lowers the technical threshold. Attached Figure Description

[0044] Figure 1 This is a system architecture diagram of the railway box culvert design and drawing system of the present invention;

[0045] Figure 2 This is a flowchart of the system program development for this invention;

[0046] Figure 3 This is a flowchart of the pre-trained model training process. Detailed Implementation

[0047] The present invention will now be described in detail with reference to specific embodiments.

[0048] This invention provides a railway culvert design and drawing system and method based on a hybrid intelligent algorithm, which realizes a closed-loop process of intelligent verification of culvert design parameters, automatic prediction of reinforcement, and automated drawing. It can not only improve design efficiency and quality, but also reduce the difficulty of railway culvert design and drawing.

[0049] like Figure 1 As shown, the present invention provides a railway culvert design and drawing system based on a hybrid intelligent algorithm, including a data input layer, a pre-trained model layer, an engineering quantity calculation layer, and a CAD drawing layer: the data input layer supports designers to input railway culvert design parameters and performs preliminary format verification on the input parameters.

[0050] The pre-trained model layer is the core of this system's intelligence, comprising two sub-models: the M1 unsupervised anomaly detection model and the M2 reinforcement prediction model, both built upon processed standardized data. The technical implementation logic of this layer and its relationships with other modules can be discussed in conjunction with the attached... Figure 1 Railway box culvert design and drawing system architecture diagram and appendix Figure 3 Understanding the pre-trained model training flowchart: (See attached diagram) Figure 1 As shown, the pre-trained model layer is located between the "data input layer" and the "engineering quantity calculation layer." It receives standardized parameters from the data input layer and passes the output results to subsequent layers. The M1 unsupervised anomaly detection model is pre-trained based on the Isolation Forest algorithm. Its input consists of the structural dimensions and stress parameters from the data input layer. It outputs anomaly scores and anomaly identifiers, alerting designers to make corrections. The M2 reinforcement prediction model is pre-trained based on the XGBoost multi-output regression algorithm. Its input consists of corrected compliance parameters. It outputs estimated reinforcement parameters for each part of the culvert, presenting the results as reinforcement area and corresponding reinforcement details.

[0051] The engineering quantity calculation layer constructs an engineering quantity calculation module to calculate the engineering quantities related to the culvert and output the relevant quantities to the CAD drawing layer.

[0052] This invention's CAD drawing layer is based on Python for secondary development of AutoCAD. It calls the pythoncom, win32com, and win32api libraries to achieve communication and low-level interaction between Python and AutoCAD. Based on the accurate engineering quantity data output from the pre-trained model layer and the engineering quantity calculation layer, it automatically draws standard railway culvert drawings through various interactive drawing code blocks independently developed in this invention. The drawings include culvert plan and elevation, wing wall plan and elevation, and reinforcement layout. This CAD drawing layer supports one-click saving and archiving of completed drawings in DWG format. It also supports designers to re-execute the drawing process after modifying culvert design parameters, thereby quickly updating and redrawing the entire set of drawings without manual modification.

[0053] Among them, the interactive drawing function is a newly developed exclusive code module (not an existing general function / interface) specifically designed for the drawing standards and structural design characteristics of railway culvert engineering. It refers to the core execution code for drawing various culvert drawings.

[0054] The interactive drawing functions are divided into culvert structure drawing code modules, wing wall structure drawing code modules, detailed structure drawing code modules, and reinforcement layout drawing code modules according to their drawing functions. All types of code modules are written based on the AutoCAD underlying interface encapsulation of pythoncom and win32com, respectively completing the drawing of culvert plan and elevation, wing wall plan and elevation, detailed structure, and reinforcement layout of each part.

[0055] In the interactive drawing function:

[0056] 1) The culvert structure drawing code module is a proprietary Python code segment developed specifically for drawing railway culvert box girder drawings. Its core function is to convert real-world culvert box girder structures into standardized views such as plan and elevation in CAD drawings. This code module receives box girder dimension parameters after verification from the preceding module. Through built-in coordinate calculation logic, it accurately calculates the CAD drawing coordinates of each contour point of the culvert box girder according to the drawing order specified in railway culvert drawings. Based on the AutoCAD underlying interface of pythoncom and win32com, it generates the corresponding CAD polyline contour of the box girder. Simultaneously, the code module embeds standardized annotation rules to automatically complete the dimension annotations (such as length, width, and thickness values) of each part of the box girder, ultimately completing the automated drawing of the culvert box girder's plan and elevation views.

[0057] 2) The wing wall drawing code module is a proprietary Python code segment developed specifically for drawing railway culvert wing wall drawings. Its core function is to convert real-world culvert wing wall structures into standardized views such as plan and elevation in CAD drawings. This code module receives wing wall dimension parameters after verification from the preceding module. Through built-in coordinate calculation logic, it accurately calculates the CAD coordinates of each contour point of the culvert box body according to the drawing order specified in railway culvert drawings. Based on the AutoCAD underlying interfaces of pythoncom and win32com, it generates the corresponding CAD polylines and line contours of the wing wall. Simultaneously, it automatically completes the dimension annotations for each part of the wing wall (wing wall length, width, and deflection angle), ultimately completing the automated drawing of the culvert wing wall plan and elevation views.

[0058] 3) The reinforcement drawing code module is a proprietary Python code segment developed specifically for the drawing requirements of railway culvert wing wall drawings. Its core function is to achieve accurate and automated drawing of culvert reinforcement layout drawings. Based on the reinforcement density, layout form (e.g., single-layer arrangement, double-layer staggered), diameter, length, bending angle, and spacing of the reinforcement, as output by the preceding reinforcement module and manually corrected, this code module, combined with the drawing scale and AutoCAD coordinate system, accurately calculates the CAD drawing coordinates of key points such as the start point, end point, and bending point of each reinforcement. Based on the AutoCAD underlying interface of pythoncom and win32com, it generates the CAD line outline corresponding to each reinforcement according to the railway engineering drawing scale and completes the visual drawing of the overall reinforcement layout according to the reinforcement layout form. Simultaneously, the code module embeds standardized reinforcement annotation rules, automatically annotating the specification number (e.g., "Φ20"), spacing, quantity, length, and reinforcement number of each reinforcement, ensuring that the reinforcement layout drawing conforms to the railway culvert reinforcement drawing specifications and accurately reflects the actual reinforcement layout status.

[0059] like Figure 2As shown, the present invention also provides a railway culvert design and drawing method based on a hybrid intelligent algorithm, which specifically includes the following steps:

[0060] Step 1: Construct an unsupervised anomaly detection model for M1 and a reinforcement prediction model for M2 to provide intelligent support for subsequent culvert design; including the following steps:

[0061] S1.1: Acquisition and processing of training data, generating training and test sets;

[0062] Design data for numerous culverts along multiple major railway lines were collected and analyzed. A dataset was constructed through a three-step process: data acquisition, standardization, and engineering cleaning, providing reliable input for model training. Figure 3 As shown, the details are as follows:

[0063] 1.1.1: Historical Data Collection: A hybrid collection mode of "real project data + standardized generated data" is adopted to ensure data coverage and compliance, and to generate more training data for the model; real project data refers to the data extracted from the approved design documents, while standardized generated data refers to virtual samples randomly sampled according to the specifications, which supplements and enriches the training and testing datasets.

[0064] 1.1.2: Data Cleaning and Processing: Perform an engineered cleaning process on the collected data to remove invalid information and standardize the format, including handling missing values, removing outliers, and standardizing units;

[0065] 1.1.3 Feature Engineering Processing: To enhance the model's ability to learn engineering principles, derived features are constructed from standardized data based on culvert design principles. The derived features constructed in this application represent all required features, specifically: aperture-to-clear height ratio, fill height-to-top slab thickness ratio, and web thickness-to-bottom slab thickness ratio. These three features accurately reflect the matching relationship of culvert structural dimensions; a reinforcement sufficiency label, which can determine reinforcement compliance. Each derived feature strengthens the engineering attributes of the data features from the dimensions of structural proportions and reinforcement compliance, providing high-value feature input for subsequent model training.

[0066] S1.2: Model Training and Saving: The Isolation Forest algorithm is used to train the M1 unsupervised anomaly detection model; the XGBoost multi-output regression algorithm is used to train the M2 reinforcement prediction model. A labeled dataset is constructed with reinforcement parameters as output labels, and the two trained models are saved.

[0067] As attached Figure 3 As shown in the "Model Training Phase," the two sub-models share the "preprocessed dataset," but employ differentiated training paths. The specific implementation process is as follows:

[0068] 1.2.1: As attached Figure 3As shown in the left branch of the "Model Training Phase," the M1 model uses the Isolation Forest algorithm, which is an existing algorithm framework. This invention does not improve the algorithm itself, but rather adapts the existing algorithm framework to the characteristics of railway culvert engineering design parameters and customizes the core parameters to train a dedicated M1 unsupervised anomaly detection model suitable for anomaly detection of railway culvert design parameters. The specific core parameters are as follows: ① Number of decision trees (n_estimators) = 200; ② Number of samples per tree (max_samples) = 256; ③ Decision tree depth limit (max_depth) = 10 layers to prevent overfitting caused by excessive learning of local features by a single tree; ④ Anomaly ratio preset (contamination) = 0.1, matching the actual occurrence probability of abnormal parameters in railway engineering (approximately 10%).

[0069] 1.2.2: As attached Figure 3 As shown in the right-hand branch of the "Model Training Phase," the M2 model employs the XGBoost multi-output regression algorithm. This algorithm is an existing framework, and this invention does not improve the algorithm itself. Instead, it adapts the existing framework to the engineering characteristics of railway culvert design, customizing the optimal hyperparameters and combining it with a standardized culvert dataset constructed in this invention for dedicated training. This results in a dedicated M2 model suitable for predicting reinforcement parameters in railway culverts. This M2 model requires the construction of a one-to-one labeled dataset of input features and output labels based on preprocessed standardized data. Reinforcement prediction is achieved by learning the mapping relationship between the two. Specifically, the mapping relationship is a deterministic engineering mathematical mapping between the input features of the railway culvert and the output labels of the reinforcement parameters for various parts of the culvert. Its core formula is as follows:

[0070] The hyperparameters are optimized using Bayesian optimization, with the core objective being to minimize the iterative objective function of XGBoost. XGBoost fits the residuals from the previous round to each decision tree, and the objective function for the t-th round is:

[0071]

[0072] Symbol explanation: n: total sample size; : The actual reinforcement value of the i-th sample (e.g., the area of ​​the main reinforcement on the inner side of the top slab); : The cumulative reinforcement prediction value of the i-th sample in the first t-1 rounds of the model; : The prediction contribution of the t-th decision tree to the i-th sample (the residual of the previous model fitting); The squared error loss function is expressed as follows: It is suitable for continuous value regression scenarios of reinforcement parameters; : The regularization term of the t-th decision tree, used to control model complexity and avoid overfitting.

[0073] To improve training efficiency, the loss function is adjusted. exist Performing a second-order Taylor expansion (ignoring the constant term) transforms the objective function into an easily optimized form:

[0074]

[0075] Symbol explanation: Loss function in The first derivative (gradient) at a given point reflects the direction of the current prediction error; Loss function in The second derivative at a given point (Hessian matrix) reflects the rate of change of the gradient, making the loss reduction more accurate and adapting to the limited sample size of engineering data.

[0076] Regularization term To control the complexity of the decision tree and prevent the model from memorizing non-standard culvert data for a specific route, the formula is:

[0077]

[0078] Symbol explanation: T: The total number of leaf nodes in the t-th decision tree; : The weight of the j-th leaf node (the reinforcement prediction contribution value corresponding to this node); : The optimal hyperparameter is the minimum loss threshold for node splitting. If the loss is below this threshold, node splitting stops to reduce invalid computation. : Optimal hyperparameters, L2 regularization coefficients, penalize excessively large node weights, and prevent model overfitting.

[0079] Regarding the expanded objective function By taking the derivative and setting it to zero, we can obtain the optimal weight of the j-th leaf node (which directly determines the reinforcement prediction output):

[0080]

[0081] Symbol explanation: : The set of samples contained in the j-th leaf node (i.e., culvert samples that the model classifies as having the same working conditions); numerator :gather The sum of gradients for all samples reflects the overall prediction bias for this type of operating condition; the denominator is the sum of gradients for all samples within the range. :gather The sum of the Hessian matrix and the regularization coefficients of all samples within the model balances prediction accuracy and model generalization ability; its role is to ensure that the prediction contribution of each decision tree conforms to the true distribution law of the reinforcement parameters of railway culverts.

[0082] 1.2.3: Model Quantitative Evaluation: The dataset was divided into training and test sets in a 7:3 ratio, and the model was trained until convergence. Since the M2 model needs to output 9 types of reinforcement parameters simultaneously, a multi-output regression loss function was used to measure the overall error. RMSE (Root Mean Square Error) and coefficient of determination were used to evaluate accuracy, with values ​​[0,1]. The closer the value is to 1, the stronger the model's explanatory power for reinforcement patterns. On the test set, the final average RMSE was 0.1415; the prediction error of reinforcement parameters for each part was controlled within 10%, meeting the accuracy requirements of engineering design.

[0083] 1.2.4: Model Saving: Save the two trained sub-models in binary format, and associate them with meta-information such as model version number, training data batch, hyperparameter combination and evaluation metrics, so as to facilitate system calls and subsequent incremental training.

[0084] Step 2: Parameter input and preliminary verification, realizing design parameter acquisition and initial screening for format compliance; specifically including:

[0085] S2.1: Designers can input structural parameters and stress parameters one by one, or input them in batches through a file.

[0086] S2.2: The preliminary format verification system automatically initiates a format verification mechanism to perform preliminary verification on the input parameters, ensuring they are numerical and excluding invalid inputs such as text and symbols. This invention utilizes a desktop interactive interface developed based on a Python visualization library to implement parameter input and preliminary verification. If the input is invalid, a pop-up window alerts the user and redirects them to the corresponding input box. After successful verification, the "Parameter Submission" button and the reinforcement parameter preview window are activated.

[0087] Step 3: Anomaly detection and correction using the M1 unsupervised anomaly detection model. This involves using a pre-trained M1 model to identify the rationality of parameter combinations and prevent anomalous parameters from entering the design phase. Specifically, this includes:

[0088] S3.1: As Figure 2 As shown, the culvert design parameters obtained in the parameter input stage are standardized and then input into the M1 unsupervised anomaly detection model. The model uses a weighted sampling algorithm optimized for the distribution characteristics of culvert parameters. By strengthening the sampling weight of key design parameters (such as aperture and top slab thickness), the anomaly score of parameter combinations is calculated (the value ranges from 0 to 100%, and the higher the score, the higher the risk of the parameter combination deviating from the conventional design range of the project). If the anomaly score is ≤10%, it indicates that the M1 model has passed the detection of the input culvert design parameters, and some parameters such as dimensions become the fixed parameters and enter the subsequent modules; otherwise, the input culvert design parameters enter the anomaly detection module.

[0089] S3.2: Based on the processing results of step S3.1, if the anomaly score > the preset engineering safety threshold (10%), calculate the SHAP value of each input parameter and sort them in descending order of anomaly degree. Prompt the user to check the most abnormal parameters and make subsequent modifications; output clear correction prompts to the user; an example prompt is: "Abnormal parameter: aperture; Cause of anomaly: the ratio of fill height to aperture deviates from the normal range, and the top plate thickness significantly exceeds the standard design range"; for example, if the input parameters are "aperture 20m, top plate thickness 0.15m, fill height 12m". The system implementation process is as follows: In step 3, the M1 model calculation anomaly score is 45%, and the top three anomalies in the SHAP value ranking are aperture, backfill height, and top plate thickness; a pop-up window reminds that "top plate thickness of 0.15m violates Article 6.2.1 of TB10003-2016 (minimum 0.2m); backfill height of 12m violates Article 5.3.2 (maximum 10m)"; the designer corrects the parameters to "top plate thickness 0.25m, backfill height 9m" and resubmits; repeat steps 2-7 to finally complete the design drawing and ensure design compliance.

[0090] S3.3: Parameter Correction and Secondary Detection: After correcting abnormal parameters based on the prompts, the designer repeats S3.1-S3.2 until the abnormal scores of all culvert parameters are ≤ the preset engineering safety threshold. If the designer needs to retain abnormal parameters due to special working conditions, they can also submit a "Parameter Special Case Application" and note the specific engineering scenario reason. The system will record the parameter combination of the special case and archive it for reference in the design of similar working conditions in the future.

[0091] Step 4: Generate culvert reinforcement schemes using the M2 reinforcement prediction model; learn the mapping relationship between "compliance parameters - reinforcement parameters" through the pre-trained M2 model to achieve intelligent generation of reinforcement schemes; specifically including:

[0092] S4.1: Import the culvert design parameters output after anomaly detection and correction in step 3 into the pre-trained M2 reinforcement prediction model; the M2 reinforcement prediction model calls the pre-stored railway culvert reinforcement training weights, which are trained based on the "compliance parameters - reinforcement parameters" labeled dataset and are adapted to the requirement of synchronous prediction of multi-dimensional reinforcement parameters.

[0093] S4.2: The M2 reinforcement prediction model outputs 9 types of reinforcement estimation parameters, specifically including the area of ​​main reinforcement and auxiliary reinforcement in different parts of the culvert. These parameters, together with the finalized parameters such as dimensions obtained in step 3, constitute the finalized design parameters of the current railway culvert. Based on the preset engineering reinforcement specifications and parameter matching rules, the system automatically generates recommended reinforcement combinations according to the output reinforcement area, providing designers with direct scheme references.

[0094] S4.3: Reinforcement Parameter Window Activation After the reinforcement prediction results are generated, the system automatically activates the "Reinforcement Parameter Input Window" preset in step 1, and fills the predicted parameters into the corresponding fields for designers to view and adjust.

[0095] Step 5: Based on the modified and confirmed design parameters, realize the automated calculation of the project quantity to provide data support for project pricing;

[0096] This invention employs Python code to develop modules for calculating quantities for different types of culverts and different parts of culverts (culvert body, wing walls, and ancillary works). The system automatically matches the calculation module based on design parameters and uses built-in functions to calculate quantities. These built-in calculation functions are modified and refined using pre-trained model layers, incorporating railway culvert design specifications and general quantity calculation formulas. Quantity data is temporarily stored in the system database, linked to the current design task ID, and supports categorized filtering and output. The output is an Excel-formatted quantity table categorized by railway culvert design ID, along with the quantities temporarily stored in the program's memory, serving as input for the CAD drawing layer.

[0097] Step 6: Based on the structural parameters, load parameters, reinforcement parameters, and temporarily stored quantities of the railway culvert obtained in Steps 2, 3, 4, and 5, generate standard drawings of the railway culvert and output the quantities of the culvert, specifically including:

[0098] S6.1: CAD Environment Initialization and Parameter Parsing: The system automatically calls the pythoncom, win32com, and win32api libraries to connect to AutoCAD (supports versions 2014 and above), initializes the drawing environment (sets layers and scale); at the same time, it parses the final parameters, reinforcement data, and engineering quantity details into CAD-recognizable drawing commands (such as coordinates, line types, and text height); it converts the finalized structural parameters and reinforcement parameters into CAD drawing coordinates and dimensions (e.g., converting "hole diameter 4m" into a 400mm drawing size at a 1:100 scale), and converts "rebar area" into rebar diameter and spacing using the formula (rebar area = π × diameter squared / 4 × 1000 / spacing);

[0099] S6.2: Batch drawing of standard drawings;

[0100] According to railway engineering drawing standards, the system automatically draws drawings in the order of "overall drawing - cross-sectional drawing - detailed drawing - table": ① Plan and elevation drawings: Draw the outer contour based on the aperture, culvert length, and wing wall dimensions, and mark the mileage and elevation; ② Cross-sectional drawings: Draw the structural dimensions of the top slab / web / bottom slab, and mark the thickness and material grade; ③ Reinforcement layout drawings: Draw the main reinforcement according to the reinforcement parameters, and use different color layers to distinguish the reinforcement in different parts (distinguishing colors: red for top slab, blue for web, green for bottom slab), stirrups, and mark the diameter and spacing; ④ Detailed structural drawings: Draw the structural details of key parts such as inner ribs and settlement joints; ⑤ Table drawing: Call the engineering quantity table drawing function to automatically generate the reinforcement table and engineering quantity table, and insert them into the specified position on the drawing.

[0101] S6.3: Drawing Output: Designers can save CAD drawings as dwg / dxf format with one click. It supports saving a single set of drawings individually or exporting multiple sets of drawings with corresponding parameters in batches. The default save path is associated with the task ID folder.

[0102] S6.4: Drawing Optimization and Saving: After drawing is completed, the layer is automatically organized (the structural layer is locked and the annotation layer is unlocked). Designers can adjust the drawing scale (1:100 / 1:200). Finally, the drawing is saved in dwg format according to the naming rule of "line name-culvert number-design date". Specific Implementation

[0103] The following example, using the design of a reinforced concrete box culvert on a passenger dedicated railway line, fully reproduces the implementation process of the system and method of this invention, facilitating understanding and reproduction by those skilled in the art:

[0104] 1) Basic Information of the Implementation Example

[0105] Design requirements: A reinforced concrete box culvert for a passenger dedicated railway line has a design aperture of 4m, a net height of 2.5m, a backfill height of 6m, and is designed to withstand a concentrated load of 300kN. Parameter verification, reinforcement design, quantity calculation, and drawing preparation are required.

[0106] System environment: Intel Core i5-10400 processor, 16GB memory, AutoCAD 2018, and this system.

[0107] 2) Implementation process

[0108] Step 1 (Parameter Input and Preliminary Verification): The designer manually enters the parameters (aperture diameter 4m, net height 2.5m, top plate thickness 0.3m, web plate thickness 0.28m, bottom plate thickness 0.32m, fill height 6m, concentrated load 300kN) through the data input module, and the system verification shows no abnormalities;

[0109] Step 2 (M1 Model Anomaly Detection): The M1 Isolation Forest model is called to calculate an anomaly score of 7.2% (≤10% threshold). The parameters are deemed compliant, and no anomaly alert is issued.

[0110] Step 3 (M2 model reinforcement prediction): Call the M2 model to output reinforcement results, including 9 types of parameters such as the main reinforcement on the inner side of the top slab (3289.56 mm²) and the main reinforcement on the inner side of the web (3856.23 mm²);

[0111] Step 4 (Parameter Finalization): Based on their engineering experience, the designers fine-tune the inner main reinforcement of the top slab to 3300mm². After the system verifies compliance, they click "Parameter Finalization" to generate the final parameters.

[0112] Step 5 (Quantity Calculation): The quantity calculation layer matches the small box culvert calculation module to calculate the concrete volume of the culvert body, the concrete volume of the wing walls, the total weight of the steel reinforcement, and the quantity of ancillary works, and generates an Excel quantity table.

[0113] Step 6 (Automatic CAD Drawing): The system connects to AutoCAD 2018 and draws a CAD drawing at a 1:100 scale, named "Box Culvert Structure Drawing," which includes plan and elevation views, three types of cross-sections, and generates a quantity table on the right side of the drawing. The final export file is "XX Passenger Dedicated Line - Box Culvert 001 - Construction Drawing.dwg". Simultaneously, it generates matching reinforcing bars according to the latest reinforcement parameters. Figure 1 Zhang, the right side of the drawing generated a reinforcement table, which was finally exported as "XX Passenger Dedicated Line-Box Culvert 001-Reinforcement Drawing.dwg".

[0114] Implementation results: The entire process took 30 minutes, which is significantly more efficient than the traditional design process (about 1 day). There were no compliance errors, the reinforcement prediction error was 8.5%, and the engineering design requirements were met.

[0115] The content of this invention is not limited to the embodiments listed. Any equivalent modifications made by those skilled in the art to the technical solutions of this invention by reading this specification are covered by the claims of this invention.

Claims

1. A railway culvert design and drawing method based on a hybrid intelligent algorithm, characterized in that: Specifically, the following steps are included: Step 1: Construct an unsupervised anomaly detection model for M1 and a reinforcement prediction model for M2; Step 2: Parameter input and preliminary verification; Step 3: Anomaly detection and correction using the M1 unsupervised anomaly detection model, identifying the rationality of parameter combinations through a pre-trained M1 model; Step 4: Generate the culvert reinforcement scheme using the M2 reinforcement prediction model; Step 5: Based on the modified and confirmed design parameters, automate the calculation of project quantities; Step 6: Based on the structural parameters, load parameters, reinforcement parameters, and temporarily stored quantities of the railway culvert obtained in Steps 2, 3, 4, and 5, generate standard drawings of the railway culvert and output the quantities of the culvert.

2. The railway culvert design and drawing method based on a hybrid intelligent algorithm according to claim 1, characterized in that: Step 1 includes the following steps: S1.1: Acquisition and processing of training data, generating training and test sets; S1.2: Model Training and Saving: The Isolation Forest algorithm is used to train the M1 unsupervised anomaly detection model; the XGBoost multi-output regression algorithm is used to train the M2 reinforcement prediction model. A labeled dataset is constructed with reinforcement parameters as output labels, and the two trained models are saved.

3. The railway culvert design and drawing method based on a hybrid intelligent algorithm according to claim 2, characterized in that: Step 2 specifically includes: S2.1: Designers can input structural parameters and stress parameters one by one, or input them in batches through a file. S2.2: Perform preliminary validation on the input parameters.

4. The railway culvert design and drawing method based on a hybrid intelligent algorithm according to claim 3, characterized in that: Step 3 specifically includes: S3.1: After standardizing the culvert design parameters obtained in the parameter input stage, input them into the M1 unsupervised anomaly detection model; the model uses a weighted sampling algorithm to calculate the anomaly score of the parameter combination; S3.2: Based on the processing result of step S3.1, if the anomaly score is greater than the preset engineering safety threshold, calculate the SHAP value of each input parameter and sort them in descending order of anomaly degree, and output a clear correction prompt to the user so that subsequent modifications can be made; S3.3: After correcting the abnormal parameters based on the prompts, repeat S3.1-S3.2 until the abnormal scores of all culvert parameters are less than or equal to the preset engineering safety threshold.

5. The railway culvert design and drawing method based on a hybrid intelligent algorithm according to claim 4, characterized in that: Step 4 specifically includes: S4.1: Import the culvert design parameters output after anomaly detection and correction in step 3 into the pre-trained M2 reinforcement prediction model; the M2 reinforcement prediction model calls the pre-stored railway culvert reinforcement training weights, which are trained based on the labeled dataset. S4.2: The M2 reinforcement prediction model outputs 9 types of reinforcement prediction parameters.

6. The railway culvert design and drawing method based on a hybrid intelligent algorithm according to claim 5, characterized in that: Step 6 specifically includes: S6.1: CAD Environment Initialization and Parameter Parsing; S6.2: Batch drawing of standard drawings; S6.3: Drawing output; S6.4: Drawing optimization and saving.

7. A railway culvert design and drawing system based on a hybrid intelligent algorithm, characterized in that: It includes a data input layer, a pre-trained model layer, an engineering quantity calculation layer, and a CAD drawing layer. The data input layer allows designers to input railway culvert design parameters and performs preliminary format verification on the input parameters; The pre-trained model layer calls the pre-trained hybrid intelligent model, which includes the M1 unsupervised anomaly detection model and the M2 reinforcement prediction model. The engineering quantity calculation layer constructs an engineering quantity calculation module to calculate the engineering quantities related to the culvert and outputs the relevant quantities to the CAD drawing layer.

Images