Rail transit simulation configuration model generation method based on AIGC and XR

Abstract

The invention relates to the technical field of rail transit simulation modeling and production and education fusion, in particular to a rail transit simulation configuration model generation method based on AIGC and XR, which comprises the following steps of: preliminarily checking generated model component parameters; scene assembly is carried out according to the actual line layout; dynamically correcting a train operation logic and signal control mechanism; model component and scene adjustment is supported based on modular design. According to the AIGC and XR-based rail transit simulation configuration model generation method, resource blocks with relatively high topic correlation are screened through a Bayesian probability formula according to a topic segmentation and feature enhancement technology of an AIGC platform, feature attribute overlapping interference is eliminated, CNN is introduced to verify model component parameters, and constraint fitting is performed in combination with a loss function; the XR simulation resource extraction and model parameter precision is improved; the virtual scene and the real physical space are aligned by combining the SLAM space positioning and data intercommunication technology, so that XR virtual-real deep fusion is realized, and the immersive interaction experience is improved.

Description

Technical Field

[0001] This invention relates to the field of rail transit simulation modeling and industry-education integration technology, specifically a method for generating rail transit simulation configuration models based on AIGC and XR. Background Technology

[0002] The method for generating simulation configuration models for rail transit encompasses the comprehensive application of technologies such as AI-generated content (AIGC), extended reality (XR), deep learning, and natural language processing (NLP) in rail transit teaching and training, operation simulation, and emergency drill scenarios. Its core lies in the integration of AI-driven simulation resource processing and XR immersive interaction, which can form a full-chain technical solution of resource extraction, model generation, scene interaction, and demand adaptation. It serves the teaching and training goals of promoting reality through virtualization in the integration of industry and education in rail transit, and can also be extended to other fields such as employee training for rail transit operating companies and simulation testing before the opening of new lines.

[0003] However, the current application of simulation modeling and teaching / training resources in the field of rail transit has the following shortcomings, specifically:

[0004] 1) The accuracy of XR simulation resource extraction for rail transit is insufficient, and the problem of feature overlap is quite prominent;

[0005] 2) Traditional modeling methods are inefficient, lack automation, and rely heavily on manual intervention;

[0006] 3) The lack of virtual-real interaction capabilities results in a weak connection between the immersive experience and the real scene;

[0007] 4) Poor scenario adaptability, making it difficult to quickly respond to diverse needs;

[0008] Therefore, based on the aforementioned technical problems, a method for generating rail transit simulation configuration models based on AIGC and XR is proposed. Summary of the Invention

[0009] (a) Technical problems to be solved

[0010] To address the shortcomings of existing technologies, this invention provides a method for generating rail transit simulation configuration models based on AIGC and XR. This method has the advantages of improving the accuracy of XR simulation resource extraction and model parameters, and achieving deep integration of XR virtual and real data. It solves the problems of low resource extraction accuracy, weak virtual-real interaction, and insufficient scene adaptability in existing simulation modeling and teaching and training resource applications mentioned above.

[0011] (II) Technical Solution

[0012] To achieve the above objectives, the present invention provides the following technical solution: a method for generating a rail transit simulation configuration model based on AIGC and XR, comprising the following steps:

[0013] S1. Covers static and Dynamic core data is collected from multiple sources, preprocessed based on the AIGC platform, and a standardized dataset is generated.

[0014] S2. The deep learning generation framework based on the AIGC platform generates model components in modules according to the type of rail transit elements, and introduces a convolutional neural network to perform preliminary verification of the parameters of the generated model components;

[0015] S3. Based on the XR engine, the 3D model components generated by AIGC are assembled into a scene according to the actual line layout, and spatial positioning technology is used to align the virtual simulation scene with the real physical space.

[0016] S4. Construct a bidirectional long short-term memory network to dynamically correct the train operation logic and signal control mechanism, and optimize the XR interaction logic based on user operation data using reinforcement learning (RL).

[0017] S5. The AIGC platform uses NLP to parse user requirement text, automatically extracts key requirement parameters, and supports model component and scenario adjustment based on modular design, enabling configuration model output and deployment.

[0018] Preferably, in step S1, XR simulation resource data is collected from the rail transit system. The XR simulation resource data ensures that it covers the static structure and dynamic operation information required for simulation. It is divided into static resource data and dynamic operation data. The static resource data includes spatial structure data of the track layout, geographic coordinate data of station locations, basic parameters of equipment 3D models, and layout data of platform facilities.

[0019] The dynamic operation data includes train timetables, route planning data, train speed curve data, and text and time-series data used to simulate the scheduling and control process of the rail transit system. The text and time-series data includes train scheduling rules, emergency response plans, and signal control logic.

[0020] The collected XR simulation resource data is preprocessed using the AIGC platform to achieve data structuring and initialization. Based on the DOM tree, the resource data is pre-divided into blocks by topic-related tags, and the independence of the feature attributes of each block is ensured to eliminate attribute overlap interference.

[0021] Preferably, the association probability between each block and the topic block is calculated using the Bayesian probability formula, and blocks with high relevance to the topic are selected. The association probability between the blocks and the topic block is then quantified. The association probability calculation formula is as follows:

[0022]

[0023] in The association probability represents the probability that a known block B contains a feature attribute. In this case, the probability that block B belongs to topic block S is: This is used to quantify the correlation between chunks and topics. The closer the probability value is to 1, the stronger the topic correlation of the chunk.

[0024] B represents a single block result after pre-blocking the XR simulation resource data, and each B corresponds to a segment of structured resource data;

[0025] This represents a feature attribute contained in block B. All feature attributes of block B constitute a feature attribute vector. , S represents the theme block of the rail transit XR simulation resource, that is, the reference block that can reflect the core content of the resource;

[0026] The prior probability represents the probability that any block B belongs to the topic block S without considering feature attributes. It can be determined by statistically analyzing the proportion of topic blocks in historical block data.

[0027] The conditional probability indicates that, given that block B belongs to topic block S, block B contains the feature attribute. The probability is used to measure the strength of the association between a topic block and a feature attribute;

[0028] Marginal probability represents the probability that any block in the partitioning of all rail transit XR simulation resources contains the characteristic attribute. The probability can be determined by statistically analyzing all blocks containing [a certain number of elements]. The proportion of the number of blocks to the total number of blocks is calculated;

[0029] Set association probability The threshold is used to retain blocks with an association probability higher than the threshold as valid data blocks, while blocks with an association probability lower than the threshold are judged as non-topic blocks and removed. Finally, the preprocessed set of valid resource data is output.

[0030] Preferably, in step S2, based on the AIGC platform's ability to process XR simulation resource tags, the extracted resource features are mapped to model component attributes using tags, specifically including:

[0031] 1) The effective block features extracted by topic segmentation are categorized by label according to the model component type. The effective block features include train size data, track topology parameters and signal control logic. The model component types include train, track, signal system and platform facilities.

[0032] 2) Standardize the categorized feature labels and unify the label format;

[0033] By utilizing the inverse pooling operation in the AIGC platform's deep learning architecture, key feature attributes of the mapped model components are amplified to enhance feature discriminative power. The formula for feature attribute amplification is as follows:

[0034]

[0035] in This represents the amplified result of the feature attributes of the i-th model component;

[0036] This represents the inverse pooling operation, used to restore and amplify the dimensionality and numerical precision of feature attributes;

[0037] Represents the original feature attributes of the i-th model component;

[0038] This represents the output result of the convolution operation on the feature attributes of the (i-1)th model component;

[0039] The amplified feature attributes are concatenated with the features of adjacent model components through convolution operations to form a complete feature set of the model components, ensuring the correlation between features between components. The feature attribute concatenation formula is as follows:

[0040]

[0041] in This represents the final feature output of the i-th model component after two convolution operations;

[0042] This refers to the convolution operation in the deep learning architecture of the AIGC platform, used to integrate different feature attributes;

[0043] This represents the amplified feature of the i-th component. Original features of the (i+1)th component Perform dimensional stitching;

[0044] Using the standard parameters of actual rail transit equipment as a benchmark, the features of the AIGC-enhanced model components are fitted to ensure consistency between the component features and the actual rail transit scenario. The steps include:

[0045] 1) Fit the fused features to the rail transit industry-education integration XR simulation resources, construct a loss function to constrain the output results, and express the loss function as follows:

[0046]

[0047] in This represents the loss value for fitting the features of the model components. The smaller the loss value, the smaller the deviation between the fitting result and the actual standard parameters.

[0048] This is a constant used to adjust the weights of the two fitting scenarios;

[0049] This indicates the status of the feature fitting result of the model component. When x=1, it means that the fitting result is a non-empty set, that is, the effective features have been successfully extracted and the fitting has been completed. When x=0, it means that the fitting result is an empty set, that is, the feature extraction or fitting has failed.

[0050] This represents the feature output of the model components after the convolution operation, which is used to participate in the calculation of the loss value, ensuring that the loss constraint is strongly correlated with the feature attributes.

[0051] 2) Set the number of iterations and continuously adjust the feature parameters of the model components to optimize the loss value. Approaching 0, the final output is a set of model component features that meet the criteria.

[0052] Preferably, in step S3, based on the hierarchical structure of the DOM tree and the topic segmentation results of the XR simulation resources, the valid XR simulation resources extracted through the AIGC platform are imported into the XR basic engine according to the classification of static resources and dynamic resources, as follows:

[0053] 1) Complete node mapping of the equipment 3D model parameters in the XR engine. The equipment 3D model parameters include tracks, platforms, and signals to ensure that the initial position of the model in the virtual space conforms to the actual line layout logic and realize the import of static resources.

[0054] 2) Convert the time-series data into script parameters that the XR engine can recognize. The time-series data includes train timetables, speed curves, and signal control logic to achieve dynamic resource import.

[0055] Based on the DOM tree node hierarchy, the resource structure is decomposed into a three-level hierarchy: root node - functional node - attribute node, thus completing the structured parsing of the resources. Specifically:

[0056] 1) Root node: corresponds to the overall rail transit simulation scenario;

[0057] 2) Functional nodes: Corresponding to the functional partitions within the scene, based on the Bayesian probability results of topic segmentation, blocks with topic relevance ≥ 0.3 are selected as the data source for functional nodes;

[0058] 3) Attribute nodes: correspond to the specific resource attributes under the function nodes;

[0059] The parsed root node-functional node-attribute node are mapped to the scene layer-object layer-parameter layer of the XR engine to ensure that the resource structure matches the XR environment hierarchy. Specifically, this includes:

[0060] 1) Scene layer: Load the overall scene coordinate system corresponding to the root node, and define the XR virtual coordinate system based on the actual geographic coordinates of the track line;

[0061] 2) Object layer: Instantiate the block resources corresponding to the functional nodes into virtual objects in the XR engine;

[0062] 3) Parameter layer: Assigns the resource parameters corresponding to the attribute nodes to the component parameters of the virtual object;

[0063] Based on the enhanced feature attributes of the XR simulation resources, the component assembly and rendering configuration of the virtual scene are completed. The specific steps are as follows:

[0064] 1) Static component assembly: Based on the track layout data, the virtual track objects are spliced ​​in the XR scene layer in the order of main branches, and the platform and signal are placed as static objects at the specified coordinate positions.

[0065] 2) Dynamic component association: Associating train operation data with train virtual objects;

[0066] 3) Based on the texture features of XR simulation resources, an adaptive texture resolution algorithm is used to adjust the rendering accuracy;

[0067] Based on the AIGC platform's deep learning framework and feature fitting logic, this system achieves the fusion and interaction of virtual and real scenes using spatial alignment and data interoperability algorithms. It also configures basic interactive functions according to the interaction requirements of XR simulation resources. Specific operations include:

[0068] 1) Multi-view interaction configuration: Based on the multi-dimensional features of resource segmentation, global view, device view and first-person view are set in the XR environment, which is triggered by the user through the view switching button of the XR controller;

[0069] 2) Operation Interaction Configuration: The signal control logic is mapped to interactive operations. Users can click on the virtual signal machine with the XR controller to trigger the signal switching operation. The operation result is fed back to the virtual scene in real time and synchronized to the real device through the data interoperability algorithm, realizing the operation-feedback-synchronization closed loop.

[0070] Preferably, in step S4, based on the resource topic segmentation and feature enhancement results of the AIGC platform, time-series feature data that meets the optimization requirements are selected, including:

[0071] 1) Train operation characteristics include time series. The speed corresponding to the time is Location is ;

[0072] 2) Signal control characteristics include the signal state sequence. The state transition time is ;

[0073] The continuous features are normalized using the extremum standardization method, and the formula is as follows:

[0074]

[0075] Construct the input sequence as The feature vector at each time step is represented as ;

[0076] Based on the time-series dependency between train operation and signal control, a Bi-LSTM model is constructed for dynamically correcting logic deviations. The model structure specifically includes:

[0077] Input layer: Receives the preprocessed temporal feature sequence The input dimension is 3, including velocity, position, and signal status;

[0078] Bi-LSTM layer: It contains one forward LSTM and one backward LSTM layer, with a hidden layer dimension of 64, to capture the forward and backward temporal dependencies of the sequence, including the impact of train speed changes on subsequent signal states and the constraints of signal states on historical speeds.

[0079] Fully connected layer: The bidirectional features output by Bi-LSTM are concatenated and mapped to the corrected feature dimensions, which are 3-dimensional and correspond to the corrected values ​​of velocity, position and signal state respectively.

[0080] Output layer: Outputs the corrected timing sequence as follows ,in .

[0081] Preferably, a loss function incorporating temporal smoothness constraints is designed, and the Bi-LSTM model parameters are optimized based on gradient descent, wherein the correction loss function is expressed as:

[0082]

[0083] in This represents the mean square error term, to ensure that the correction value is close to the true value and ;

[0084] This represents a timing smoothing term to ensure the continuity of train operation and signal control. ;

[0085] These are the weighting coefficients;

[0086] This represents the t-th step correction value of the Bi-LSTM output;

[0087] This is the true reference value at step t;

[0088] Indicates Euclidean distance;

[0089] The Adam optimizer is used to update the model parameters, including the weights W and biases b. The gradient descent optimization formula is as follows: ,in For the set of model parameters;

[0090] The learning rate;

[0091] This represents the gradient of the loss function with respect to the parameters in the k-th iteration. It occurs when the number of iterations reaches its maximum, or after 5 consecutive iterations. When the decrease is less than 0.0001, stop the iteration and output the corrected time series sequence;

[0092] The time-series data corrected by Bi-LSTM is applied to the XR simulation model, and a feedback mechanism is established for continuous optimization, specifically including:

[0093] 1) Adjust the corrected train speed ,Location Update the virtual train model to ensure a smooth running trajectory;

[0094] 2) The corrected signal state Synchronize to the virtual signal machine;

[0095] 3) Collect deviation data in XR simulation in real time. If Δp exceeds the threshold, add the deviation data as new samples to the training set and retrain the Bi-LSTM model.

[0096] Preferably, in step S5, the user-inputted scenario requirements are parsed using natural language processing technology and transformed into quantifiable configuration parameters. The user-input scenario description is then segmented and stop word removed to extract core keywords. Based on the XR simulation resource topic tag library, the keywords are mapped to resource topics. The TF-IDF algorithm is used to calculate the association weight between keywords and scenario types, expressed as:

[0097]

[0098] in Keywords In terms of scenario requirements The word frequency, i.e., the number of occurrences divided by the total number of words;

[0099] Inverse document frequency and N is the total number of scene types. For keywords The number of scene types;

[0100] The output scene feature vector is , Let be the association weight between the i-th topic and the scene;

[0101] Based on scene feature vectors, suitable model components are selected from the component library, and component adjustment parameters are calculated. The model component library is categorized according to topic segmentation logic, specifically as follows:

[0102] 1) The track component library includes straight tracks, curved tracks, and turnouts;

[0103] 2) The train component library includes Type A cars, Type B cars, and driverless trains;

[0104] 3) The signal component library includes color light signal generators, transponders, and ATP equipment;

[0105] The formula for calculating the matching degree between components and scenes is as follows: ,in This represents the matching degree between component c and scene s, with a value range of [0,1].

[0106] This indicates that component c and the i-th topic Similarity;

[0107] Select Components with a parameter value ≥0.7 are included in the candidate set. For each candidate component, the parameter adjustment value is calculated based on the scenario requirements. The algorithm formula for dynamically adjusting component parameters is as follows:

[0108]

[0109] in These are the adjusted component parameters;

[0110] These are the basic parameters of the component. For adjustment coefficients;

[0111] This is the parameter weight matrix, representing the weight of the influence of the topic on the parameter;

[0112] The adjusted components are spliced ​​together according to the coordinate parameters of the track components. A correlation is established between the train operating parameters and the control logic of the signal components. The compatibility index between the component parameters is calculated and expressed as:

[0113]

[0114] in Representation Component and Compatibility index;

[0115] Representation Component The core parameters;

[0116] For parameter matching coefficients;

[0117] Based on the application scenario requirements, output model formats adapted to different platforms and complete deployment configuration.

[0118] (III) Beneficial Effects

[0119] Compared with existing technologies, this invention provides a method for generating rail transit simulation configuration models based on AIGC and XR, which has the following advantages:

[0120] 1. This method for generating a rail transit simulation configuration model based on AIGC and XR uses the topic segmentation and feature enhancement technology of the AIGC platform to select resource blocks with high topic relevance through Bayesian probability formula to eliminate interference from overlapping feature attributes. At the same time, CNN is introduced to verify the parameters of model components, and combined with loss function constraint fitting, the accuracy of XR simulation resource extraction and model parameter accuracy is improved.

[0121] 2. The method for generating a rail transit simulation configuration model based on AIGC and XR automatically generates high-precision 3D model components according to the type of rail transit elements through the deep learning generation framework of the AIGC platform. Then, the component features are automatically fused by unpooling amplification and convolutional stitching to achieve automated modeling.

[0122] 3. This method for generating a rail transit simulation configuration model based on AIGC and XR combines SLAM spatial positioning and data interoperability technology to achieve alignment between the virtual scene and the real physical space, while supporting real-time synchronization of virtual model and real equipment data. Users can switch between global view, device view, and first-person view through an XR headset, and operate signal switching, train scheduling, and other functions through a controller, forming an operation-feedback-synchronization closed loop, realizing deep integration of XR virtual and real and enhancing the immersive interactive experience.

[0123] 4. This method for generating a rail transit simulation configuration model based on AIGC and XR analyzes user requirement text, extracts key parameters using the TF-IDF algorithm, selects model components with high matching degree, and dynamically adjusts the modularity to adapt to diverse scenarios by combining parameters. Attached Figure Description

[0124] Figure 1 This is a flowchart of the method for generating a simulation configuration model of rail transit according to the present invention;

[0125] Figure 2 This is a schematic diagram illustrating the division of thematic block feature attributes of the rail transit industry-education integration XR simulation resources of the present invention;

[0126] Figure 3 This is a schematic diagram illustrating the relationship between output loss and iteration count during the convolution operation of the feature attributes of the rail transit industry-education integration XR simulation resources.

[0127] Figure 4 This diagram illustrates the extraction effects of different methods of the present invention on XR simulation resource content for testing rail transit industry-education integration. Detailed Implementation

[0128] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. 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.

[0129] Example 1

[0130] In this embodiment, the steps for preprocessing XR simulation resource data on the AIGC platform are as follows:

[0131] 1) Read in the HTML page information corresponding to the rail transit XR simulation resources, automatically identify and remove redundant tags that do not contain the theme content, including Script tags, form tags, Style tags and if-frame tags;

[0132] 2) Abnormal Tag Correction: Automatically detects and corrects abnormal tags in HTML pages that have syntax errors or logical conflicts, including unclosed div tags and incorrectly nested tab tags, to ensure that the tag structure conforms to the Document Object Model specification;

[0133] 3) Convert the filtered and corrected XR simulation resource data into a DOM tree structure, and use the hierarchical characteristics of the DOM tree to present the node relationships of the resource data;

[0134] The steps for pre-dividing resource data into chunks based on topic-related tags according to the DOM tree are as follows:

[0135] 1) Select tags that are strongly related to the theme of rail transit XR simulation resources as the basis for block division, specifically including table tag, p tag, tr tag and div tag, among which table tag is used as table tag, corresponding to structured data such as line layout and timetable;

[0136] The 'p' tag is used as a paragraph tag, corresponding to text data such as scheduling rules and emergency plans;

[0137] The `` tag is used as the table row label, corresponding to the row-level data of the device parameters;

[0138] The `div` tag is used as the partition tag, corresponding to the component partition data of the 3D model;

[0139] 2) Nested tag processing rules: When there is a nested relationship between tags, the nearest block tag containing the core information of the rail transit XR simulation resources is used as the block division execution benchmark, and the tag that directly wraps the valid data is selected for block division first.

[0140] Example 2

[0141] In this embodiment, the coordinates of the XR virtual scene and the real physical space are aligned using a spatial alignment algorithm to ensure synchronous matching of the virtual scene's viewpoint when moving in the real space. The spatial alignment loss function is expressed as follows:

[0142]

[0143] in This represents the spatial alignment loss value, ranging from [0, +∞). A smaller value indicates higher alignment accuracy between virtual and real spaces. The optimization objective is to minimize the loss value, setting a threshold accordingly. Alignment is considered acceptable if the deviation is ≤0.05m. These are the weighting coefficients; / This represents the device coordinates in the XR virtual space, obtained through the XR headset's SLAM positioning module. / The coordinates of the device in the real physical space are obtained through an infrared positioning system; S is the virtual-to-real space scaling factor; the spatial alignment steps include:

[0144] 1) Obtain the initial coordinates of the origin in the real space and the origin in the virtual space, and set... And calculate the initial alignment loss. ;

[0145] 2) Iterative adjustment The coordinates of the virtual space origin are recalculated after each iteration. until ≤0.05 m;

[0146] 3) Collect in real time when the user moves / , substitute into the calculation , if it exceeds the threshold, trigger a new round of iteration.

[0147] Realize the real-time synchronization of data between the XR virtual model and the real rail transit equipment according to the data interconnection algorithm. The data synchronization error formula is:

[0148]

[0149] Where represents the virtual and real data synchronization error, and the value range is [0, +∞). Set ≤5% as qualified synchronization; is the data value in the XR virtual model; is the real-time data value of the real rail transit equipment, obtained from the real equipment control system through the industrial Ethernet; is the data conversion coefficient, used to unify the dimensions of virtual and real data; The steps of its data interconnection include:

[0150] 1) Synchronously collect and at an interval of 100 ms, and set the collection frequency according to the dynamic resource timing;

[0151] 2) Error calculation: Substitute into the calculation , if ≤5%, keep the data synchronized; if >5%, then assign to , realize the calibration of virtual data to real data;

[0152] 3) Exception handling: Based on the processing logic of correcting the problematic tags, when exceeds the normal range, trigger a data exception alarm, pause synchronization and retain the exception data log.

[0153] Example 3

[0154] In this example, the forward LSTM unit in the Bi-LSTM layer calculates to process the temporal dependence from the past to the future, and its formula is:

[0155]

[0156] The backward LSTM unit in the Bi-LSTM layer calculates to process the temporal dependence from the future to the past, and its formula is:

[0157]

[0158] The output concatenation is implemented by a Bi-LSTM layer, which is used to concatenate the forward and backward hidden states. The dimension is 128, and it is represented as follows:

[0159]

[0160] in , , These are the outputs of the forget gate, input gate, and output gate, respectively, with values ​​ranging from [0,1], controlling the discarding and retention of information. The cell state has a value range of [−1,1] and stores long-term temporal information; This represents the candidate cell state, with a value range of [−1, 1], and is used to update the cell state. The hidden state has a value range of [−1, 1], and the output is the feature representation of the current time step; , , , All are gated weight matrices. , , , All are gated bias vectors; It is the sigmoid activation function. It is the hyperbolic tangent activation function; [;] represents element-wise multiplication, and [;] represents vector concatenation.

[0161] Example 4

[0162] In this embodiment, according to Figure 2 The thematic block feature attribute information shown is used to segment the themes of the test rail transit industry-education integration XR simulation resources into blocks. The relationship between the XR simulation resource blocks and thematic blocks is calculated using the correlation probability formula between the XR simulation resource blocks and thematic blocks. Represented as ;

[0163] according to After dividing the AIGC platform's test rail transit industry-education integration XR simulation resource topics into blocks, the convolution kernel sizes for the AIGC platform's deep learning architecture to perform convolution operations on the block feature attributes of the rail transit industry-education integration XR simulation resource topics are set to 3×3 and 1×1, respectively. Simultaneously, let... The value is 0.1, and the relationship between the output loss and the number of iterations is as follows: Figure 3 As shown;

[0164] according to Figure 3As shown, when the number of iterations reaches 30, the loss of the fitting output result of the deep learning architecture of the AIGC platform for the feature attributes of the segmented resources of the rail transit industry-education integration XR simulation is close to 0. Therefore, the number of iterations is set to 30 to carry out resource content extraction. The rail transit industry-education integration XR simulation resources and resource segmented feature attribute files can be extracted and output, indicating that the method can effectively perform the resource content extraction task on the AIGC platform. The reason is that the feature attributes are used to segment the resource topics, solve the problem of resource feature overlap, and thus achieve the fitting output of the target resource.

[0165] Example 5

[0166] In this embodiment, accuracy is used as the evaluation index. A network small-sample data feature extraction method based on convolutional neural networks and an automated extraction method for in-service road surface based on point cloud data are set as control groups for the test. The extraction effects of different methods on the content of XR simulation resources for the integration of industry and education in rail transit are compared and analyzed. Figure 4 The test results for different methods are shown below;

[0167] according to Figure 4 The test results show that the method achieves the highest accuracy in extracting content from XR simulation resources for rail transit industry-education integration, reaching over 0.92, thus outperforming the control group. This indicates that the designed method can extract content from XR simulation resources for rail transit industry-education integration. This is because the method uses the deep learning architecture of the AIGC platform to accurately fit the block feature attributes of XR simulation resources for rail transit industry-education integration, and uses loss to constrain the fitting results, ensuring that the content extraction results have a high accuracy.

[0168] The beneficial effects of this invention are: the method for generating a rail transit simulation configuration model based on AIGC and XR, according to the topic segmentation and feature enhancement technology of the AIGC platform, uses the Bayesian probability formula to screen resource blocks with high topic relevance to eliminate the interference of overlapping feature attributes, and introduces CNN to verify the model component parameters, combined with loss function constraint fitting, to improve the accuracy of XR simulation resource extraction and model parameter;

[0169] The AIGC platform's deep learning generation framework automatically generates high-precision 3D model components based on the type of rail transit elements. Then, it uses unpooling to amplify the components and convolutional to stitch them together to achieve automatic feature fusion and automated modeling.

[0170] By combining SLAM spatial positioning and data interoperability technologies, the virtual scene is aligned with the real physical space, while supporting real-time synchronization of virtual model and real device data. Users can switch between global view, device view, and first-person view through the XR headset, and operate functions such as signal switching and train scheduling through the controller, forming an operation-feedback-synchronization closed loop, realizing deep integration of XR virtual and real and enhancing the immersive interactive experience.

[0171] By parsing user requirement text, the TF-IDF algorithm is used to extract key parameters, select model components with high matching degree, and dynamically adjust the modularity to adapt to diverse scenarios based on the parameters.

[0172] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating a simulation configuration model of rail transit based on AIGC and XR, characterized in that, Includes the following steps: S1. It covers the collection of static and dynamic core data from multiple sources, performs preprocessing operations based on the AIGC platform, and generates a standardized dataset; S2. The deep learning generation framework based on the AIGC platform generates model components in modules according to the type of rail transit elements, and introduces a convolutional neural network to perform preliminary verification of the parameters of the generated model components; S3. Based on the XR engine, the 3D model components generated by AIGC are assembled into a scene according to the actual line layout, and spatial positioning technology is used to align the virtual simulation scene with the real physical space. S4. Construct a bidirectional long short-term memory network to dynamically correct the train operation logic and signal control mechanism, and optimize the XR interaction logic based on user operation data using reinforcement learning (RL). S5. The AIGC platform uses NLP to parse user requirement text, automatically extracts key requirement parameters, and supports model component and scenario adjustment based on modular design, enabling configuration model output and deployment.

2. The AIGC and XR-based rail transit simulation configuration model generation method according to claim 1, characterized in that, In step S1, XR simulation resource data is collected from the rail transit system. The XR simulation resource data is divided into static resource data and dynamic operation data. The static resource data includes spatial structure data of the track layout, geographic coordinate data of station locations, basic parameters of equipment 3D models, and layout data of platform facilities. The dynamic operation data includes train timetables, route planning data, train speed curve data, and text and time-series data used to simulate the scheduling and control process of the rail transit system. The text and time-series data includes train scheduling rules, emergency response plans, and signal control logic. The collected XR simulation resource data is preprocessed using the AIGC platform. Based on the DOM tree, the resource data is pre-divided into blocks using topic-related tags.

3. The AIGC and XR-based rail transit simulation configuration model generation method according to claim 2, characterized in that, The association probability between each block and the topic block is calculated using the Bayesian probability formula. Blocks with high relevance to the topic are selected, and the association between the blocks and the topic block is quantified. The association probability calculation formula is as follows: wherein is the probability of association, representing the probability that the block B belongs to the topic S given that the block B contains the feature attribute is used to quantify the association of the block with the topic, the probability value closer to 1 indicates the stronger topic relevance of the block.​ B represents a single block result after pre-blocking the XR simulation resource data, and each B corresponds to a segment of structured resource data; represents a feature attribute contained in the block B, and all feature attributes of the blocks B constitute a feature attribute vector , S represents a theme block of the rail transit XR simulation resource, that is, a reference block capable of reflecting the core content of the resource; The prior probability represents the probability that any one block B belongs to the topic block S without considering the feature attribute, and can be determined by counting the proportion of the topic block in the historical block data. is the conditional probability that a block B contains a feature property given that the block B belongs to a topic block S, and is used to measure the strength of association between a topic block and a feature property. is the conditional probability that a block B contains a feature property given that the block B belongs to a topic block S, and is used to measure the strength of association between a topic block and a feature property. For marginal probability, it represents the probability that any one of the divided blocks of all rail transit XR simulation resources contains the feature attribute The probability can be calculated by counting the proportion of the number of blocks containing in all blocks. Setting the threshold value of the association probability The association probability higher than the threshold value is reserved as the valid data block, and the association probability lower than the threshold value is determined as the non-topic block and eliminated. Finally, the preprocessed valid resource data set is output.

4. The AIGC and XR-based rail transit simulation configuration model generation method according to claim 1, characterized in that, In step S2, based on the AIGC platform's ability to process XR simulation resource tags, the extracted resource features are mapped to model component attributes using tags. Specifically, this includes: 1) The effective block features extracted by topic segmentation are categorized by label according to the model component type. The effective block features include train size data, track topology parameters and signal control logic. The model component types include train, track, signal system and platform facilities. 2) Standardize the categorized feature labels and unify the label format; By utilizing the inverse pooling operation in the AIGC platform's deep learning architecture, the key feature attributes of the mapped model components are amplified. The formula for feature attribute amplification is as follows: wherein represents the magnification result of the i-th model component feature property; denotes a pooling inverse operation for restoring and amplifying the dimension and numerical precision of the feature attributes; represents the original feature attribute of the i-th model component; represents the output result of the i-1th model component feature attribute after the convolution operation; The amplified feature attributes are concatenated with the features of adjacent model components through convolution operations to form a complete set of model component features. The feature attribute concatenation formula is as follows: wherein represents the final feature output of the i-th model component after two convolution operations; represents a convolution operation in the deep learning architecture of the AIGC platform, used to integrate different feature attributes; represents the zoomed feature of the i-th component with the original feature of the i+1-th component dimensional concatenation; Based on the standard parameters of actual rail transit equipment, the features of the AIGC-enhanced model components are fitted. The steps include: 1) Fit the fused features to the rail transit industry-education integration XR simulation resources, construct a loss function to constrain the output results, and express the loss function as follows: wherein represents the loss value of model component feature fitting, the smaller the loss value, the smaller the deviation between the fitting result and the actual standard parameter; is a constant used to adjust the weight of the two fitted scenarios; a state indicating the fitting result of the model component feature, when x = 1, it indicates that the fitting result is a non-empty set, that is, the effective features are successfully extracted and the fitting is completed, when x = 0, it indicates that the fitting result is an empty set, that is, the feature extraction or fitting fails; The model component feature output after the convolution operation is used to participate in the calculation of the loss value, ensuring that the loss constraint is strongly associated with the feature attributes. 2) Set the number of iterations, by continuously adjusting the model component feature parameters, so that the loss value tends to 0, and the final output meets the model component feature set.

5. The AIGC and XR-based rail transit simulation configuration model generation method according to claim 1, characterized in that, In step S3, based on the hierarchical structure of the DOM tree and the topic segmentation results of the XR simulation resources, the valid XR simulation resources extracted through the AIGC platform are imported into the XR base engine, categorized as static resources and dynamic resources, as follows: 1) Complete node mapping of the equipment 3D model parameters in the XR engine, including tracks, platforms, and signals; 2) Convert the time-series data into script parameters that the XR engine can recognize. The time-series data includes train timetables, speed curves, and signal control logic. Based on the DOM tree node hierarchy, the resource structure is decomposed into a three-level hierarchy: root node - functional node - attribute node, thus completing the structured parsing of the resources. Specifically: 1) Root node: corresponds to the overall rail transit simulation scenario; 2) Functional nodes: Corresponding to the functional partitions within the scene, based on the Bayesian probability results of topic segmentation, blocks with topic relevance ≥ 0.3 are selected as the data source for functional nodes; 3) Attribute nodes: correspond to the specific resource attributes under the function nodes; The parsed root node-functional node-attribute node are mapped to the scene layer-object layer-parameter layer of the XR engine, specifically including: 1) Scene layer: Load the overall scene coordinate system corresponding to the root node, and define the XR virtual coordinate system based on the actual geographic coordinates of the track line; 2) Object layer: Instantiate the block resources corresponding to the functional nodes into virtual objects in the XR engine; 3) Parameter layer: Assigns the resource parameters corresponding to the attribute nodes to the component parameters of the virtual object; Based on the enhanced feature attributes of the XR simulation resources, the component assembly and rendering configuration of the virtual scene are completed. The specific steps are as follows: 1) Static component assembly: Based on the track layout data, the virtual track objects are spliced ​​in the XR scene layer in the order of main branches, and the platform and signal are placed as static objects at the specified coordinate positions. 2) Dynamic component association: Associating train operation data with train virtual objects; 3) Based on the texture features of XR simulation resources, an adaptive texture resolution algorithm is used to adjust the rendering accuracy; Based on the AIGC platform's deep learning framework and feature fitting logic, this system achieves the fusion and interaction of virtual and real scenes using spatial alignment and data interoperability algorithms. It also configures basic interactive functions according to the interaction requirements of XR simulation resources. Specific operations include: 1) Multi-view interaction configuration: Based on the multi-dimensional features of resource segmentation, global view, device view and first-person view are set in the XR environment, which is triggered by the user through the view switching button of the XR controller; 2) Operation Interaction Configuration: The signal control logic is mapped to interactive operations. Users can click on the virtual signal machine with the XR controller to trigger the signal switching operation. The operation result is fed back to the virtual scene in real time and synchronized to the real device through the data interoperability algorithm, realizing the operation-feedback-synchronization closed loop.

6. The AIGC and XR-based rail transit simulation configuration model generation method according to claim 3, characterized in that, In step S4, based on the resource topic segmentation and feature enhancement results of the AIGC platform, time-series feature data that meet the optimization requirements are selected, including: 1) Train operation features include time sequence as , time corresponding speed as , position as ; 2) the signal control feature includes a signal aspect sequence of , and a state switching time of ; The continuous features are normalized using the extremum standardization method, and the formula is as follows: The input sequence is constructed as where the feature vector at each time step is represented as ; Based on the time-series dependency between train operation and signal control, a Bi-LSTM model is constructed for dynamically correcting logic deviations. The model structure specifically includes: Input layer: receives the pre-processed time series feature sequence with input dimension 3 including velocity, position and signal state; Bi-LSTM layer: It contains one forward LSTM and one backward LSTM layer, with a hidden layer dimension of 64, to capture the forward and backward temporal dependencies of the sequence, including the impact of train speed changes on subsequent signal states and the constraints of signal states on historical speeds. Fully connected layer: The bidirectional features output by Bi-LSTM are concatenated and mapped to the corrected feature dimensions, which are 3-dimensional and correspond to the corrected values ​​of velocity, position and signal state respectively. Output layer: output the corrected timing sequence as wherein .

7. The AIGC and XR-based rail transit simulation configuration model generation method according to claim 6, characterized in that, Design a loss function that incorporates temporal smoothness constraints, and optimize the Bi-LSTM model parameters based on gradient descent. The correction loss function is expressed as: wherein represents the mean square error term, and ; denotes a timing smoothing term, and ; are weight coefficients; denotes the Bi-LSTM output at time step t; true reference value for step t; denotes the Euclidean distance; The model parameters, including the weights W and the bias b, are updated using an Adam optimizer, and the gradient descent optimization formula is wherein is a set of model parameters; learning rate; represents the gradient of the loss function with respect to the parameters at iteration k, when the number of iterations reaches the maximum number of iterations, or when the decrease in amplitude is less than 0.0001 for 5 consecutive iterations when the decrease in amplitude is less than 0.0001, the iterations are stopped and the corrected time series is output The time-series data corrected by Bi-LSTM is applied to the XR simulation model, and a feedback mechanism is established for continuous optimization, specifically including: 1) Adjust the corrected train speed ,Location Updated to virtual train model; 2) The corrected signal state Synchronize to the virtual signal machine; 3) Collect deviation data in XR simulation in real time. If Δp exceeds the threshold, add the deviation data as new samples to the training set and retrain the Bi-LSTM model.

8. The method for generating a rail transit simulation configuration model based on AIGC and XR according to claim 1, characterized in that, In step S5, natural language processing (NLP) is used to parse the user-inputted scenario requirements and convert them into quantifiable configuration parameters. The user-input scenario description is segmented and stop word removed to extract core keywords. Based on the XR simulation resource topic tag library, the keywords are mapped to resource topics. The TF-IDF algorithm is used to calculate the association weight between keywords and scenario types, expressed as: in Keywords In terms of scenario requirements The word frequency, i.e., the number of occurrences divided by the total number of words; Inverse document frequency and N is the total number of scene types. For keywords The number of scene types; The output scene feature vector is , Let be the association weight between the i-th topic and the scene; Based on scene feature vectors, suitable model components are selected from the component library, and component adjustment parameters are calculated. The model component library is categorized according to topic segmentation logic, specifically as follows: 1) The track component library includes straight tracks, curved tracks, and turnouts; 2) The train component library includes Type A cars, Type B cars, and driverless trains; 3) The signal component library includes color light signal generators, transponders, and ATP equipment; The formula for calculating the matching degree between components and scenes is as follows: ,in This represents the matching degree between component c and scene s, with a value range of [0,1]. This indicates that component c and the i-th topic Similarity; Select Components with a parameter value ≥0.7 are included in the candidate set. For each candidate component, the parameter adjustment value is calculated based on the scenario requirements. The algorithm formula for dynamically adjusting component parameters is as follows: in These are the adjusted component parameters; These are the basic parameters of the component. For adjustment coefficients; This is the parameter weight matrix, representing the weight of the influence of the topic on the parameter; The adjusted components are spliced ​​together according to the coordinate parameters of the track components. A correlation is established between the train operating parameters and the control logic of the signal components. The compatibility index between the component parameters is calculated and expressed as: in Representation Component and Compatibility index; Representation Component The core parameters; For parameter matching coefficients; Based on the application scenario requirements, output model formats adapted to different platforms and complete deployment configuration.

Images