A high-speed rail application software cloud testing method and system based on digital twinning
By combining digital twin modeling and transfer neural networks, the problems of large data volume, poor real-time performance, high cost, and insufficient multi-module collaborative testing capabilities in the application software testing of high-speed railway station ticket gates were solved. This enabled efficient and accurate cloud testing, reduced testing costs and cycles, and improved test coverage and result accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING OUTASITE TECHNOLOGY DEVELOPMENT CO LTD
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-12
AI Technical Summary
Existing testing methods for high-speed railway station ticket gate application software suffer from problems such as large test data volume, low effectiveness, poor real-time performance, high cost, insufficient multi-module collaborative testing capabilities, and poor adaptability to passenger flow scenarios.
By combining digital twin modeling, active learning strategies, and transfer neural networks, and by constructing a randomly configured network (SCN) and a transfer neural network (TNN), efficient, accurate, and low-cost cloud testing can be achieved under different traffic scenarios.
By combining digital twin modeling and transfer neural networks, efficient and accurate testing under different traffic flow scenarios is achieved, reducing testing costs and cycles, improving test coverage and result accuracy, and meeting the real-time testing needs of complex scenarios.
Smart Images
Figure CN122195846A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital twin technology, and more specifically to a cloud testing method and system for high-speed rail application software based on digital twins. Background Technology
[0002] Currently, with the continuous improvement of the high-speed rail network and the enhancement of its intelligent level, the ticket gates at high-speed rail stations, as key equipment for passenger passage, directly impact passenger travel experience and station operational efficiency due to the stability, reliability, and functionality of their application software. High-speed rail station ticket gates integrate multiple modules such as cameras, displays, voice and light alerts, and gate mechanisms. The corresponding application software needs to achieve complex functions such as multi-module collaborative control, real-time data processing, and abnormal scenario response.
[0003] However, existing testing methods for high-speed rail application software mostly adopt physical testing or simple simulation testing modes: physical testing requires building a real gate environment, which has problems such as high equipment cost, incomplete coverage of test scenarios, and long test cycle; simple simulation testing mostly models single functional modules, ignores the coupling relationship between modules, and requires processing massive amounts of redundant data during the testing process, resulting in low testing efficiency, high resource consumption, and high testing costs.
[0004] Specifically, the existing technology has the following shortcomings: Low validity of test data: The gate operation data and simulated passenger operation data collected during the test contain a large number of worthless samples, which not only increases the data processing burden, but may also lead to bias in test results and affect the accuracy of software problem localization. Poor real-time performance of tests: Traditional test algorithms have a slow iteration speed, making it difficult to adapt to the real-time response test requirements of ticket gate application software. Especially in complex test scenarios such as dense passenger flow during peak hours and multiple abnormal scenarios occurring concurrently, test results cannot be fed back in a timely manner. High testing costs: Physical testing requires a large investment in gate equipment, site resources and manpower. In addition, data storage, processing and test model updates during the testing process consume a lot of computing resources, resulting in high overall testing costs. Insufficient multi-module collaborative testing capabilities: The functions of each hardware module of the ticket gate are closely related, and existing testing methods are difficult to simulate the complex scenario of multi-module collaborative work, and cannot fully verify the collaborative control capabilities of the application software. Poor adaptability to pedestrian traffic: Pedestrian traffic varies significantly at different times (peak / off-peak / low-peak) and in different scenarios (holidays / weekdays). Existing testing methods are difficult to quickly transfer testing experience under different pedestrian traffic conditions, and test models need to be repeatedly built, which further increases testing costs and time.
[0005] Therefore, how to propose a cloud testing method and system for high-speed rail application software based on digital twins, overcome the defects in existing technologies, and efficiently and cost-effectively complete the comprehensive testing of high-speed rail station ticket gate application software under different passenger flow scenarios is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0006] In view of this, the present invention provides a cloud testing method and system for high-speed rail application software based on digital twins, aiming to solve the technical problems existing in the testing of high-speed rail station ticket gate application software, such as large test data volume, low validity, poor test real-time performance, high cost, insufficient multi-module collaborative testing capabilities, and poor adaptability to passenger flow scenarios. By combining digital twin modeling, active learning strategies, and transfer neural networks, efficient, accurate, and low-cost cloud testing under different passenger flow scenarios is achieved. To achieve the above objectives, the present invention adopts the following technical solution: A cloud testing method for high-speed rail application software based on digital twins includes: Collect historical operational data of high-speed railway station ticket gates, application software testing data, and passenger flow data from multiple scenarios to construct a historical dataset; An initial digital twin model of the ticket gate was constructed based on a randomly configured network (SCN). The SCN was trained using a historical dataset to obtain an initial test model. By actively learning and operating, using the initial test model, value is screened from samples that have not participated in the test, suitable value samples are selected and processed to obtain valuable test data, and the initial test model is updated based on the valuable test data to obtain a suitable digital twin test model. Construct a transfer neural network (TNN), train the TNN based on valuable test data from the source pedestrian traffic scenario, and perform test knowledge transfer under different pedestrian traffic scenarios; Based on the applicable digital twin testing model and the trained transfer neural network (TNN), the real-time test data stream of the target traffic scenario is filtered and adapted, valuable real-time test data is saved, the testing model is continuously optimized, and the application software testing is completed.
[0007] Optionally, the initial model includes a digital mapping module for a camera, display screen, voice and light alerts, and gate device.
[0008] Optionally, a training set is set according to the historical dataset. The training set includes feature data and label data corresponding to the feature data. The random configuration network SCN is trained using the training set to obtain an initial test model.
[0009] Optionally, the historical operational data and application software testing-related data include: Data from cameras, displays, voice and light alerts, turnstiles, application software, and pedestrian traffic in various scenarios.
[0010] Optionally, training a randomly configured network (SCN) using historical datasets to obtain an initial test model includes: S21. Given a test objective function, the objective function is: Where x represents the test input data and y represents the expected test output result; let the hidden layer of the randomly configured network SCN be the... Each node has been generated; calculate the current network output.
[0011] S22. Calculate the current network residual vector.
[0012] S23. Obtain residual vector parameters based on the current network residual vector processing, and perform a judgment operation based on the residual vector parameters. Or the preset maximum number of nodes has not been reached. When this happens, the number of random configuration networks (SCNs) is increased. The hidden layer node determines the first... Input weights and biases of each hidden layer node; S24. Calculate the output weights of the hidden layer nodes.
[0013] S25. Calculate the current network output result again based on the output weights; S26. Based on the training set, repeat steps S21 to S25 in a loop to construct the initial test model. .
[0014] Optionally, the determination of the first The input weights and biases of each hidden layer node include: ; ; in, , Represents the hidden layer node The output, express Transpose of; Represents the hidden layer node Candidate input weights, Represents the hidden layer node The candidate bias; λ∈(0,1) is the introduced constraint; Represents a sequence of nonnegative real numbers
[0015] Will satisfy The candidate node parameters are used as the parameters of the Hth hidden layer node. Indicates the constraint value. Indicates the constraint value The One portion, This indicates that the components of the constraint value are summed. Indicates to Activate, This indicates the hidden layer node. Candidate input weights transpose, Features representing the input The One portion, Represents the current network residual vector The One portion, Represents the current network residual vector The Transpose of each component.
[0016] Optionally, the step of actively learning, using an initial test model, performing value screening on samples not participating in the test, selecting and processing applicable value samples, and obtaining valuable test data includes: S31. Determine the current initial test model.
[0017] S32. If so, calculate the sample uncertainty for a single model. To filter and obtain valuable test data for individual models; S33. If not, calculate the contribution value of the sample to different models and filter out the valuable test data for multiple models.
[0018] S34. Save the valuable test data of the single model and the valuable test data of the multiple models to constitute the valuable test data. .
[0019] Optionally, constructing the transfer neural network (TNN) includes: S41. Utilize the feature extraction layer of TNN to extract valuable test data from the source scene. Feature extraction is performed to obtain common features of the scene and unique features of the source scene.
[0020] S42. Minimize the distribution difference between the source scene and the target scene through the domain adaptation layer, and use the maximum mean difference as the domain adaptation loss function.
[0021] S43. Establish a mapping relationship between pedestrian flow parameters and test model parameters through the traffic scene mapping layer.
[0022] S44. Using the source scene test accuracy and domain adaptation loss as joint optimization objectives, train the TNN parameters to obtain a transfer test model adapted to the target scene. .
[0023] Optionally, the continuous optimization of the test model and the completion of application software testing include: S51. Collect real-time test data streams under target passenger flow scenarios, including passenger operation data, gate module response data, and flow dynamic monitoring data; S52. Input the real-time data stream into the applicable digital twin test model. Make an initial value judgment and input it into the transfer test model. Perform scene adaptation corrections to obtain real-time valuable data judgment results for the target scene.
[0024] S53, based on Synchronous updates and ; S54, based on the optimized and Complete the functional testing, performance testing, and abnormal scenario testing of the application software under the target scenario, and output the test report.
[0025] Optionally, a cloud testing system for high-speed rail application software based on digital twins includes: Data Acquisition Module: Used to collect historical operating data of high-speed railway station ticket gates, application software testing data, and passenger flow data from multiple scenarios to build a historical dataset; Training module: Used to build an initial digital twin model of the ticket gate based on the Randomly Configured Network (SCN), and to train the SCN using historical datasets to obtain the initial test model; Update module: It is used to actively learn and operate, use the initial test model to screen for value in samples that have not participated in the test, select and process applicable value samples, process them to obtain valuable test data, update the initial test model based on the valuable test data, and obtain an applicable digital twin test model. Knowledge Transfer Module: Used to build a transfer neural network (TNN), train the TNN based on valuable test data from the source pedestrian traffic scenario, and perform test knowledge transfer under different pedestrian traffic scenarios; The testing module is used to filter and adapt real-time test data streams for target pedestrian traffic scenarios based on applicable digital twin testing models and trained transfer neural networks (TNNs), save valuable real-time test data, continuously optimize the testing model, and complete application software testing. As can be seen from the above technical solution, compared with the prior art, the present invention discloses a cloud testing method and system for high-speed rail application software based on digital twins, which has the following beneficial effects: This invention discloses a cloud testing method for high-speed rail application software based on digital twins, comprising: collecting historical operating data of high-speed rail station ticket gates, application software testing-related data, and multi-scenario passenger flow data to construct a historical dataset; constructing an initial digital twin model of the ticket gates based on a Randomized Network (SCN), training the SCN using the historical dataset to obtain an initial test model; through active learning operations, using the initial test model, performing value screening on samples not participating in the test, selecting and processing applicable value samples to obtain valuable test data, updating the initial test model based on the valuable test data to obtain an applicable digital twin test model; constructing a Transfer Neural Network (TNN), training the TNN based on the valuable test data from the source passenger flow scenarios, and performing test knowledge transfer under different passenger flow scenarios; Based on the applicable digital twin testing model and the trained transfer neural network (TNN), the real-time test data stream of the target traffic scenario is filtered and adapted, valuable real-time test data is saved, the testing model is continuously optimized, and the application software testing is completed. This invention achieves knowledge transfer for testing across different traffic flow scenarios through transfer neural networks, eliminating the need to repeatedly build test models for each target scenario. This significantly reduces testing costs and timelines for special scenarios such as extreme peaks and sudden surges in passenger flow. It employs an active learning strategy to filter valuable data and eliminate redundant samples. Combined with the lightweight design of transfer neural networks, it reduces data processing volume and computational resource consumption, improving testing efficiency. The randomly configured network (SCN) supports incremental learning, enabling rapid model updates. The transfer neural network can quickly adapt to dynamically changing traffic scenarios, meeting the real-time testing needs of complex scenarios such as peak hours and sudden surges in passenger flow. It accurately maps each hardware module of the turnstile using a digital twin model, and combines this with the transfer neural network's adaptation to module collaboration logic under different traffic conditions, comprehensively verifying the multi-module collaborative control capabilities of the application software. This significantly improves test coverage and result accuracy. By replacing some physical testing with cloud testing, transfer learning reduces repetitive modeling workload, and data filtering optimizes resource consumption, reducing the overall lifecycle cost of testing from multiple dimensions, including equipment investment, labor costs, and computational resources. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0027] Figure 1 This invention provides a schematic flowchart of a cloud testing method for high-speed rail application software based on digital twins.
[0028] Figure 2 This invention provides a structural framework diagram of a cloud testing system for high-speed rail application software based on digital twins. Detailed Implementation
[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the 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.
[0030] This invention discloses a cloud testing method for high-speed rail application software based on digital twins, such as... Figure 1 As shown, it includes: Collect historical operational data of high-speed railway station ticket gates, application software testing data, and passenger flow data from multiple scenarios to construct a historical dataset; An initial digital twin model of the ticket gate was constructed based on a randomly configured network (SCN). The SCN was trained using a historical dataset to obtain an initial test model. By actively learning and operating, using the initial test model, value is screened from samples that have not participated in the test, suitable value samples are selected and processed to obtain valuable test data, and the initial test model is updated based on the valuable test data to obtain a suitable digital twin test model. Construct a transfer neural network (TNN), train the TNN based on valuable test data from the source pedestrian traffic scenario, and perform test knowledge transfer under different pedestrian traffic scenarios; Based on the applicable digital twin testing model and the trained transfer neural network (TNN), the real-time test data stream of the target traffic scenario is filtered and adapted, valuable real-time test data is saved, the testing model is continuously optimized, and the application software testing is completed.
[0031] Furthermore, the initial model includes a digital mapping module for a camera, a display screen, voice and light alerts, and a gate device.
[0032] Furthermore, a training set is set according to the historical dataset. The training set includes feature data and label data corresponding to the feature data. The random configuration network SCN is trained using the training set to obtain an initial test model.
[0033] Furthermore, the historical operational data and application software testing-related data include: Camera data, display screen data, voice and light alert data, turnstile data, application software operation data, and multi-scenario pedestrian flow data; Camera data: face verification success rate, ticket recognition accuracy, recognition response time, and abnormal image data; Display screen data: information display response speed, display content accuracy, and abnormal display logs; Voice and light alert data: voice broadcast accuracy, audio-visual alert synchronization, volume and brightness parameters, and abnormal warning trigger records; Turnstile device data: door opening and closing response time, channel control accuracy, number of fault triggers, and mechanical motion coordination parameters; Application software runtime data: module communication latency, data processing rate, exception and error logs, and concurrent request processing capacity data; The multi-scenario passenger flow data includes: passenger density, average passage time, number of concurrent ticket check requests, rate of change of passenger flow, and differences in passenger behavior, such as rapid passage or hesitation.
[0034] Furthermore, the step of training a randomly configured network (SCN) using historical datasets to obtain an initial test model includes: S21. Given a test objective function, the objective function is: Where x represents the test input data and y represents the expected test output result; Suppose that the hidden layer of the randomly configured network SCN is... Each node has been generated. Calculate the current network output: ； , , , express A function represented by each hidden layer node. Indicates network input. Represents hidden layer nodes The output weights, This represents the activation function. Indicates the hidden layer number 1 The input weights of each node, This represents the transpose of the input weights. Indicates the hidden layer number 1 The bias of each node The maximum number of nodes is preset. S22. Calculate the current network residual vector:
[0035] in, For the residual vector, This represents the function corresponding to the expected test output result. Indicates the first Each residual vector Indicates the total number of models. This represents the maximum index value of the residual vector; S23. Obtain residual vector parameters based on the current network residual vector processing, and perform a judgment operation based on the residual vector parameters. The preset error was not met. Or the preset maximum number of nodes has not been reached. When this happens, the number of random configuration networks (SCNs) is increased. The hidden layer node determines the first... Input weights and biases of each hidden layer node; S24. Calculate the output weights of the hidden layer nodes. : ; in, express Moore-Penrose generalized inverse, The output set of hidden layer nodes. This indicates the expected output label corresponding to the test input data; S25. Recalculate the current network output result: ; S26. Based on the training set, repeat steps S21 to S25 in a loop to construct the initial test model. .
[0036] Furthermore, the determination of the first The input weights and biases of each hidden layer node include: ; ; in, , Represents the hidden layer node The output, express Transpose of; Represents the hidden layer node Candidate input weights, Represents the hidden layer node The candidate bias; λ∈(0,1) is the introduced constraint; Let represent a sequence of non-negative real numbers, where ≤1-λ, the sequence of non-negative real numbers approximates 0 infinitely. ; Will satisfy The candidate node parameters are used as the parameters of the Hth hidden layer node. Indicates the constraint value. Indicates the constraint value The One portion, This indicates that the components of the constraint value are summed; Indicates to Activate, This indicates the hidden layer node. Candidate input weights transpose, Features representing the input The One portion, Represents the current network residual vector The One portion, Represents the current network residual vector The Transpose of each component.
[0037] Furthermore, the active learning operation, utilizing the initial test model, performs value screening on samples not participating in the test, selects and processes applicable value samples, and obtains valuable test data, including: S31. Determine the current initial test model. Is it a single model? S32. If so, then use the following logic to calculate the sample uncertainty for a single model: ; in, This refers to unused test sample data from the collected data. For a single model to correspond to a model, For the single model corresponding to the model For the unused test sample data The predicted probability; to filter out valuable test data for individual models; S33. If not, calculate the contribution value of the sample to different models, and filter out valuable test data for multiple models. The contribution value of the sample to different models is obtained using the following logic: ; in, For the first in multiple models One model, For the first in the multi-model A model For the unused test sample data The predicted probability, The total number of models, Indicates to Find the natural logarithm. Indicates to Perform summation; S34. Save the valuable test data of the single model and the valuable test data of the multiple models to constitute the valuable test data. .
[0038] Furthermore, the construction of the transfer neural network (TNN) specifically includes: First, we define and design the transfer neural network. The transfer neural network (TNN) uses a lightweight convolutional neural network (LightCNN) as its basic architecture. It includes a feature extraction layer, a domain adaptation layer, a traffic scene mapping layer, and a fine-tuning output layer. It is used to transfer the test knowledge of the source traffic scene, i.e., a typical scene that has been fully tested, to the test scene of the target traffic scene.
[0039] Next, the source and target scenarios are defined. The source scenario is a traffic flow scenario with sufficient test data, including low-peak traffic (0-5 people / minute), off-peak traffic (5-20 people / minute), and peak traffic (20-50 people / minute). The target scenario is a specific traffic flow scenario to be tested, including extreme peak traffic (more than 50 people / minute), sudden traffic surge scenario (short-term traffic surge), and time period transition scenario (off-peak to peak / peak to off-peak).
[0040] S41. Feature Extraction: Utilizing the feature extraction layer of a TNN to extract valuable test data from the source scene. Feature extraction is performed to obtain common features of the scene and unique features of the source scene. The feature expression is as follows: ; in, For feature extraction function, These are common feature vectors. This represents the unique feature vector of the source scene. S42. Domain Adaptation: The domain adaptation layer minimizes the distribution difference between the source and target scenes, using the maximum mean difference (MMD) as the domain adaptation loss function. ; in, The number of samples in the source scene. To calibrate a small number of samples for the target scenario, Let i be the feature of the i-th sample in the source scene. For the target scenario Individual sample features; S43. Traffic Scenario Mapping: A mapping relationship between traffic flow parameters and test model parameters is established through the traffic scenario mapping layer. The mapping function is: ; in, For mapping functions, For the target scene's pedestrian traffic parameters, To adapt the model parameters to the target scene; S44. TNN Training and Fine-tuning: TNN parameters are trained using source scene test accuracy and domain adaptation loss as joint optimization objectives. ; in, These are the weighting coefficients. To account for the accuracy loss in the source scene test; the TNN is fine-tuned using a small number of calibrated samples in the target scene to obtain a transfer test model adapted to the target scene. .
[0041] Furthermore, the continuous optimization of the test model and the completion of application software testing include: S51. Collect real-time test data streams under target passenger flow scenarios, including passenger operation data, gate module response data, and flow dynamic monitoring data; S52, Input real-time data stream Make a preliminary value judgment and input the following: By performing scene adaptation and correction, we obtain real-time valuable data judgment results for the target scene: ; in, For the target scene's real-time data stream, Provide valuable real-time test data for the target scenario; S53, based on Synchronously update digital twin test model With transfer neural networks This ensures that the model's accuracy in adapting to the target scene continues to improve; S54, based on the optimized and Complete the functional testing, performance testing, and abnormal scenario testing of the application software under the target scenario, and output the test report.
[0042] In a specific implementation, a cloud testing system for high-speed rail application software based on digital twins, such as... Figure 2 As shown, it includes: Data Acquisition Module: Used to collect historical operating data of high-speed railway station ticket gates, application software testing data, and passenger flow data from multiple scenarios to build a historical dataset; Training module: Used to build an initial digital twin model of the ticket gate based on the Randomly Configured Network (SCN), and to train the SCN using historical datasets to obtain the initial test model; Update module: It is used to actively learn and operate, use the initial test model to screen for value in samples that have not participated in the test, select and process applicable value samples, process them to obtain valuable test data, update the initial test model based on the valuable test data, and obtain an applicable digital twin test model. Knowledge Transfer Module: Used to build a transfer neural network (TNN), train the TNN based on valuable test data from the source pedestrian traffic scenario, and perform test knowledge transfer under different pedestrian traffic scenarios; Test module: Based on the applicable digital twin test model and the trained transfer neural network (TNN), it filters and adapts the real-time test data stream of the target traffic scenario, saves valuable real-time test data, continuously optimizes the test model, and completes application software testing.
[0043] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0044] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A cloud testing method for high-speed rail application software based on digital twins, characterized in that, include: Collect historical operational data of high-speed railway station ticket gates, application software testing data, and passenger flow data from multiple scenarios to construct a historical dataset; An initial digital twin model of the ticket gate was constructed based on a randomly configured network (SCN). The SCN was trained using a historical dataset to obtain an initial test model. By actively learning and operating, using the initial test model, value is screened from samples that have not participated in the test, suitable value samples are selected and processed to obtain valuable test data, and the initial test model is updated based on the valuable test data to obtain a suitable digital twin test model. Construct a transfer neural network (TNN), train the TNN based on valuable test data from the source pedestrian traffic scenario, and perform test knowledge transfer under different pedestrian traffic scenarios; Based on the applicable digital twin testing model and the trained transfer neural network (TNN), the real-time test data stream of the target traffic scenario is filtered and adapted, valuable real-time test data is saved, the testing model is continuously optimized, and the application software testing is completed.
2. The cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, The initial model includes a digital mapping module for a camera, display screen, voice and light alerts, and a turnstile device.
3. The cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, A training set is set up based on the historical dataset. The training set includes feature data and label data corresponding to the feature data. The random configuration network SCN is trained using the training set to obtain an initial test model.
4. The cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, The historical operational data and application software testing-related data include: Data from cameras, displays, voice and light alerts, turnstiles, application software, and pedestrian traffic in various scenarios.
5. A cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, The process of training a randomly configured network (SCN) using historical datasets to obtain an initial test model includes: S21. Given a test objective function, the objective function is: Where x represents the test input data and y represents the expected test output result; let the hidden layer of the randomly configured network SCN be the... Each node has been generated; calculate the current network output. S22. Calculate the current network residual vector; S23. Obtain residual vector parameters based on the current network residual vector processing, and perform a judgment operation based on the residual vector parameters. If the residual vector parameters do not reach the preset error... Or the preset maximum number of nodes has not been reached. When this happens, the number of random configuration networks (SCNs) is increased. The hidden layer node determines the first... Input weights and biases of each hidden layer node; S24. Calculate the output weights of the hidden layer nodes; S25. Calculate the current network output result again based on the output weights; S26. Based on the training set, repeat steps S21 to S25 in a loop to construct the initial test model. .
6. A cloud testing method for high-speed rail application software based on digital twins according to claim 5, characterized in that, The determination of the first The input weights and biases of each hidden layer node include: ； ； in, , Represents the hidden layer node The output, express transpose; Represents the hidden layer node Candidate input weights, Represents the hidden layer node The candidate bias; λ∈(0,1) is the introduced constraint; Represents a sequence of non-negative real numbers; Will satisfy The candidate node parameters are used as the parameters of the Hth hidden layer node. Indicates the constraint value. Indicates the constraint value The One portion, This indicates that the components of the constraint value are summed. Indicates to Activate, This indicates the hidden layer node. Candidate input weights transpose, Features representing the input The One portion, Represents the current network residual vector The One portion, Represents the current network residual vector The Transpose of each component.
7. The cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, The process of active learning, utilizing an initial testing model, involves value screening from samples not yet tested, selecting and processing applicable value samples to obtain valuable test data, including: S31. Determine whether the current initial test model is a single model; S32. If so, calculate the uncertainty of the sample with respect to a single model in order to screen out valuable test data for a single model; S33. If not, calculate the contribution value of the sample to different models and filter out valuable test data for multiple models. S34. Save the valuable test data of the single model and the valuable test data of the multiple models to constitute the valuable test data. .
8. The cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, The construction of the transfer neural network (TNN) includes: S41. Utilize the feature extraction layer of TNN to extract valuable test data from the source scene. Feature extraction is performed to obtain common features of the scene and unique features of the source scene; S42. Minimize the distribution difference between the source scene and the target scene through the domain adaptation layer, and use the maximum mean difference as the domain adaptation loss function. S43. Establish the mapping relationship between pedestrian flow parameters and test model parameters through the traffic scene mapping layer; S44. Using the source scene test accuracy and domain adaptation loss as joint optimization objectives, train the TNN parameters to obtain a transfer test model adapted to the target scene. .
9. A cloud testing method for high-speed rail application software based on digital twins according to claim 1, characterized in that, The continuous optimization of the test model and the completion of application software testing include: S51. Collect real-time test data streams under target passenger flow scenarios, including passenger operation data, gate module response data, and flow dynamic monitoring data; S52. Input the real-time data stream into the applicable digital twin test model. Make an initial value judgment and input it into the transfer test model. Perform scene adaptation corrections to obtain real-time valuable data judgment results for the target scene; S53, based on Synchronous updates and ; S54, based on the optimized and Complete the functional testing, performance testing, and abnormal scenario testing of the application software under the target scenario, and output the test report.
10. A cloud testing system for high-speed rail application software based on digital twins, characterized in that: include: Data Acquisition Module: Used to collect historical operating data of high-speed railway station ticket gates, application software testing data, and passenger flow data from multiple scenarios to build a historical dataset; Training module: Used to build an initial digital twin model of the ticket gate based on the Randomly Configured Network (SCN), and to train the SCN using historical datasets to obtain the initial test model; Update module: It is used to actively learn and operate, use the initial test model to screen for value in samples that have not participated in the test, select and process applicable value samples, process them to obtain valuable test data, update the initial test model based on the valuable test data, and obtain an applicable digital twin test model. Knowledge Transfer Module: Used to build a transfer neural network (TNN), train the TNN based on valuable test data from the source pedestrian traffic scenario, and perform test knowledge transfer under different pedestrian traffic scenarios; Test module: Based on the applicable digital twin test model and the trained transfer neural network (TNN), it filters and adapts the real-time test data stream of the target traffic scenario, saves valuable real-time test data, continuously optimizes the test model, and completes application software testing.