Ergonomic integrated intelligent teaching system for rail transit intelligent operation and maintenance

Abstract

The invention relates to the technical field of intelligent teaching, and provides an engineering integrated intelligent teaching system for rail transit intelligent operation and maintenance. In the system, a teaching data center module aggregates and processes teaching data; the teaching twinborn body module constructs a three-layer twinborn model and generates a multi-dimensional teaching task; the engineering and learning integrated teaching engine module automatically arranges an optimal teaching activity sequence through reinforcement learning according to student ability portraits and teaching tasks; the element universe scene construction module constructs an immersive simulation training scene according to the sequence; the ability evolution evaluation module analyzes trainee training data and generates an ability increment graph; and the teaching feedback optimization module feeds back the atlas to the task generator to drive a new round of teaching. According to the invention, teaching dynamic generation, intelligent arrangement, accurate evaluation and closed-loop optimization in the field of rail transit intelligent operation and maintenance are realized, and the problems of insufficient intelligence, mismatching of teaching strategies, mismatching of engineering and single teaching quality evaluation in rail transit teaching are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent teaching, more particularly, to a work-study integrated intelligent teaching system for rail transit intelligent operation and maintenance. BACKGROUND

[0002] Currently, the rail transit industry is undergoing a comprehensive transformation from traditional operation and maintenance to "intelligent operation and maintenance" characterized by unmanned driving, intelligent monitoring, and predictive maintenance. The industry upgrade has set unprecedentedly high standards for the compound ability of technical and skilled personnel, and there is an urgent need for a vocational education system to cultivate high-quality personnel who are both theoretically knowledgeable and practically skilled. Under this background, the work-study integrated teaching mode has become the core path for vocational colleges to cultivate such personnel. This mode emphasizes the deep integration of real work processes and theoretical learning, but in the teaching practice of this specific high-end field of rail transit intelligent operation and maintenance, it faces a series of technical bottlenecks that need to be solved.

[0003] The traditional teaching mode has the following deficiencies in this field: First, the teaching scene is seriously disconnected from real working conditions. The core equipment of rail transit, such as the train traction system and signal interlocking equipment, is costly, complex in structure, and operates in a high-risk environment, making it difficult to carry out deep skill training with high failure rates and high risks in physical training. Existing virtual simulation software is mostly limited to preset and isolated operation process drills, and cannot dynamically simulate the evolution of equipment hidden faults, nor can it realistically reproduce multi-system coupled faults and the resulting chain of operational incidents, resulting in students mastering individual operations but lacking systematic thinking and emergency decision-making capabilities to deal with complex real working conditions. Second, teaching design is highly dependent on subjective experience and lacks individualization and adaptability. Teachers mainly rely on personal experience and general outlines for teaching design, making it difficult to accurately grasp the significant differences in knowledge base, skill gaps, and learning pace among different students, and unable to achieve personalized teaching for each individual. There is often a mismatch between teaching objectives, teaching activities, and the actual ability level of students, leading to low teaching efficiency and making it difficult for "work" and "learning" to form an effective closed loop. Third, the teaching evaluation dimension is single and the feedback is lagging. The existing evaluation system mainly focuses on whether the final operation result is correct or not, or the theory test score, which is a terminal evaluation. It severely lacks multi-dimensional and quantitative tracking of students' process capabilities, collaboration and communication skills, and innovative thinking in the complete work process of "task understanding, plan development, decision implementation, check control, evaluation feedback, and innovation transfer". The evaluation results are usually lagging, and cannot provide immediate and accurate data support for real-time adjustment and optimization of the teaching process.

[0004] In the prior art, such as the patent "Rail transit intelligent operation and maintenance training system based on cloud computing" with publication number CN120690080A, a method is provided for optimizing training project scheduling and cloud computing resource allocation by quantitatively analyzing the performance of students. However, this system is essentially a scheduling platform that focuses on resource management and load optimization. Its core lies in prioritizing and allocating resources to an existing, static training project library, and it does not address fundamental teaching issues such as the dynamic generation of teaching scenarios, the intelligent arrangement of personalized teaching strategies, and the multi-dimensional process evaluation of complex professional capabilities. Other related technical solutions also focus on single aspects such as online learning management, fixed process simulation, or simple behavior analysis, and fail to build a complete intelligent teaching system that deeply integrates domain knowledge, dynamic twin environments, intelligent decision engines, and closed-loop evaluation feedback.

[0005] Therefore, there is an urgent need in the field for an innovative technical solution to systematically address the above-mentioned problems and achieve the leap from concept to high-quality implementation of integrated work-study teaching in the field of rail transit intelligent operation and maintenance. SUMMARY

[0006] To solve the technical problems mentioned in the above background art, the present application provides an integrated work-study intelligent teaching system for rail transit intelligent operation and maintenance, which adopts the following technical solutions: An integrated work-study intelligent teaching system for rail transit intelligent operation and maintenance, comprising: A teaching data hub module for aggregating and processing heterogeneous data related to rail transit operation and maintenance teaching, including static standard data, dynamic teaching process data, and external teaching resource data; A teaching twin module for dynamically constructing a three-layer twin model containing device mechanisms, system operations, and teaching tasks based on the heterogeneous data, and generating a set of multi-dimensional composite teaching tasks; An integrated work-study teaching engine module with an adaptive task generator based on a reinforcement learning hybrid algorithm, for automatically arranging an optimal teaching activity sequence based on the student's ability profile and the set of multi-dimensional composite teaching tasks; A meta-universe scene construction module for calling corresponding virtual materials based on the optimal teaching activity sequence, and constructing a simulation scene for interactive virtual operation training for students; A capability evolution evaluation module for collecting training data of students in the simulation scene and comparing it with the standard job model in the teaching twin module to generate a capability increment map; A teaching feedback optimization module for feeding the capability increment map back to the adaptive task generator to drive the generation of a new round of optimal teaching activity sequence.

[0007] Further, the teaching data center module specifically comprises: A static standard library for storing and associating rail transit operation and maintenance capability standards, course standard atlas and industry specifications in the four-dimensional integration of "post-course-race-certificate", and the course standard atlas establishes a semantic mapping relationship between course knowledge points, skill points and entries in the capability standard through ontology modeling; A dynamic process library for real-time collection and structured storage of multi-modal data in the whole teaching process, the multi-modal data including operation sequence logs of students in virtual operation training, dialogue texts generated by interaction with intelligent teaching assistants, role-playing and communication records in collaborative tasks, and key node annotation data of students' operation process by teachers based on standard operation models, the standard operation model being a pre-set operation specification model in the teaching twin body module; An external resource library for indexing and associating a parameterized rail transit equipment three-dimensional model library, a fault case knowledge graph, maintenance procedure video clips and safety warning education materials.

[0008] Further, the three-layer twin model in the teaching twin body module specifically comprises: A device-level mechanism twin layer for fault simulation and state prediction based on physical simulation models and historical operation data of key rail transit equipment, generating a set of potential hidden fault modes; A system-level business twin layer connected with the device-level mechanism twin layer, for associating and deducing the set of hidden fault modes with device topology relationship and operation scheduling rules according to the operation regulation knowledge graph, generating a virtual business event chain affecting operation safety; A teaching-level task twin layer connected with the system-level business twin layer, for receiving the virtual business event chain and combining the teaching target in the static standard library, and through task decomposition and reorganization algorithm, the event chain is decomposed into a multi-dimensional composite teaching task set including fault disposal, scheduling coordination and safety briefing.

[0009] Further, the teaching-engineering integrated teaching engine module specifically comprises: A student ability portrait generator for outputting a "safety-efficiency-cost" three-dimensional ability vector based on student historical training data and real-time eye movement, operation sequence, voice instruction multi-modal information through graph neural network; A course difficulty adjuster for monitoring the variance of the reward function, and if the variance exceeds the threshold, the difficulty of subsequent tasks is dynamically adjusted according to the "recent development zone" principle; A knowledge graph gap diagnosis interface for comparing the four-dimensional knowledge graph of "post-course-race-certificate" with the three-dimensional ability vector of the student to generate a gap vector; An adaptive task generator, which is built-in a reinforcement learning hybrid algorithm unit, is used to splice the three-dimensional capability vector and the embedding vector of the teaching task set as a state, take the task difficulty, task type, and resource combination as an action space, and take the capability increment expectation as a reward function, and output a preliminary teaching activity sequence through an Actor-Critic architecture.

[0010] Further, after the adaptive task generator outputs the preliminary teaching activity sequence, the adaptive task generator further includes content rearrangement and task replacement on the preliminary teaching activity sequence based on the gap vector, generates a revised teaching activity sequence, and outputs an optimal teaching activity sequence based on the task difficulty adjustment requirement fed back by the course difficulty adjuster.

[0011] Further, the metaverse scene construction module is specifically used for: receiving the optimal teaching activity sequence, and analyzing the requirements for virtual environments, device states, fault phenomena, and interactive objects in the optimal teaching activity sequence; According to the analysis result, the matching parameterized three-dimensional model, material map, animation script, and physical engine configuration parameter are intelligently retrieved and called from the external resource library; Based on a real-time rendering game engine, the called materials are real-time instantiated, spatially laid out, and logically bound according to the teaching logic, to generate an immersive simulation scene that is synchronized with the optimal teaching activity sequence and is interactive; The immersive simulation scene supports fine operation training of the trainee in a first-person perspective through a personal terminal, and supports global situation deduction and role-playing in a third-person perspective in a collaborative sand table.

[0012] Further, the capability evolution evaluation module specifically includes: A data acquisition unit is used to extract trajectory data, step completion time sequence, tool selection sequence, and communication content with other trainees from the background log of virtual operation training; A multi-dimensional comparative analysis unit is used to compare the trajectory data, time sequence, and tool selection sequence with the preset optimal path, standard working hours, and standard tool set in the standard operation model, respectively, to calculate operation specification score, efficiency score, and decision accuracy score; the communication content is subjected to natural language processing to analyze its collaboration intention and problem solving logic, to generate collaboration communication and innovative thinking scores; The atlas generation unit is configured to generate a visual ability increment atlas in the form of a radar chart and an ability node growth chart according to the collaborative communication and innovative thinking scores of the multidimensional comparative analysis unit and based on the item response theory to calculate the ability level changes of the students in each ability dimension relative to the previous assessment.

[0013] Further, the teaching feedback optimization module specifically includes: The strategy adjustment unit is configured to analyze the ability increment atlas, and identify common weak abilities of the student group and specific ability short boards of the individuals. The resource rematching unit is configured to rematch corresponding resource lists from the external resource library and the static standard library of the teaching data platform module according to the identified common weak ability dimensions of the student group and the specific ability short boards of the individuals, wherein the resource lists include cases, micro-lessons, simulation faults or extended reading materials. The task regeneration instruction unit is configured to package and send the updated common weak ability dimensions of the student group and the specific ability short boards of the individuals and the rematched resource lists to the adaptive task generator of the work-integrated teaching engine module, so as to trigger the adaptive task generator to generate a new round of optimal teaching activity sequence for the common weak abilities and the specific ability short boards in a new round of strategy iteration, wherein the new round of optimal teaching activity sequence includes reinforcement training tasks and personalized learning path suggestions.

[0014] Further, the system further includes a cross-platform collaboration and resource sharing module, which is in communication connection with the teaching data platform module and the work-integrated teaching engine module, and specifically includes: The collaborative interaction unit is configured to realize voice / video interaction, operation picture synchronization and text collaborative annotation between the students and the instructors or between the students by using the WebRTC real-time communication protocol, and support multi-role permission hierarchical management. The learning data synchronization unit is configured to synchronize the learning progress, training records and ability evaluation results of the students to the cloud in real time based on the distributed database of the cloud-edge collaboration architecture, so as to ensure data consistency after multi-terminal login. The intelligent resource pushing unit is configured to filter and match customized learning resources from the external resource library of the teaching data platform based on the collaborative filtering algorithm in combination with the weak links in the ability portrait of the students, and push the customized learning resources to the student terminals in real time through a message pushing interface.

[0015] Further, all the modules of the work-integrated intelligent teaching system for rail transit intelligent operation are deployed in the same cloud-edge collaboration architecture, and the cloud-edge collaboration architecture includes: a central cloud for storing historical operation and maintenance big data, the heterogeneous data and cross-campus teaching resources; an edge cloud deployed in a practical training base for carrying out real-time calculation and low-latency rendering of teaching twin bodies; a terminal edge node embedded with a personal immersive operation terminal and AR glasses for locally completing pose tracking and gesture recognition; a unified time sequence database adopting a time sequence structured multi-tag (TSMT) data model to tag and store real operation and maintenance events, teaching task events and student behavior events.

[0016] The present application has the following beneficial effects: the present application deeply integrates field knowledge graph, dynamic digital twin, reinforcement learning decision engine and immersive virtual simulation technology, and constructs a closed-loop intelligent teaching ecology capable of self iteration. The system breaks through the transformation of the teaching scene from preset static cases to dynamic generation based on real data and mechanism models, can simulate complex implicit fault chains and system-level business events in intelligent operation and maintenance of rail transit, and provides a high-fidelity training environment with infinite approximation to real work pressure and complexity for students. Meanwhile, the system intelligently arranges and generates personalized teaching task sequences deeply integrating technical operation and process cooperation according to multi-dimensional capability portraits and real-time twin events of students through reinforcement learning algorithm, realizes a fundamental leap from "resource scheduling" to "strategy generation", and achieves precise "one person one strategy" adaptive teaching. In the evaluation aspect, the system collects multi-modal behavior data of students in the immersive environment throughout the process, and performs multi-dimensional and process deep analysis based on the item response theory to generate a capability increment graph capable of quantitatively reflecting the growth trajectory of students in the core professional capability dimension, and realizes the evolution from single result evaluation to value-added and developmental evaluation. Finally, the above evaluation results are fed back to the task generation engine and twin model of the system in real time, driving the continuous dynamic optimization of teaching strategies and training content, forming a strong closed loop with self-learning and self-evolution capabilities, and fundamentally solving the systematic problems such as distorted teaching situation, insufficient individualization, single and lagging evaluation method, and disconnection between theory and practice in traditional vocational education, and providing a quantifiable, traceable and optimized solution for cultivating compound high-end technical and skilled talents meeting the needs of industrial upgrading. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0018] Figure 1A work-study integrated intelligent teaching system block diagram for rail transit intelligent operation is provided for an embodiment of the present application.

[0019] Figure 2 A student virtual operation score generated by a graph generation unit in a capability evolution evaluation module and a radar chart evaluation schematic diagram of each capability dimension are provided for an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of a work-study integrated intelligent teaching system for rail transit intelligent operation according to the present application are described in detail as follows in combination with the preferred embodiments and the drawings. Different "one embodiment" or "another embodiment" in the following description do not necessarily refer to the same embodiment. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs.

[0021] EMBODIMENT The specific scheme of the work-study integrated intelligent teaching system for rail transit intelligent operation provided by the present application is specifically described below in combination with the drawings.

[0022] The specific scene targeted by the embodiment of the present application is: for the core dilemma faced by the field of rail transit intelligent operation in implementing work-study integrated advanced skill training, that is, how to reproduce the complex fault handling scene in real operation with high fidelity in a safe and controllable environment, and realize the process training and precise evaluation of the complex ability of students such as systematic diagnosis, emergency decision-making and collaborative work.

[0023] Please refer to Figure 1 which shows a work-study integrated intelligent teaching system block diagram for rail transit intelligent operation provided by an embodiment of the present application, which includes: A teaching data center module 101, a teaching twin module 102, a work-study integrated teaching engine module 103, a meta-universe scene construction module 104, a capability evolution evaluation module 105, and a teaching feedback optimization module 106.

[0024] The teaching data middle station module 101 is used for aggregating and processing heterogeneous data of rail transit operation and maintenance teaching, and the heterogeneous data includes static standard data, dynamic teaching process data and external teaching resource data. The teaching data middle station module specifically includes: a static standard library, used for storing and associating rail transit operation and maintenance capability standards, course standard maps and industry specifications in four-dimensional integration of “post course competition certificate”, the course standard map establishes a semantic mapping relationship between course knowledge points, skill points and entries in the capability standard through ontology modeling; a dynamic process library, used for real-time collection and structured storage of multi-modal data in the whole teaching process, the multi-modal data includes operation sequence logs of students in virtual operation training, dialogue texts generated by interaction with an intelligent teaching assistant, role playing and communication records in collaborative tasks, and key node annotation data of the operation process of students by teachers based on a standard operation model, and the standard operation model is a preset operation specification model of the teaching twin body module; an external resource library, used for indexing and associating a parameterized rail transit equipment three-dimensional model library, a fault case knowledge map, a maintenance procedure video segment and safety warning education materials.

[0025] To realize the function of the teaching data middle station module 101, in the specific implementation, the module constructs a unified data governance framework integrating the static standard library, the dynamic process library and the external resource library. The static standard library converts the capability standard entries corresponding to “post course competition certificate”, course knowledge points and industry specification provisions into a knowledge graph with semantic association through ontology modeling technology. For example, for the capability standard of “train traction system fault troubleshooting”, the graph establishes a mapping relationship between it and the “traction inverter working principle” knowledge point in the “Urban Rail Transit Vehicle Maintenance” course, the safety test regulations in the local standard “DB11 / T2390-2025 Urban Rail Transit Urban Rapid Transit Vehicle Maintenance Technical Specification”, and the related competition topics of “Rail Transit Intelligent Operation and Maintenance Skills Competition”, and forms a structured capability requirement description.

[0026] The dynamic process library captures and stores multi-modal data streams generated in the teaching process in real time through data collection interfaces deployed in the virtual training environment. When the student operates in the virtual train maintenance scene, the system not only records the operation step sequence and tool use sequence, but also synchronously collects the voice dialogue text with the intelligent teaching assistant, the instruction text sent through the communication interface in the collaborative task, and the standard compliance annotation added by the teacher through the management terminal in the playback video to the key operation node. These data are time-stamped and stored in association with specific teaching tasks, equipment twin body instances and student identities.

[0027] The external resource library manages parameterized three-dimensional models and multimedia materials. For example, a three-dimensional model of a metro bogie is associated with the geometric parameters, material properties, and physical simulation parameters of its detachable components. An operation video of "abnormal wear detection of pantograph carbon slide plate" is cut into step segments and linked to the corresponding nodes in the fault case knowledge graph.

[0028] The core processing logic of the teaching data center module is to associate and fuse the above three types of heterogeneous data. When the engineering-integrated teaching engine module requests to generate a teaching task about "intermittent fault of train network control system", the data center will determine the knowledge points such as "fault code analysis" and "network topology analysis" that need to be covered according to the ability map in the static standard library; at the same time, it queries the historical operation records of the target student in the dynamic process library to identify the weak links in the "logical judgment" dimension; and then retrieves matching fault case three-dimensional models, network topology graph materials and related technical manual fragments from the external resource library. This cross-library semantic retrieval and association matching enables the subsequent generated teaching tasks to accurately interface with the job capability requirements, adapt to the individual learning situation of the students, and provide high-fidelity practical training resources.

[0029] In this embodiment, the static standard data ensures the normativity of the teaching objectives, the dynamic process data realizes the objectivity of the learning situation analysis, and the external resource data guarantees the richness of the teaching scene. These three types of data support each other in data processing logic: the standard data defines the dimensions and targets for analyzing the dynamic data, the analysis results of the dynamic data drive the personalized selection and invocation of the resource data, and the associability of the resource data enables the instructions generated based on the standard and dynamic data to be presented concretely. This mutual interaction between data supports the dynamic construction of complex teaching scenarios by the subsequent teaching twin body module and the intelligent generation of individualized teaching strategies by the engineering-integrated teaching engine module.

[0030] Based on the heterogeneous data aggregated and structured by the teaching data center module, the intelligent evolution of the teaching process has a solid data foundation. However, simple data invocation cannot automatically generate dynamic training scenarios that match the complexity of real operation and maintenance and adapt to the individual ability development of students, so the system introduces the teaching twin body module 102. As the core engine connecting the bottom-layer data and the top-layer teaching application, this module is responsible for dynamically fusing and constructing an interactive, deductive, and trainable virtual teaching environment from the standardized knowledge, real-time process information, and parameterized resources from the data center, according to the physical laws of equipment, the operation logic of systems, and teaching objectives.

[0031] The teaching twin module 102 is used for dynamically constructing a three-layer twin model including device mechanism, system business and teaching task based on the heterogeneous data, and generating a multi-dimensional composite teaching task set. The three-layer twin model in the teaching twin module specifically includes: a device-level mechanism twin layer, which is used for simulating faults and predicting states based on a physical simulation model of rail transit key equipment and historical operation data, and generating a set of potential hidden fault modes; a system-level business twin layer connected with the device-level mechanism twin layer, which is used for associating and deducing the set of hidden fault modes with device topology relationship and operation scheduling rules according to an operation and maintenance regulation knowledge graph, and generating a virtual business event chain affecting operation safety; and a teaching-level task twin layer connected with the system-level business twin layer, which is used for receiving the virtual business event chain, combining a teaching target in the static standard library, and decomposing and recombining the event chain into a multi-dimensional composite teaching task set including fault disposal, scheduling coordination and safety briefing.

[0032] Specifically, the teaching twin module 102 realizes the conversion from data to dynamic teaching situation by constructing a three-layer coupled twin model of device mechanism, system business and teaching task. The device-level mechanism twin layer drives a physical simulation model of key equipment such as a traction motor and a signal relay to run according to historical and real-time monitoring data from a dynamic process library. The model can simulate the state evolution of the equipment under specific working conditions and potential degradation conditions. When the simulation parameters and the measured data trend continuously deviate, a hidden fault mode such as “the vibration amplitude of the traction motor bearing increases exponentially but does not exceed the threshold” is marked and generated. This fault mode is transmitted to the system-level business twin layer as a basic event.

[0033] The system-level business twin layer is embedded with an operation and maintenance regulation knowledge graph, which defines the topological connection relationship between devices and the operation scheduling logic. After receiving the set of hidden fault modes from the device layer, the layer performs association and deduction according to the graph. For example, a hidden fault of “abnormal current waveform of a turnout switch machine” can deduce a series of virtual business event chains such as “failure of route arrangement”, “inaccurate train station marking” and “decline of interval passing capacity” according to the defined interlocking relationship and train operation rules in the knowledge graph. This deduction process simulates the cascading amplification effect of a single device anomaly in the operation system.

[0034] The teaching level task twin layer receives the above virtual service event chain and decomposes the task in combination with the teaching objectives corresponding to "post-course-race-certificate" from the static standard library. This layer converts the abstract event chain into a teaching task set containing specific actions, roles and resource requirements through task decomposition and reorganization algorithms. For the aforementioned "turnout abnormality triggering route failure" event chain, the algorithm will automatically generate a multi-dimensional composite teaching task set according to the ability requirements in the teaching objective library for "fault emergency disposal", "train organization adjustment" and "safety risk management". The task set may include: a turnout site electrical inspection and data measurement task performed by one student, a route rearrangement and train adjustment task performed by another student on a simulation dispatch terminal, and a fault information reporting and safety protection setting task completed by both parties in coordination. The task set clearly defines the execution order, logical dependency relationship, required three-dimensional model and procedure document of each sub-task, as well as the corresponding ability evaluation dimensions.

[0035] The running logic of the three-layer twin model is tightly coupled and the data flow is progressively passed down layer by layer. The physical simulation of the device mechanism layer provides the system service layer with original input conforming to engineering laws, avoiding the defect of service deduction deviating from physical reality; the rule-based deduction of the system service layer places the device fault under real operational constraints, generating a complex situation with real post authenticity; the algorithmic decomposition of the teaching task layer converts the complex situation into specific teaching activities that can be trained and evaluated, and directly relates to the curriculum standards. In this way, the technical problems of single and static virtual training scenarios, disconnection with real operational dynamic complexity, and insufficient matching of teaching tasks and comprehensive post requirements are solved. By simulating the dynamic process from the implicit occurrence of a fault to the system impact, and decomposing it into a composite task according to the teaching law, the student can conduct systematic ability training covering the complete work process and involving multi-post collaboration in a highly realistic virtual environment, thereby effectively improving the depth and effectiveness of integrated teaching.

[0036] The multi-dimensional composite teaching task set generated by the teaching twin module 102 provides a rich and structured content basis for carrying out integrated teaching. However, effectively converting these tasks into teaching activity sequences that adapt to the individual ability level of a specific student and achieve optimal training results within limited teaching resources still requires a core engine with intelligent decision-making capabilities. Therefore, the system introduces an integrated teaching engine module 103. This module takes the task set output by the teaching twin module as "raw materials" and the student's ability profile as "individualized blueprint", and performs dynamic planning and optimization through the built-in reinforcement learning hybrid algorithm. Its function is to combine static task lists with dynamic student status to intelligently arrange a training path that is easy to difficult, sequential and highly adapted to the student's real-time ability shortfalls, thereby converting the preset standardized teaching tasks into dynamic and individualized teaching activity execution plans.

[0037] The engineering-integrated teaching engine module 103 is internally provided with an adaptive task generator based on a reinforcement learning hybrid algorithm, which is used to automatically arrange an optimal teaching activity sequence according to the learner ability profile and the multi-dimensional composite teaching task set. Specifically, the engineering-integrated teaching engine module includes: a learner ability profile generator, which is used to output a “safety-efficiency-cost” three-dimensional ability vector based on historical training data of the learner and real-time multi-modal information such as eye movement, operation sequence and voice instruction through a graph neural network; a course difficulty adjuster, which is used to monitor the variance of the reward function, and if the variance exceeds the threshold, the difficulty of the subsequent task is dynamically adjusted according to the “recent development zone” principle; a knowledge graph gap diagnosis interface, which is used to compare the “post-course-competition-certificate” four-dimensional knowledge graph with the three-dimensional ability vector of the learner to generate a gap vector; and an adaptive task generator internally provided with a reinforcement learning hybrid algorithm unit, which is used to splice the three-dimensional ability vector and the embedded vector of the teaching task set as a state, take the task difficulty, task type and resource combination as an action space, and take the ability increment expectation as a reward function, and output a preliminary teaching activity sequence through an Actor-Critic architecture. After the adaptive task generator outputs the preliminary teaching activity sequence, it further includes content rearrangement and task replacement of the preliminary teaching activity sequence based on the gap vector to generate a revised teaching activity sequence, and outputs an optimal teaching activity sequence based on the task difficulty adjustment requirement fed back by the course difficulty adjuster.

[0038] More specifically, the engineering-integrated teaching engine module 103 realizes the complete intelligent decision-making process from learner state perception to personalized teaching sequence generation through the progressive cooperation of its internal components. The learner ability profile generator continuously receives and processes multi-modal data streams from the training scene, such as the tool operation sequence of the learner in the virtual train brake system maintenance task, the time consumption of each step, the discussion content of the fault reason with the voice assistant, and the gaze duration on the key instrument panel recorded by the VR glasses. The graph neural network is trained to model these time series and associated data, outputting a quantitative three-dimensional ability vector, such as [safety: 0.82, efficiency: 0.58, cost: 0.70], which objectively depicts the learner’s immediate state in terms of procedure compliance, operation proficiency and resource utilization.

[0039] The knowledge graph gap diagnosis interface compares this three-dimensional vector with the “post-course-competition-certificate” four-dimensional knowledge graph stored in the teaching data platform. The graph has preset target values for each ability dimension and their associated knowledge and skill nodes. By calculating the difference, the interface generates a structured gap vector, such as explicitly identifying that the learner has a significant deficiency in the “brake cylinder piston stroke adjustment” skill point with a quantitative gap value of 0.35, which indicates the specific learning goal to be strengthened.

[0040] The reinforcement learning hybrid algorithm unit built into the adaptive task generator initiates the first round of decision-making, which splices the three-dimensional ability vector with the embedding vector extracted from the teaching task set to form the initial state input. The algorithm defines the action space in terms of task difficulty level, task type (such as diagnosis, operation, collaboration), and resource combination (such as whether to enable AR guidance), and constructs a reward function aimed at maximizing the predicted ability increment. Through iterative exploration of the Actor-Critic architecture, the algorithm outputs a preliminary teaching activity sequence that meets the basic teaching constraints, such as [Task A: Cognitive Learning of Brake Principle, Task B: Basic Brake Pad Replacement Drill].

[0041] Subsequently, the generator initiates the optimization correction phase. Based on the gap vector provided by the knowledge graph gap diagnosis interface, the system rearranges the content and replaces tasks in the preliminary sequence. For the "brake cylinder adjustment" gap mentioned above, the algorithm may replace the "basic brake pad replacement drill" task in the sequence with a more suitable "brake cylinder stroke measurement and adjustment" specialized training task, generating a revised teaching activity sequence. In this process, the course difficulty adjuster runs in parallel, monitoring the variance of the reward function during reinforcement learning training. If the variance is detected to consistently exceed the pre-set threshold, indicating a systematic deviation in the adaptation of the current task difficulty to the student group, a global difficulty adjustment parameter is generated based on the "zone of proximal development" principle, such as "uniformly reducing the baseline difficulty coefficient of all subsequent practical tasks by 10%". The adaptive task generator receives this adjustment requirement and performs final difficulty calibration on the revised sequence, outputting the optimal teaching activity sequence.

[0042] In this module, the student ability profile generator and the knowledge graph gap diagnosis interface provide quantitative and accurate student state and learning goal data, the reinforcement learning algorithm unit provides the core ability to optimize the sequence under complex constraints, and the course difficulty adjuster ensures the macro adaptability and teaching reasonableness of the optimization process. These technical features functionally support and interact with each other, collectively forming a technical means that can dynamically respond to individual learning differences and intelligently generate personalized teaching paths. This solution effectively addresses the technical problem of teaching design relying on experience and being difficult to dynamically adapt to individual student ability gaps and learning pace in the background technology. Through data-driven intelligent decision-making, it realizes the transition from general teaching arrangements to "one strategy for one person" precise teaching, significantly improving the relevance and efficiency of rail transit intelligent operation skill training.

[0043] The optimal teaching activity sequence output by the engineering-education integrated teaching engine module 103 clearly defines "what to train" and "in what logical order to train". However, to convert these abstract teaching instructions into specific training environments that are perceptible, interactive, and operable for students, an execution interface that can achieve virtual-real integration and scenario construction is needed. Therefore, the system introduces the meta-universe scenario construction module 104.

[0044] The metaverse scene construction module 104 is configured to call corresponding virtual materials according to the optimal teaching activity sequence, and construct a simulation scene for interactive virtual operation training of the trainee. Specifically, the metaverse scene construction module 104 is configured to receive the optimal teaching activity sequence, analyze the requirements for the virtual environment, device state, fault phenomenon and interactive object in the sequence, intelligently retrieve and call the matched parameterized three-dimensional model, material map, animation script and physical engine configuration parameter from the external resource library according to the analysis result, based on the real-time rendering game engine, real-time instantiate, spatial layout and logic binding of the called materials according to the teaching logic, and generate an immersive simulation scene that is synchronized with the optimal teaching activity sequence and is interactive. The immersive simulation scene supports the trainee to perform fine operation training through the first-person perspective of the personal terminal, and supports global situation deduction and role playing in the third-person perspective of the collaborative sand table.

[0045] Specifically, the metaverse scene construction module 104 converts the abstract teaching logic into a concrete and interactive immersive training environment according to the dynamic instructions of the optimal teaching activity sequence. After receiving the optimal teaching activity sequence data from the engineering-teaching integrated teaching engine module, the module first analyzes the specific requirements of each teaching step in the sequence for the virtual environment constituent elements. These requirements include the target training site type, the involved rail transit device object and its initial state, the fault phenomenon characteristics to be simulated, and the interactive object or role expected to appear in the step. For example, for a teaching task step of “train pantograph carbon slide plate abnormal wear detection and replacement”, the analysis result will explicitly require a virtual roof work environment containing the overhead line system and the pantograph, a pantograph three-dimensional model with the state parameter displayed as “carbon slide plate thickness below threshold and existence of eccentric wear”, a fault phenomenon particle effect simulating “spark” or “abnormal noise”, and a virtual tool kit and safety protection equipment available for the trainee to use.

[0046] Based on the requirement description generated by the analysis, the module intelligently retrieves and matches from the external resource library. The parameterized three-dimensional models in the resource library are associated with metadata tags such as device name, movable parts, physical properties and configurable state variables; the fault case knowledge graph links the animation scripts, sound files and physical engine parameters required to be loaded to represent specific faults. The system compares the requirement keywords such as “pantograph” and “carbon slide plate wear” with the resource metadata through a semantic matching algorithm, retrieves the basic model with high matching degree, the corresponding wear material map, the disassembly and assembly animation sequence, and the physical parameter set simulating the dynamic contact between the pantograph and the overhead line system.

[0047] The retrieved materials are input into a real-time rendering game engine for scene synthesis. The engine instantiates and spatially lays out the materials in real time according to the teaching logic, for example, accurately positioning the pantograph model at the specified coordinates on the virtual train roof, placing the tool model in the tool car for easy access, and setting a series of logical trigger points in the scene according to the pre-set work flow in the teaching sequence. These trigger points are bound to the student's operation actions on the personal terminal (such as view movement, virtual handle grabbing), and when the student's virtual line of sight focuses on a specific part of the carbon slide plate or performs a disassembly action, the scene will trigger interactive responses such as displaying accurate measurement readings, playing disassembly animations, or providing force feedback. The scene built by this module has multi-modal interaction capabilities, supporting students to perform fine operations in first-person perspective through VR headsets and handsets, such as using virtual measuring instruments for size measurement and using wrenches for bolt disassembly. At the same time, the scene also supports a collaborative sand table mode, providing a third-person global overhead view for multiple students to perform task division and collaboration, train dispatching simulation, or safety monitoring role-playing. The scene state is strictly synchronized with the teaching sequence, and the operation results of the previous step are used as the initial conditions for the next step.

[0048] By retrieving, assembling, and binding parameterized resources in real time according to dynamic teaching instructions, this module enables on-demand generation and high configurability of teaching scenes, avoiding the limitations of single scenes and fixed processes in traditional simulation software. The close binding of teaching logic, virtual object state, and interactive behavior enables abstract knowledge and skill requirements to be directly perceived and trained in a concrete operating environment. The high-fidelity, highly interactive, and immersive scenes generated by this module not only provide students with a "learning by doing" practice platform, but more importantly, provide a native environment for the subsequent ability evolution evaluation module to collect students' full-process, multi-dimensional behavior data, making it possible to quantify the analysis of operation specifications, decision logic, and collaboration processes.

[0049] The immersive simulation scene generated by the metaverse scene construction module 104 provides students with a concrete operating space to perform "optimal teaching activity sequences" and continuously generates a large amount of native, real-time behavior data during the process. However, these data are only records of the training process and need to be systematically processed and analyzed to be transformed into scientific evaluations of students' ability growth status and training effectiveness. Therefore, the system introduces the ability evolution evaluation module 105. This module, as an "analysis engine" of training effectiveness, receives and processes data streams from the virtual scene background in real time, and according to the standard operation model and multi-dimensional evaluation system pre-set in the teaching twin module, quantitatively analyzes and compares students' operation processes and interactive performances in the simulation environment. Its role is to aggregate and sublimate the discrete, time-series behavior traces of students in virtual training into structured, measurable ability evolution maps, thereby providing objective and precise data feedback for the closed-loop optimization of teaching strategies.

[0050] The capability evolution evaluation module 105 is configured to collect training data of the trainee in the simulation scene and compare the training data with a standard operation model in the teaching twin module to generate a capability increment atlas. The capability evolution evaluation module 105 specifically includes a data collection unit configured to extract trajectory data of the trainee's operation, a time sequence of completion of each teaching sub-step, a tool selection sequence, and communication content with other trainees from a background log of virtual operation training; a multi-dimensional comparison and analysis unit configured to compare the trajectory data, the time sequence, and the tool selection sequence with a preset optimal path, standard working hours, and standard tool set in the standard operation model, respectively, to calculate an operation specification score, an efficiency score, and a decision accuracy score; perform natural language processing on the communication content to analyze a collaboration intention and a problem solving logic, and generate a collaboration communication and innovative thinking score; and an atlas generation unit configured to generate a visual capability increment atlas in the form of a radar chart and a capability node growth chart based on the collaboration communication and innovative thinking score of the multi-dimensional comparison and analysis unit and a change in a capability level of the trainee in each capability dimension relative to a previous evaluation based on a project response theory, wherein the trainee in each capability dimension includes six dimensions of "task understanding, plan making, decision implementation, check control, evaluation feedback, and transfer innovation".

[0051] Specifically, the capability evolution evaluation module 105 realizes quantitative evaluation and growth tracking of the comprehensive professional capability of the trainee by collecting and analyzing behavior data of the trainee in the virtual scene. The data collection unit continuously extracts a raw data stream generated by the trainee in the training process from a background running log of the metaverse scene construction module. The data includes a moving and operation trajectory coordinate sequence of the virtual avatar of the trainee in the three-dimensional space, a time stamp sequence of time consumed for completing each teaching sub-step, a sequence record of selecting and using different tools from a virtual tool box, and content text of communication with other trainees or intelligent roles through a text or voice channel in a collaborative task.

[0052] After receiving the above data, the multi-dimensional comparative analysis unit compares it with the pre-defined standard operation model in the teaching twin model module. The standard operation model includes the optimal operation path for a specific teaching task, the standard working time range of each step, the recommended tool selection sequence, and the safety specification points. For example, for the task of “replacing train brake pads”, the unit calculates the spatial similarity between the actual movement trajectory of the trainee and the optimal path (such as the shortest safe route from the tool cart to the brake cylinder) to obtain the operation specification score; compares the actual completion time with the standard working time to calculate the efficiency score; matches the actual tool sequence selected by the trainee (such as whether to use a torque wrench to loosen the bolts first, and then use a special puller) with the standard tool set and usage sequence to obtain the decision accuracy score. At the same time, the unit processes the collected communication content through natural language processing, analyzes the intent classification, information integrity, term accuracy of the sentence, and whether the proposed solution exceeds the regular process, to generate scores in the two dimensions of collaborative communication and innovative thinking. These processes convert discrete behavior records into initial scores in the six pre-set ability dimensions of “task understanding, planning, decision implementation, inspection control, evaluation feedback, and innovation transfer”.

[0053] The atlas generation unit is responsible for integrating the dimension scores and calculating the ability increment. The unit applies the item response theory model, which uses the dimension scores obtained in the current evaluation period and the trainee's historical evaluation data as inputs. The item response theory model can estimate the potential ability value of the trainee in each ability dimension, which eliminates the impact of specific task difficulty differences on the score. By comparing the ability value calculated in this evaluation with the ability value in the last evaluation, the unit calculates the “ability level change” in each dimension, i.e. the ability increment. For example, if a trainee's ability value in the “decision implementation” dimension increases from 60% to 80%, the increment is recorded as 20%. Finally, the system visualizes the trainee's current absolute ability value in the six dimensions in the form of a radar chart, and intuitively displays the increment change in each dimension compared to the last evaluation in the form of a node growth chart, which together constitute a complete trainee ability increment atlas. As shown in the Figure 2 The virtual operation score of the trainee and the evaluation radar chart of each ability dimension generated by the atlas generation unit in the ability evolution evaluation module are shown in the figure, which intuitively presents the trainee's virtual operation performance: in the left “trainee virtual operation score”, the “operation efficiency” and “decision accuracy” scores are above 80 points, and the “innovative thinking” score is only 25 points, reflecting strong practical ability but insufficient collaborative innovation; in the right “trainee ability dimension”, “planning” and “inspection control” are above 80%, and “forward innovation” is 20%, indicating that the trainee's early planning ability meets the standard, but collaborative innovation is weak. This atlas can provide direct basis for subsequent reinforcement of collaborative teaching tasks.

[0054] The setting of the module directly solves the problems of single teaching evaluation system, missing of process data collection, lagging of evaluation results and inability to reflect ability growth in the background art. By directly collecting full-process and multi-modal original behavior data from the immersive virtual environment, the module realizes the evaluation of the operation process of the students instead of only the final results; by fine comparison with the standard operation model, it converts subjective experience judgment into objective quantitative score; especially by introducing the item response theory to calibrate and compare the cross-period ability value, it changes the evaluation focus from static "ability level" to dynamic "ability growth", realizing the value-added evaluation.

[0055] The ability increment atlas generated by the ability evolution evaluation module 105 provides a fine quantitative diagnosis of the effectiveness of the teaching process. However, the evaluation result itself is not the end point, and its core value lies in driving the self-optimization and iteration of the teaching system. In order to enable the evaluation conclusion to be transformed into practical strategies to improve the pertinence and effectiveness of subsequent teaching, the system introduces a teaching feedback optimization module 106. As a hub connecting "evaluation" and "redetermination", the module receives and analyzes the common short board of the student group and the individual specific deficiency revealed by the ability increment atlas, and then generates specific teaching strategy adjustment instructions and resource rematching suggestions. Its role is to convert the data insights generated in the evaluation link into executable teaching intervention measures and feed back to the upstream engineering-teaching integrated teaching engine module, thereby starting a new round of more targeted and personalized teaching design cycle. The operation of the teaching feedback optimization module marks the completion of the complete closed loop from "teaching implementation-effect evaluation" to "evaluation feedback-strategy optimization", realizing the adaptive evolution of the teaching process.

[0056] The teaching feedback optimization module 106 is configured to feed back the ability increment atlas to the adaptive task generator to drive generation of a new round of optimal teaching activity sequence. Specifically, the teaching feedback optimization module 106 comprises a strategy adjustment unit configured to analyze the ability increment atlas and identify common weak abilities of the student group and specific ability short boards of individuals; a resource rematching unit configured to rematch corresponding resource lists from an external resource library and a static standard library of the teaching data center module according to the identified common weak ability dimensions of the student group and the specific ability short boards of individuals, the resource lists including cases, micro-lessons, simulation faults or extended reading materials; and a task regeneration instruction unit configured to package and send the updated common weak ability dimensions of the student group and the specific ability short boards of individuals and the rematched resource lists to the adaptive task generator of the engineering-teaching integrated teaching engine module, triggering the adaptive task generator to generate a new round of optimal teaching activity sequence in a new round of strategy iteration, the new round of optimal teaching activity sequence including reinforcement training tasks and personalized learning path suggestions.

[0057] The teaching feedback optimization module 106 drives the closed-loop optimization of teaching strategies and the dynamic adaptation of learning resources according to the ability increment atlas generated by the ability evolution evaluation module. After the strategy adjustment unit receives the atlas data, it first clusters and statistically analyzes the ability increments in the six dimensions of "task understanding, plan making, decision implementation, check control, evaluation feedback, and transfer innovation" in the atlas. This unit identifies the dimensions that have generally negative growth or slow growth in a specific group of students, marking them as common weak ability dimensions; at the same time, it also identifies the individual's specific ability short board that deviates significantly from the average level of the group in certain dimensions. For example, analysis may find that the current student group generally scores low in the "check control" dimension, while a student has a prominent short board in the "decision implementation" dimension.

[0058] The resource rematching unit starts a new round of resource association retrieval according to the weak link identification output by the strategy adjustment unit. This unit accesses the static standard library and external resource library in the teaching data center module, and retrieves learning materials with higher matching degree for the target weak ability dimension according to semantic association and preset mapping rules. For the common weak dimension "check control", the system may match a series of micro-course videos and three-dimensional interactive cases about "standardized process of track transportation equipment point inspection" and "hidden defect features easily overlooked in fault troubleshooting" from the external resource library; for the individual's short board in "decision implementation", it may filter out complex fault disposal simulation tasks that emphasize "multi-scheme comparison and risk assessment" from the fault case knowledge graph associated with the static standard library. The resource retrieval process not only considers the relevance of content themes, but also evaluates whether the difficulty gradient of the material and the teaching form are suitable for targeted reinforcement training.

[0059] The task regeneration instruction unit encapsulates the updated learning situation data into a structured instruction package. This instruction package contains a list of identified common weak dimensions, specific ability short board descriptions for each learner, and a list of recommended alternative material resources for each type of weak link recommended by the resource rematching unit. This instruction package is sent to the adaptive task generator of the engineering-integrated teaching engine module. The adaptive task generator quantifies the weak link information in the instruction package as part of the reinforcement learning state vector in the new round of policy iteration. When generating a new optimal teaching activity sequence, the algorithm will prioritize teaching task combinations that can effectively train the identified weak links and prefer to use related resources recommended in the instruction package to construct these tasks, while meeting the original constraint conditions. For example, for a group with weak "inspection control", the generator may arrange a virtual inspection task sequence that contains multiple hidden defects and requires the learner to check them one by one according to the standard process; for an individual with a short board in "decision implementation", a decision-making task that requires him to weigh and choose among multiple maintenance plans may be inserted into his personal learning path.

[0060] This technical means directly solves the problem of evaluation result feedback lag, lack of data basis for teaching strategy adjustment, and inability to form a continuous optimization closed loop in the background art. By automatically analyzing the evaluation map into operational strategy adjustment instructions and resource matching suggestions, and directly feeding back to the task generation engine, the system realizes the automation link from learning effect diagnosis to teaching intervention implementation. The analysis function of the strategy adjustment unit allows teaching optimization to consider both group rules and individual differences; the resource rematching unit ensures that the intervention measures have high relevance content support; and the task regeneration instruction unit completes the conversion from optimization intention to specific teaching action.

[0061] In addition, the system also includes a cross-platform collaboration and resource sharing module, which is in communication connection with the teaching data platform module 101 and the engineering-integrated teaching engine module 103, and specifically includes: a collaborative interaction unit that uses the WebRTC real-time communication protocol to realize voice / video interaction, operation screen synchronization, and text collaborative annotation between learners and instructors, and supports multi-role permission hierarchical management; a learning data synchronization unit that uses a distributed database based on a cloud-edge collaboration architecture to synchronize the learning progress, training records, and ability evaluation results of learners to the cloud in real time, ensuring data consistency after multiple terminal logins; an intelligent resource pushing unit that uses a collaborative filtering algorithm to filter matching customized learning resources from the external resource library of the teaching data platform based on the weak links in the learner's ability profile, and pushes them to the learner's terminal in real time through a message pushing interface.

[0062] Specifically, the collaborative interaction unit deploys a service based on the WebRTC protocol, and when a student or a tutor accesses the system through different terminals, the unit establishes a point-to-point real-time audio and video communication channel and a data channel. In a collaborative training task, for example, when performing a "depot interlocking fault emergency handling" deduction, students playing the roles of dispatchers, on-site maintenance personnel, and safety monitors can join the same virtual collaboration space through their respective personal terminals. The unit not only transmits audio and video streams, but also synchronizes the thumbnail views of key operation screens on each terminal or the annotation information on the shared whiteboard, and supports permission management, such as allowing only the tutor role to make annotations on the shared screen or freeze the operation of a certain student. This low-latency interaction supports instant communication and collaborative decision-making among distributed roles.

[0063] The learning data synchronization unit interacts with the distributed database deployed in the cloud-edge collaborative architecture and is responsible for maintaining the consistency of global data. When a student completes a virtual operation training on a local edge terminal, the unit synchronizes the operation log, temporary evaluation score, and other process data generated during this training to the cloud center database. At the same time, when the student logs in from another terminal, the unit pulls the latest learning progress, complete ability profile, and historical training records from the cloud to ensure the continuity of the learning experience. This synchronization mechanism allows students to seamlessly switch between immersive terminals in a training room and desktop terminals on a personal computer, while maintaining synchronized updates of teaching progress and evaluation data.

[0064] The intelligent resource pushing unit, as an extension of personalized learning support, operates based on the student ability profile data from the engineering-education integrated teaching engine module. The unit uses a collaborative filtering algorithm that not only analyzes the weaknesses identified in the target student's ability profile, but also references learning resources commonly accessed or highly rated by other student groups with similar ability profile characteristics. For example, if a target student has a low ability value in the "traction system fault diagnosis" dimension, and the algorithm finds that other students with similar weaknesses generally show an ability improvement after learning a specific "traction inverter IGBT module fault principle and troubleshooting" micro course, the unit will mark this micro course resource as high correlation from the external resource library. The pushing unit then sends this resource link and related recommendation reasons to the terminal interface currently logged in by the student in real time through the integrated message pushing interface. The triggering time of the push can be combined with the teaching session, such as triggering immediately after the student completes a related task but the evaluation shows that there is a defect in their decision-making logic.

[0065] Thus, the problems of teaching being limited to fixed sites and single-machine devices, learning process data being isolated, and after-school learning lacking targeted guidance are solved. The collaborative interaction unit aggregates physically dispersed students and instructors into the same teaching environment through a standardized real-time communication protocol, enabling complex operation and maintenance process training that requires multiple roles to be carried out remotely, breaking through the time and space limitations of traditional training. The learning data synchronization unit enables continuous recording and multi-terminal adaptation of the learning process through cloud-edge data synchronization, providing a technical foundation for ubiquitous learning. The intelligent resource pushing unit uses collaborative filtering algorithms to combine group learning behavior patterns with individual capability gaps, enabling more socially intelligent resource recommendations and promoting autonomous deep learning after class.

[0066] Further, all modules are deployed in the same cloud-edge collaborative architecture, which includes a central cloud for storing historical operation and maintenance big data, heterogeneous data, and cross-school teaching resources; an edge cloud deployed in a training base for carrying out real-time calculation and low-latency rendering of teaching twins; a terminal edge node embedded in a personal immersive operation terminal and AR glasses for locally completing pose tracking and gesture recognition; and a unified time series database using a time series structured multi-label (TSMT, a storage engine structure designed specifically for time series databases) data model to label and store real operation and maintenance events, teaching task events, and student behavior events.

[0067] Specifically, all functional modules of the system are deployed in the same cloud-edge collaborative architecture, which is hierarchically processed according to data characteristics and computing requirements to optimize overall performance. The central cloud, as a data aggregation and macro-management node, centrally stores historical operation and maintenance big data generated from long-term operation of rail transit lines, heterogeneous data from the teaching data platform module after refinement and integration, and standardized teaching resources accumulated from different schools or enterprises. These data have the characteristics of large volume, relatively low update frequency, but high global sharing value. The edge cloud is deployed inside the training base of specific vocational schools or enterprise training centers, which is close to users in physical location and responsible for carrying out tasks with extremely high requirements for real-time calculation and response delay. The iterative operation of device-level physical simulation in the teaching twin module and the high-frame-rate rendering of complex three-dimensional models in the metaverse scene construction module are completed in the edge cloud to ensure the smoothness and real-time interaction of the student's operation in the virtual environment and avoid dizziness or operation asynchronization caused by network delay. The terminal edge node refers to the personal immersive operation terminal or AR glasses used directly by students, which has built-in sensors and processors for locally tracking and recognizing the student's head pose and hand movements, and uploading the processed simplified action instruction stream rather than the original video stream to the edge cloud, thereby greatly reducing the uplink data transmission bandwidth requirement and further reducing the end-to-end interaction delay.

[0068] The unified time series database, as the core of data organization throughout the cloud-edge three layers, uses a time series structured multi-label data model to store various events generated by the system. This model timestamps each data record with a main label and adds multiple structured dimension labels, such as event type, associated device ID, involved student ID, and assigned teaching task ID. For example, a record might be labeled as "timestamp T, event type: student operation behavior, device: train B brake unit, student: A, task: brake pad replacement training," with the content being the specific operation parameters of student A at that time. This storage method allows heterogeneous but strongly time-correlated data streams, such as sensor data from real operations, task scheduling instructions generated by the teaching engine, and behavior logs generated by students on the terminal, to be uniformly ingested, aligned, stored, and efficiently correlated. When teaching backtracking analysis or causal inference is needed, the system can quickly retrieve and reconstruct the complete event sequence from real device state changes, virtual task triggers, and student operation responses within a specific time period based on timestamps and multi-dimensional labels.

[0069] The cloud-edge collaborative architecture and its unified time series data model solve the problems of centralized system computing load, high response delay, and difficulty in analyzing multi-source heterogeneous data in the background art. By sinking computationally intensive and real-time demanding tasks to the edge, the quality of experience for immersive training is effectively guaranteed; by using the unified time series data model to standardize the description and correlated storage of cross-layer, multi-source events, a reliable and efficient data foundation is provided for fine-grained analysis and intelligent decision-making throughout the teaching process. This architecture and data model are not isolated; they are closely coupled with and support the functional requirements of the upper-layer application modules: the real-time computing requirements of the upper-layer modules define the necessity of edge cloud deployment, their reliance on historical data and cross-scenario knowledge calls for central cloud storage, and the process data they generate relies on the unified time series model for standardized governance and value extraction.

[0070] The embodiment of the present application achieves remarkable technical effects through the implementation of the above technical solutions. The system successfully constructs a complete technical chain from data aggregation, dynamic twin modeling, intelligent task arrangement, immersive training, multi-dimensional capability evaluation to closed-loop feedback optimization, and realizes the full-link intelligentization of the teaching process. Compared with the prior art, the system not only solves specific problems such as static presetting of teaching scenes, resource fragmentation, and single evaluation method, but more importantly, it drives through data and models to dynamically and accurately integrate "work" and "learning" in teaching activities. Trainees can conduct systematic training covering the complete work process and targeting at individual capability short boards in a highly realistic virtual operation and maintenance environment, and their comprehensive professional capabilities are continuously improved in a quantifiable and traceable manner. At the same time, the burden of teaching design of teachers is reduced, and the formulation and adjustment of teaching strategies are supported by objective data. The system provides a complete technical solution with good operability and generalizability for the large-scale, personalized and efficient cultivation of high-end technical and skilled personnel in the field of intelligent operation and maintenance of rail transit, effectively promoting the substantial landing and innovative development of the work-learning integrated teaching mode.

[0071] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, not representing the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0072] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A work-integrated learning intelligent teaching system for rail transit intelligent operation and maintenance, characterized in that, Comprise: Teaching data platform module for aggregating heterogeneous data of rail transit operation and maintenance teaching, including static standard data, dynamic teaching process data and external teaching resource data; Teaching twin module for dynamically constructing a three-layer twin model containing device mechanism, system business and teaching tasks based on the heterogeneous data, and generating a multi-dimensional composite teaching task set; Work-integrated learning teaching engine module with an adaptive task generator based on a reinforcement learning hybrid algorithm for automatically arranging an optimal teaching activity sequence according to the student's ability profile and the multi-dimensional composite teaching task set; Meta-universe scene construction module for calling corresponding virtual materials according to the optimal teaching activity sequence to construct a simulation scene for interactive virtual operation training for students; Ability evolution evaluation module for collecting training data of students in the simulation scene and comparing it with the standard operation model in the teaching twin module to generate an ability increment map; Teaching feedback optimization module for feeding the ability increment map back to the adaptive task generator to drive the generation of a new round of optimal teaching activity sequence. 2.The rail transit intelligent operation-oriented work-integrated learning intelligent teaching system according to claim 1, characterized in that, The teaching data platform module specifically comprises: Static standard library for storing and associating rail transit operation and maintenance capability standards, course standard maps and industry specifications in four-dimensional integration of "post, course, competition and certificate", and establishing semantic mapping relationships between course knowledge points, skill points and items in the capability standards through ontology modeling; Dynamic process library for real-time collection and structured storage of multi-modal data in the whole teaching process, including operation sequence logs of students in virtual operation training, dialogue texts generated by interaction with intelligent teaching assistants, role-playing and communication records in collaborative tasks, and key node annotation data of students' operation process based on the standard operation model by teachers, and the standard operation model is a pre-set operation specification model in the teaching twin module; External resource library for indexing and associating parameterized rail transit equipment 3D model library, fault case knowledge graph, maintenance procedure video clips and safety warning education materials. 3.The rail transit intelligent operation-oriented engineering-integrated intelligent teaching system according to claim 2, characterized in that, The three-layer twin model in the teaching twin module specifically comprises: Device-level mechanism twin layer for fault simulation and state prediction based on physical simulation models and historical operation data of key rail transit equipment to generate a set of potential hidden fault modes; System-level business twin layer connected with the device-level mechanism twin layer for associating and deducing the set of hidden fault modes with device topology relationship and operation scheduling rules based on operation regulation knowledge graph to generate a virtual business event chain affecting operation safety; Teaching-level task twin layer connected with the system-level business twin layer for receiving the virtual business event chain and combining teaching targets in the static standard library to decompose and reorganize the event chain into a multi-dimensional composite teaching task set containing fault disposal, scheduling coordination and safety briefing.

4. The intelligent teaching system for the integration of engineering and learning for rail transit intelligent operation according to claim 1, characterized in that, The work-integrated learning teaching engine module specifically comprises: A student ability profile generator is configured to output a "safety-efficiency-cost" three-dimensional ability vector based on student historical training data and real-time eye movement, operation sequence, and voice instruction multi-modal information through a graph neural network; A course difficulty adjuster is configured to monitor a reward function variance, and if the variance exceeds a threshold, dynamically adjust the difficulty of subsequent tasks according to the "zone of proximal development" principle; A knowledge graph gap diagnosis interface is configured to compare a "post-course examination certificate" four-dimensional knowledge graph with a three-dimensional ability vector of a student, and generate a gap vector; An adaptive task generator is configured to use the three-dimensional ability vector and the embedding vector of the teaching task set as a state, the task difficulty, the task type, and the resource combination as an action space, and the ability increment expectation as a reward function, and output a preliminary teaching activity sequence through an Actor-Critic architecture.

5. The ergonomics-integrated intelligent teaching system for rail transit intelligent operation according to claim 4, characterized in that, After the adaptive task generator outputs the preliminary teaching activity sequence, it further includes content rearrangement and task replacement of the preliminary teaching activity sequence based on the gap vector, to generate a revised teaching activity sequence, and outputs an optimal teaching activity sequence based on the task difficulty adjustment requirement fed back by the course difficulty adjuster. 6.The rail transit intelligent operation-oriented work-integrated learning intelligent teaching system according to claim 2, characterized in that, The metaverse scene construction module is specifically configured to: receive the optimal teaching activity sequence, and analyze the requirements for virtual environments, device states, fault phenomena, and interactive objects in the sequence; intelligently retrieve and call matching parameterized three-dimensional models, material maps, animation scripts, and physical engine configuration parameters from the external resource library according to the analysis results; based on a real-time rendering game engine, instantiate, spatially layout, and logically bind the called materials in real time according to the teaching logic, to generate an immersive simulation scene that is synchronized with the optimal teaching activity sequence and is interactive; wherein the immersive simulation scene supports fine operation training of a student in a first-person perspective through a personal terminal, and supports global situation deduction and role-playing in a third-person perspective in a collaborative sand table.

7. The ergonomics-integrated intelligent teaching system for rail transit intelligent operation according to claim 2, characterized in that, The ability evolution evaluation module specifically includes: a data acquisition unit configured to extract trajectory data, step completion time series, tool selection sequences, and communication content with other students from the background logs of virtual operation training; a multi-dimensional comparative analysis unit configured to compare the trajectory data, time series, and tool selection sequences with the optimal path, standard working hours, and standard tool set preset in the standard operation model, respectively, to calculate operation specification score, efficiency score, and decision accuracy score; and perform natural language processing on the communication content to analyze the collaboration intent and problem-solving logic, and generate collaboration communication and innovative thinking scores; a graph generation unit configured to calculate the ability level changes of a student in each ability dimension relative to the previous evaluation based on the project response theory according to the collaboration communication and innovative thinking scores of the multi-dimensional comparative analysis unit, and generate a visual ability increment graph in the form of a radar chart and an ability node growth chart, wherein the student has six ability dimensions including "task understanding, plan development, decision implementation, inspection control, evaluation feedback, and innovation transfer". 8.The rail transit intelligent operation-oriented work-integrated learning intelligent teaching system according to claim 2, characterized in that, The teaching feedback optimization module specifically comprises: A strategy adjustment unit for analyzing the ability increment atlas, identifying common weak abilities of the student group and specific ability short boards of individuals; A resource re-matching unit for re-matching corresponding resource lists, including cases, micro-lessons, simulation faults or extended reading materials, from the external resource library and static standard library of the teaching data center module according to the identified common weak ability dimensions of the student group and specific ability short boards of individuals; A task regeneration instruction unit for packaging and sending the updated common weak ability dimensions of the student group and specific ability short boards of individuals and the re-matched resource lists to the adaptive task generator of the work-integrated learning teaching engine module to trigger it to generate a new round of optimal teaching activity sequences for the common weak abilities and specific ability short boards in a new round of strategy iteration, the new round of optimal teaching activity sequences including reinforcement training tasks and personalized learning path suggestions. 9.The rail transit intelligent operation-oriented work-integrated learning intelligent teaching system according to claim 1, characterized in that, It also includes a cross-platform collaboration and resource sharing module, which is in communication connection with the teaching data center module and the work-integrated learning teaching engine module, specifically comprising: A collaborative interaction unit that uses the WebRTC real-time communication protocol to realize voice / video interaction, operation screen synchronization and text collaborative annotation between students and instructors, and supports multi-role permission hierarchical management; A learning data synchronization unit based on a distributed database of cloud-edge collaboration architecture, which synchronizes the learning progress, training records and ability evaluation results of students to the cloud in real time, ensuring data consistency after multi-terminal login; An intelligent resource pushing unit based on collaborative filtering algorithm, which screens and matches customized learning resources from the external resource library of the teaching data center according to the weak links in the student ability portrait, and pushes them to the student terminal in real time through the message pushing interface.

10. The ergonomics-integrated intelligent teaching system for rail transit intelligent operation and maintenance according to any one of claims 1-9, characterized in that, All modules are deployed in the same cloud-edge collaboration architecture, which comprises: A central cloud for storing historical operation and maintenance big data, the heterogeneous data and cross-school teaching resources; An edge cloud deployed in the training base for carrying out real-time calculation and low-latency rendering of teaching twins; A terminal edge node embedded in a personal immersive operation terminal and AR glasses for local completion of pose tracking and gesture recognition; A unified time series database using a time series structured multi-label TSMT data model to label and store real operation and maintenance events, teaching task events and student behavior events.

Images