Immersive generative teaching scene intelligent pushing system facing regional industrial demand

Abstract

The application belongs to the technical field of big data analysis, and specifically discloses an immersive generative teaching scene intelligent pushing system facing regional industrial demand. The system comprises: an industrial demand dynamic sensing module, which is used for real-time collection and vectorization processing of regional industrial data; a teaching element decoupling and knowledge graph construction module, which is used for constructing a teaching knowledge graph dynamically associated with industrial demand; an immersive scene generation engine, which is used for dynamically synthesizing a three-dimensional interactive teaching scene based on the knowledge graph; and an intelligent matching and accurate pushing module, which is used for matching the learner's ability and the industrial demand based on multi-objective optimization, and accurately pushing the adapted teaching scene. Through the above scheme, the application realizes dynamic accurate matching of teaching content and regional industrial demand and personalized immersive teaching pushing.

Description

Technical Field

[0001] This invention belongs to the field of big data analytics technology, specifically relating to an intelligent push system for immersive, generative teaching scenarios tailored to regional industry needs. Background Technology

[0002] Against the backdrop of the integration of higher education and industry, leveraging digital technologies to build an efficient and precise talent cultivation system has become crucial for promoting regional economic transformation and upgrading. Developing teaching resources and constructing scenarios tailored to specific industry needs is a core element in achieving deep integration of industry and education and enhancing the relevance of talent cultivation.

[0003] Existing technologies primarily create immersive learning environments by constructing digital teaching platforms or introducing virtual simulations, but their logic for generating and pushing teaching scenarios has significant limitations. Existing systems typically generate scenarios based on pre-set general teaching syllabi or static knowledge bases, failing to effectively access and analyze dynamically changing regional industry data, resulting in a disconnect between teaching content and real-world industry technology frontiers and job skill requirements.

[0004] Meanwhile, capturing industry demand relies heavily on periodic manual surveys and report analysis, resulting in information lag and coarse granularity. This makes it impossible to map the information accurately and in real-time onto the design of specific teaching scenarios, creating a dual disconnect between educational supply and industry demand at both the data and application layers. This disconnect makes immersive teaching difficult to effectively support the strategic goals of higher education institutions in accurately serving regional industrial development and responding to rapid technological iterations. Building a system that can intelligently perceive industry needs and dynamically generate teaching scenarios has become an urgent technical challenge. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent push system for immersive generative teaching scenarios tailored to the needs of regional industries, comprising:

[0006] The dynamic perception module for industry demand is used to collect and structure regional industry data from multiple heterogeneous data sources in real time to generate a set of demand feature vectors that represent the current dynamic demand of regional industries.

[0007] The teaching element decoupling and knowledge graph construction module is used to standardize and decouple knowledge, skills and contexts in the teaching field, and construct a dynamically expandable teaching knowledge graph. The teaching knowledge graph semantically associates teaching elements with demand feature vectors from the industry demand dynamic perception module.

[0008] An immersive scene generation engine is used to dynamically synthesize highly immersive and targeted 3D interactive teaching scenes based on the teaching knowledge graph and real-time industry needs.

[0009] The intelligent matching and precise push module is used to achieve the optimal matching between industry needs, teaching resources and learner profiles, and to control the precise push of the scenes synthesized by the immersive scene generation engine.

[0010] Preferably, the industry demand dynamic perception module includes an industry data crawling unit, a multimodal data fusion unit, and a demand feature vectorization unit;

[0011] The industry data crawling unit establishes data connections with regional industry policy release platforms, key enterprise recruitment portals, technology patent databases, and industry analysis report websites through a preset application programming interface, and performs data crawling tasks at a set time frequency.

[0012] The multimodal data fusion unit cleans, deduplicatizes, and standardizes the format of the captured text, tables, and structured data. It also transforms unstructured text reports into structured skill requirements, technology trends, and job description data through named entity recognition and keyword extraction algorithms in natural language processing.

[0013] The demand feature vectorization unit maps the fused structured data into high-dimensional feature vectors based on a preset industry demand classification system. The industry demand classification system includes at least technical skills dimension, process flow dimension, safety standard dimension, and soft quality requirements dimension. The vectorization process uses a word embedding model to convert text descriptions into numerical vectors and normalizes the numerical indicators.

[0014] Preferably, the teaching element decoupling and knowledge graph construction module includes a teaching element atomization unit, a relationship mining unit, and a knowledge graph dynamic update unit;

[0015] The atomic units of the teaching elements are based on the national higher education teaching standards. They decompose course knowledge points, practical skills points, and typical work scenarios into the smallest indivisible teaching elements, and assign each element a unique identifier and metadata tag.

[0016] The association mining unit, based on a large number of teaching outlines, textbooks and excellent teaching cases, uses graph neural network algorithms to automatically mine and establish predecessor and successor relationships, collaborative teaching relationships and contextual dependencies among teaching elements, forming an initial teaching knowledge graph.

[0017] The graph dynamic update unit receives the demand feature vector from the industry demand dynamic perception module. Through semantic similarity calculation and graph structure matching algorithm, it associates the industry demand vector with the teaching element nodes in the knowledge graph and assigns weights to the associated edges. The weights represent the correlation strength between the teaching elements and the corresponding industry demands. At the same time, the graph dynamic update unit incrementally expands the knowledge graph according to the newly associated industry demand nodes.

[0018] Preferably, the immersive scene generation engine includes a scene logic orchestration unit, a 3D asset intelligent invocation unit, and a physical and interaction rule injection unit;

[0019] The scene logic orchestration unit extracts a set of related teaching element nodes and their relationships from the teaching knowledge graph according to the scene generation instructions issued by the intelligent matching and precise push module, and automatically generates the teaching narrative logic and task flow script corresponding to the teaching elements.

[0020] The intelligent 3D asset calling unit is connected to a pre-built 3D teaching asset library. All assets in the 3D teaching asset library are affixed with semantic tags corresponding to teaching element identifiers. Based on the script and semantic tags output by the scene logic arrangement unit, the intelligent 3D asset calling unit intelligently retrieves and calls the required 3D models, audio, video and interface elements from the asset library through a tag matching algorithm.

[0021] The physical and interactive rule injection unit injects physical attributes that conform to the real industrial environment into the called 3D model, including gravity, collision, and material texture. Based on the teaching task process script, it pre-sets interactive logic at key operation points in the scene, thereby generating interactive, high-fidelity immersive teaching scene instances.

[0022] Preferably, the intelligent matching and precise push module includes a learner ability profiling unit, a multi-objective optimization matching unit, and a push strategy execution unit;

[0023] The learner competency profiling unit continuously collects learners' historical behavioral data within the teaching platform, including course completion rate, practical training results, skills assessment results, and interactive operation records in immersive scenarios. It constructs and dynamically updates learners' multi-dimensional competency profile vectors through competency modeling algorithms.

[0024] The multi-objective optimization matching unit takes the set of demand feature vectors, the teaching knowledge graph, and the learner ability profile vectors as inputs to establish a multi-objective matching optimization model. The first optimization objective of the multi-objective matching optimization model is to maximize the matching degree between the industry demand features covered by the push scenario and the gaps in the learner's ability to be improved. The second optimization objective is to minimize the estimated cognitive load of the learner in order to master the teaching elements required for the scenario. The third optimization objective is to maximize the resource adaptability between the teaching resources required for the scenario and the existing training conditions of the college. The multi-objective optimization matching unit uses a multi-objective evolutionary algorithm to solve the model and outputs a set of Pareto optimal teaching scenario generation scheme sequences.

[0025] The push strategy execution unit selects the final execution plan from the optimal plan sequence according to the preset push strategy, generates a scene generation instruction containing the target teaching element set and scene complexity parameters, and sends it to the immersive scene generation engine. At the same time, the push strategy execution unit pushes the generated scene to the target learner's terminal or the designated immersive training classroom through the teaching management platform.

[0026] Preferably, the word embedding model used by the demand feature vectorization unit incorporates an industry domain dictionary as prior knowledge during training. The industry domain dictionary is automatically constructed and updated by continuously analyzing high-frequency terms and co-occurrence relationships in industry policy documents and technology white papers.

[0027] Preferably, the pre-built 3D teaching asset library connected to the 3D asset intelligent calling unit adopts a microservice-based distributed architecture for storage and management. In addition to geometric data and texture data, each 3D asset object is associated with a structured attribute description file. The attribute description file defines in detail the interactive components, state variables and acceptable interactive operation instruction set of the asset in the teaching scenario.

[0028] Preferably, in the multi-objective matching optimization model established by the multi-objective optimization matching unit, the gap between the learner's ability profile vector and the industry demand feature vector is defined as the ability gap vector, and the matching degree is measured by calculating the cosine similarity between the industry demand vector associated with the teaching elements covered by the scenario and the ability gap vector.

[0029] Cognitive load prediction is calculated based on the complexity weight of teaching elements and the average time spent on related content of learners' historical learning elements.

[0030] Resource suitability is calculated by comparing the list of assets required for the scenario with the list of assets in the school's inventory, and taking into account equipment reservation status and venue scheduling constraints.

[0031] Preferably, the push strategy preset by the push strategy execution unit includes an aggressive strategy, a robust strategy, and a balanced strategy;

[0032] The aggressive strategy prioritizes the solution with the highest matching degree; the robust strategy prioritizes the solution with the lowest cognitive load prediction.

[0033] The balancing strategy seeks a balance between matching degree and cognitive load; the push strategy can be set globally by teacher administrators according to the teaching plan or individually specified for specific learner groups.

[0034] Preferably, the system further includes a closed-loop evaluation module for teaching effectiveness, which is used to collect learners' real-time operation data in immersive scenarios, assessment scores after scenario completion, and practical evaluations from industry mentors. By comparing the changes in learners' ability profiles before and after scenario learning, the module calculates the teaching effectiveness gain value of scenario push and feeds this gain value back to the intelligent matching and precise push module for dynamically adjusting learners' ability profile vectors and optimizing the weight parameters of the model.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] 1. This invention, by constructing a dynamic industry demand perception module, achieves real-time, automated collection and structured processing of multi-source heterogeneous regional industry data. It transforms traditional, manual, time-consuming, and coarse-grained industry surveys into continuous, precise, and data-driven demand perception capabilities. This system can directly capture subtle changes in cutting-edge industry technologies and job skill requirements, quantifying them into calculable demand feature vectors. This solves the problem of the disconnect between teaching data sources and industry data sources, providing a solid data foundation for the dynamic alignment of teaching content.

[0037] 2. This invention creatively designs a module for decoupling teaching elements and constructing a knowledge graph. This decouples the fixed curriculum system into flexibly recombinable knowledge atoms, and establishes semantic connections between these atoms and dynamic industry needs through knowledge graph technology. This structure transforms teaching resources from a static, closed system into a dynamic, intelligently expanding, and evolving knowledge network that adapts to changing industry demands. The immersive scene generation engine synthesizes scenes on demand based on this dynamic network, ensuring that each generated teaching scene embeds the latest industry demand information, achieving precise and dynamic matching between teaching content and industry needs.

[0038] 3. This invention introduces a personalized matching mechanism based on multi-objective optimization through an intelligent matching and precise push module. This personalized matching mechanism not only considers the matching between industry needs and learners' skill gaps, but also coordinates learners' cognitive load with the actual resource allocation of institutions, thereby seeking the globally optimal push solution under multiple constraints. This refined decision-making process transcends simple tag filtering or rule-based push, generating immersive learning tasks most suitable for learners at different levels and progress levels, improving the pertinence and effectiveness of teaching interventions, and truly achieving the unity of large-scale education and personalized training.

[0039] 4. This invention establishes a complete closed loop from scenario-based delivery to effect feedback and model optimization through a closed-loop evaluation module for teaching effectiveness. The system can objectively evaluate the teaching value of each delivery using learners' real behavioral and outcome data, and continuously optimize the matching algorithm and delivery strategy accordingly. This data-driven self-evolutionary capability enables the system to continuously adapt to changes in industry needs, differences in learner groups, and changes in the teaching environment, maintaining the accuracy and effectiveness of its delivery scenarios in the long term, thus constructing a continuously vibrant intelligent teaching ecosystem. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the framework of the knowledge graph construction and evolution principle of the dynamic correlation between teaching elements and industry needs in this invention;

[0042] Figure 3 This is a multi-objective optimization matching logic framework diagram of the intelligent matching and precise push module in this invention;

[0043] Figure 4 This is a flowchart of the dynamic synthesis logic process of the immersive scene generation engine based on knowledge graph and asset library in this invention.

[0044] Figure 5 This is a flowchart illustrating the multi-source heterogeneous data processing and characterization process of the dynamic perception module for industry demand in this invention. Detailed Implementation

[0045] Example 1: Reference Figures 1 to 5The intelligent push system for immersive generative teaching scenarios proposed in this invention, tailored to regional industry needs, consists of four core functional modules: a dynamic perception module for industry needs, a decoupling module for teaching elements and a knowledge graph construction module, an immersive scenario generation engine, and an intelligent matching and precise push module. It can be further integrated with a closed-loop evaluation module for teaching effectiveness to achieve system self-optimization. The modules interact with each other through standardized data interfaces, forming a complete technical closed loop from industry data perception, teaching knowledge modeling, immersive scenario synthesis to personalized push execution.

[0046] First, the dynamic industry demand sensing module is responsible for real-time collection, structured processing, and feature vectorization of multi-source heterogeneous data from the regional industrial environment. The operational logic of the dynamic industry demand sensing module is shown in the attached diagram. Figure 5 As shown, it contains three sub-units: an industry data crawling unit, a multimodal data fusion unit, and a demand feature vectorization unit.

[0047] The industry data crawling unit establishes stable data connections with regional industry policy release platforms, key enterprise recruitment portals, technology patent databases, and industry analysis report websites through a pre-defined application programming interface (API). The unit automatically executes data crawling tasks at a set time frequency (e.g., every 24 hours or every 72 hours) to ensure the timeliness of the acquired industry information. The types of data crawled include, but are not limited to, skill requirement fields in job postings, descriptions of technological development directions in industry policy documents, key technical terms in patent abstracts, and explanations of emerging technological processes in industry white papers.

[0048] The multimodal data fusion unit receives the raw data stream from the industry data crawling unit and performs cleaning, deduplication, and format standardization on it. For unstructured text data, the multimodal data fusion unit uses named entity recognition models and keyword extraction algorithms from natural language processing to transform the raw text into structured skill requirements, technology trends, and job descriptions. For example, a job description for a "New Energy Vehicle Battery Management System Development Engineer" can be broken down into structured fields such as "Skill Requirements: Embedded C Language Programming, CAN Bus Protocol, BMS Algorithm," "Technology Trends: High Energy Density Batteries, Thermal Runaway Early Warning," and "Soft Skills Requirements: Cross-departmental Collaboration Ability, Rapid Learning Ability." All structured data is uniformly stored in an intermediate data buffer, awaiting subsequent vectorization processing.

[0049] The demand feature vectorization unit, based on a pre-defined industry demand classification system, maps the fused structured data into high-dimensional feature vectors. This industry demand classification system includes at least four dimensions: technical skills, process flow, safety standards, and soft skills requirements. Each dimension has several fine-grained labels; for example, the "technical skills" dimension may include sub-items such as "Python programming," "PLC control," and "CAD modeling." The vectorization process uses a word embedding model to convert text descriptions into numerical vectors; for numerical indicators (such as "requires more than 3 years of work experience"), normalization is performed to ensure they fall within the range of 0 to 1.

[0050] Finally, the demand feature vectorization unit outputs a set of demand feature vectors representing the current dynamic demands of industries in the region. Each vector corresponds to a specific industry demand item, and its dimension is consistent with the total number of labels in the preset classification system. As a preferred embodiment of the invention, this word embedding model incorporates an industry domain dictionary as prior knowledge during the training phase. The domain dictionary is automatically constructed and updated by continuously analyzing high-frequency terms and their co-occurrence relationships in industry policy documents and technical white papers, thereby significantly improving the domain specialization and semantic accuracy of demand text vectorization.

[0051] Secondly, the teaching element decoupling and knowledge graph construction module is responsible for standardizing and decoupling knowledge, skills, and contexts in the teaching domain, and constructing a dynamically expandable teaching knowledge graph. The operating principle of this teaching element decoupling and knowledge graph construction module is shown in the attached figure. Figure 2 As shown, it internally includes atomic units of teaching elements, relationship mining units, and dynamic graph updating units. The atomic units of teaching elements, based on national higher education teaching standards, decompose course knowledge points, practical skills points, and typical work scenarios into the smallest indivisible teaching elements.

[0052] For example, the skill point of "CNC lathe operation" can be decoupled into multiple atomic elements such as "workpiece clamping", "tool selection", "program input", and "cutting parameter setting". Each teaching element is assigned a unique identifier (such as TE-2024-0876) and metadata tags. The metadata includes the major category, prerequisite knowledge requirements, typical application scenarios, and recommended teaching time.

[0053] The relational mining unit, based on a large amount of teaching syllabus, textbooks, and excellent teaching cases, uses graph neural network algorithms to automatically mine and establish predecessor-successor relationships, collaborative teaching relationships, and contextual dependencies among teaching elements. For example, there is a predecessor-successor relationship between the element "PLC ladder diagram programming" and the element "motor forward and reverse rotation control circuit construction"; there is a strong contextual dependency between "wearing safety protection equipment" and "high-voltage electricity operation training". These relationships are encoded as directed edges and assigned weight values ​​(e.g., 0.8 indicates strong dependency), thus forming an initial teaching knowledge graph. This teaching knowledge graph is stored in the form of a graph database, supporting efficient node querying and path traversal.

[0054] The graph dynamic update unit receives a set of demand feature vectors from the industry demand dynamic perception module. Through semantic similarity calculation and graph structure matching algorithms, it associates the industry demand vectors with teaching element nodes in the knowledge graph. Specifically, the graph dynamic update unit calls a model based on a bidirectional encoder representation. This model is pre-trained using a parallel corpus consisting of industry technical documents and higher education textbooks to ensure accurate comparison between industry demand descriptions and teaching element descriptions within the same semantic space.

[0055] When the semantic similarity between an industry demand vector and a teaching element node exceeds a preset threshold (e.g., 0.75), the system establishes a new connection edge between them and assigns weights based on the similarity value. Simultaneously, if an industry demand cannot find enough matching teaching elements in the existing graph, the graph incremental expansion mechanism is triggered. After review by teaching experts, corresponding teaching element nodes and their relationships are added, thereby achieving the dynamic evolution of the knowledge graph.

[0056] Third, the immersive scene generation engine is used to dynamically synthesize highly immersive and targeted 3D interactive teaching scenes based on teaching knowledge graphs and real-time industry needs. Its synthesis logic is shown in the attached figure. Figure 4 As shown, it includes a scene logic orchestration unit, a 3D asset intelligent invocation unit, and a physical and interaction rule injection unit.

[0057] The scenario logic orchestration unit extracts a set of relevant teaching element nodes and their relationships from the teaching knowledge graph based on the scenario generation instructions issued by the intelligent matching and precise push module. For example, if the instruction requires the generation of an "industrial robot fault diagnosis" scenario, the scenario logic orchestration unit will extract element nodes such as "robot teach pendant operation," "servo motor status reading," and "emergency stop circuit detection," and automatically generate the teaching narrative logic and task flow script based on their predecessor and successor relationships. This script defines the scenario's starting conditions, task objectives, key operation steps, error operation feedback mechanisms, and success criteria.

[0058] The 3D asset intelligent retrieval unit connects to a pre-built 3D teaching asset library. This asset library uses a microservice-based distributed architecture for storage and management. Each 3D asset object contains geometric and texture data, as well as a structured attribute description file. The attribute description file defines in detail the interactive components of the asset in the teaching scenario (such as "teaching pendant screen" and "emergency stop button"), state variables (such as "motor temperature" and "communication status"), and the acceptable set of interactive operation instructions (such as "click", "rotate", and "drag").

[0059] The 3D asset intelligent retrieval unit uses the semantic tags corresponding to the scripts and teaching element identifiers output by the scene logic orchestration unit to intelligently retrieve and call the required 3D models, audio, video, and interface elements from the asset library through a tag matching algorithm. For example, when the script contains the phrase "check the robot controller cooling fan," the system will automatically call the controller model with the semantic tag "cooling fan" and load its corresponding fan sound effect.

[0060] The physics and interaction rule injection unit injects physical properties consistent with real-world industrial environments into the invoked 3D model, including gravity, collision detection, and material textures (such as the reflectivity of metals and the coefficient of friction of plastics). Simultaneously, based on the teaching task flow script, the physics and interaction rule injection unit pre-sets interactive logic at key operation points within the scene. For example, in the "adjust inverter parameters" task, the system pre-sets parameter input boxes on the inverter panel and binds verification rules: if the input value exceeds the safe range, an alarm sound and visual warning are triggered; if the input is correct, the next operation step is unlocked. Through this process, the system ultimately generates an interactive, high-fidelity, and logically complete immersive teaching scene instance, which can be directly deployed on virtual reality headsets, augmented reality glasses, or desktop 3D simulation platforms.

[0061] Fourth, the intelligent matching and precise push module is used to achieve optimal matching between industry needs, teaching resources, and learner profiles, and to control precise push notifications within specific scenarios. Its matching logic is shown in the attached diagram. Figure 3 As shown, it includes a learner ability profiling unit, a multi-objective optimization matching unit, and a push strategy execution unit.

[0062] The learner profiling unit continuously collects learners’ historical behavioral data within the teaching platform, including course completion rate, practical training results, skills assessment results, and interactive operation records in immersive scenarios (such as the order of operation steps, number of incorrect attempts, and task completion time).

[0063] The competency modeling algorithm is the core algorithm for quantifying learners' mastery of each atomized teaching element in the teaching knowledge graph. By integrating learners' multi-source learning behavior data, it constructs a multi-dimensional competency profile vector that corresponds one-to-one with each teaching element, thereby achieving accurate quantification and dynamic updating of learners' competency status. The specific formula is as follows:

[0064] ;

[0065] in, The score represents the mastery level of the i-th teaching element (within the range [0,1]). The weights of the k-th type of behavioral data ( ), This is the standardized value of the learner's behavior data corresponding to the k-th class in the i-th element (normalized to the [0,1] interval).

[0066] Through the ability modeling algorithm, learner ability profile units construct and dynamically update learner multidimensional ability profile vectors. The dimensions of the multidimensional ability profile vectors correspond one-to-one with the teaching element nodes in the teaching knowledge graph, and the value of each dimension represents the learner's mastery of the element (the value ranges from 0 to 1, with 1 indicating complete mastery).

[0067] The multi-objective optimization matching unit uses a set of demand feature vectors, a teaching knowledge graph, and learner ability profile vectors as inputs to establish a multi-objective matching optimization model. This model is a multi-objective constrained optimization mathematical model, with its core structure consisting of "three objective functions + associated constraints," specifically including three optimization objectives:

[0068] First, maximize the match between the industry demand characteristics covered by the push scenarios and the gaps in learners' skills that need to be improved.

[0069] Second, minimize the learner's estimated cognitive load on the teaching elements required to master the scenario.

[0070] Third, maximize the resource compatibility between the teaching resources required for the scenario and the existing practical training conditions of the institution. The competency gap vector is defined as the element-wise difference (taking the positive value) between the demand feature vector and the learner competency profile vector. Matching degree. It is measured by the cosine similarity between the industry demand vector and the capability gap vector associated with the teaching elements covered by the scenario, as shown in the following formula:

[0071] ;

[0072] The weighted sum of the industry demand vectors covered by the scenario. This represents the capability gap vector. Cognitive load estimation is calculated based on the complexity weights of teaching elements (pre-calibrated by teaching experts) and the average time spent on related content of learners' historical learning elements. Resource suitability is calculated by comparing the list of assets required for the scenario with the list of school assets, taking into account equipment reservation status and venue scheduling constraints, and is a real number between 0 and 1. This multi-objective optimization matching unit uses multi-objective evolutionary algorithms such as NSGA-II to solve the model. The core revolves around "population iterative optimization" to achieve Pareto optimal solution search. The specific process is as follows: an initial population that meets the constraints of knowledge graph association and asset call feasibility is generated by encoding "teaching element combination + scenario complexity parameter"; for each solution, the fitness values ​​of matching degree, cognitive load (inverse), and resource suitability are calculated; the dominant level is divided by fast non-dominated sorting, and the diversity of solutions is ensured by combining congestion calculation; the next generation population is generated through genetic operations of selection, crossover, and mutation; the iteration is repeated until a preset number of times or fitness is stable, and finally, the Pareto optimal teaching scenario generation scheme sequence of the non-dominated level is output.

[0073] The push strategy execution unit selects the final execution plan from the optimal plan sequence based on the preset push strategy. The system presets three push strategies: aggressive strategy, robust strategy, and balanced strategy. The aggressive strategy prioritizes the plan with the highest matching degree and is suitable for the skills enhancement training stage; the robust strategy prioritizes the plan with the lowest cognitive load prediction and is suitable for the new knowledge introductory learning stage; the balanced strategy seeks a balance between matching degree and cognitive load and is suitable for the regular teaching promotion stage.

[0074] The push strategy can be set globally by the teacher administrator according to the teaching plan or individually specified for a specific group of learners. After selecting a scheme, the push strategy execution unit generates a scene generation instruction containing the set of target teaching elements and scene complexity parameters (such as the number of interaction points and the level of physical simulation accuracy), and sends it to the immersive scene generation engine. At the same time, the push strategy execution unit pushes the generated scene to the target learner's terminal or the designated immersive training classroom through the teaching management platform, and records the scene's usage log, completion status, and preliminary feedback data.

[0075] In addition, the system can integrate a closed-loop evaluation module for teaching effectiveness. This module collects real-time operational data of learners in immersive scenarios (such as operation paths and decision point selections), assessment scores after scenario completion (automatically generated by the built-in scoring engine), and practical evaluations from industry mentors (entered through an external review interface). By comparing changes in learners' competency profiles before and after scenario learning, this closed-loop evaluation module calculates the teaching effectiveness gain value for that scenario push, defined as the weighted sum of the improvement in the competency profile vector across relevant dimensions.

[0076] The teaching effectiveness gain is fed back to the intelligent matching and precise push module, which dynamically adjusts the update rate of the learner ability profile vector and the weight parameters of each objective in the multi-objective optimization model. For example, if a certain type of scenario consistently and stably generates high gain, the system will automatically increase its priority weight in the matching model, thereby achieving adaptive iterative optimization of the push strategy.

[0077] In summary, this embodiment constructs a complete technology chain from industry data perception to the generation and precise delivery of immersive teaching scenarios through the coordinated operation of four core modules. The system not only achieves dynamic alignment between teaching content and regional industry needs, but also ensures the personalization, appropriateness, and effectiveness of the delivered scenarios through multi-objective optimization and closed-loop feedback mechanisms, providing solid technical support for the intelligent transformation of higher education.

[0078] Example 2: Based on the previous examples, this example further refines the implementation mechanism of the graph dynamic update unit in the teaching element decoupling and knowledge graph construction module, and enhances the immersive scene generation engine's support for multi-person collaborative teaching scenarios.

[0079] When performing semantic similarity calculations, the graph dynamic update unit not only relies on a model based on bidirectional encoder representations but also introduces a context-aware attention mechanism. Specifically, when dealing with complex industry requirement descriptions (such as "Smart factory MES system integration engineers need to be familiar with the OPCUA protocol, database optimization, and human-machine interface design"), the system will break down the description into multiple clauses and calculate the similarity between each clause and the teaching element node.

[0080] Subsequently, an attention weighting mechanism is used to weight and fuse the matching results of different clauses, thereby avoiding the decrease in matching accuracy caused by semantic dilution from long texts. The attention weight is determined by the information entropy of the clause in the overall demand description; the higher the information entropy (i.e., the more discriminative), the greater its weight. This mechanism significantly improves the matching accuracy between complex and multifaceted industry demands and teaching elements.

[0081] In terms of immersive scene generation, the scene logic orchestration unit has been expanded to support the generation of multi-person collaborative task scripts. When the instructions issued by the intelligent matching and precise push module specify that the target learners are a group (e.g., 3 to 5 people), the scene logic orchestration unit will extract a set of teaching element nodes with division of labor and collaboration relationships from the teaching knowledge graph. For example, in the "automated production line debugging" scenario, the system will assign tasks such as "PLC program debugging," "robot trajectory planning," and "vision system calibration" to different roles. The task flow script clearly marks the operating permissions, information sharing scope, and collaborative triggering conditions of each role (e.g., "the robot role can start trajectory simulation only after the PLC program is successfully downloaded").

[0082] The 3D asset intelligent invocation unit accordingly invokes 3D assets that support concurrent multi-user interaction. A "role binding" field has been added to the attribute description files of these assets to specify that asset components are only visible or operable by specific roles. The physics and interaction rule injection unit injects a network synchronization mechanism into the scene, ensuring that the scene state remains consistent across all learner terminals.

[0083] The system employs a lightweight protocol based on state synchronization, transmitting only change events of key state variables (such as device on / off status and parameter values) rather than full-scene data, thereby reducing network bandwidth consumption. Simultaneously, the system incorporates a conflict resolution strategy: when multiple learners simultaneously operate on the same interaction point, the system determines the validity of the operation based on role priority or operation timestamp and provides feedback to other learners regarding the reason for the operation being blocked.

[0084] In multi-person scenarios, the push strategy execution unit comprehensively considers the differences in the ability profiles of team members. The input to the multi-objective optimization matching unit is expanded to a team ability profile matrix, where each row corresponds to the ability vector of a member. The optimization model adds a new objective: maximizing the balance of overall team ability improvement, i.e., minimizing the standard deviation of ability gains among members. This objective is achieved by adding a balance penalty term to the objective function. The resulting scenario not only meets industry demand matching but also promotes knowledge complementarity and collaborative ability development among team members.

[0085] Through the above enhancements, this embodiment enables the system to effectively support the increasingly important team collaboration and project-based learning models in modern higher education, further expanding the application boundaries and teaching value of the system.

[0086] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0087] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent push system for immersive generative teaching scenarios tailored to regional industry needs, characterized in that: include: The dynamic perception module for industry demand is used to collect and structure regional industry data from multiple heterogeneous data sources in real time to generate a set of demand feature vectors that represent the current dynamic demand of regional industries. The teaching element decoupling and knowledge graph construction module is used to standardize and decouple knowledge, skills and contexts in the teaching field, and construct a dynamically expandable teaching knowledge graph. The teaching knowledge graph semantically associates teaching elements with demand feature vectors from the industry demand dynamic perception module. An immersive scene generation engine is used to dynamically synthesize highly immersive and targeted 3D interactive teaching scenes based on the teaching knowledge graph and real-time industry needs. The intelligent matching and precise push module is used to achieve the optimal matching between industry needs, teaching resources and learner profiles, and to control the precise push of the scenes synthesized by the immersive scene generation engine.

2. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 1, characterized in that, The dynamic perception module for industry demand includes an industry data crawling unit, a multimodal data fusion unit, and a demand feature vectorization unit. The industry data crawling unit establishes data connections with regional industry policy release platforms, key enterprise recruitment portals, technology patent databases, and industry analysis report websites through a preset application programming interface, and performs data crawling tasks at a set time frequency. The multimodal data fusion unit cleans, deduplicatizes, and standardizes the format of the captured text, tables, and structured data. It also transforms unstructured text reports into structured skill requirements, technology trends, and job description data through named entity recognition and keyword extraction algorithms in natural language processing. The demand feature vectorization unit maps the fused structured data into high-dimensional feature vectors based on a preset industry demand classification system. The industry demand classification system includes at least technical skills dimension, process flow dimension, safety standard dimension, and soft quality requirements dimension. The vectorization process uses a word embedding model to convert text descriptions into numerical vectors and normalizes the numerical indicators.

3. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 1, characterized in that, The teaching element decoupling and knowledge graph construction module includes a teaching element atomization unit, a relationship mining unit, and a graph dynamic update unit. The atomic units of the teaching elements are based on the national higher education teaching standards. They decompose course knowledge points, practical skills points, and typical work scenarios into the smallest indivisible teaching elements, and assign each element a unique identifier and metadata tag. The association mining unit, based on a large number of teaching outlines, textbooks and excellent teaching cases, uses graph neural network algorithms to automatically mine and establish predecessor and successor relationships, collaborative teaching relationships and contextual dependencies among teaching elements, forming an initial teaching knowledge graph. The graph dynamic update unit receives the demand feature vector from the industry demand dynamic perception module. Through semantic similarity calculation and graph structure matching algorithm, it associates the industry demand vector with the teaching element nodes in the knowledge graph and assigns weights to the associated edges. The weights represent the correlation strength between the teaching elements and the corresponding industry demands. At the same time, the graph dynamic update unit incrementally expands the knowledge graph according to the newly associated industry demand nodes.

4. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 1, characterized in that, The immersive scene generation engine includes a scene logic orchestration unit, a 3D asset intelligent invocation unit, and a physical and interaction rule injection unit; The scene logic orchestration unit extracts a set of related teaching element nodes and their relationships from the teaching knowledge graph according to the scene generation instructions issued by the intelligent matching and precise push module, and automatically generates the teaching narrative logic and task flow script corresponding to the teaching elements. The intelligent 3D asset calling unit is connected to a pre-built 3D teaching asset library. All assets in the 3D teaching asset library are affixed with semantic tags corresponding to teaching element identifiers. Based on the script and semantic tags output by the scene logic arrangement unit, the intelligent 3D asset calling unit intelligently retrieves and calls the required 3D models, audio, video and interface elements from the asset library through a tag matching algorithm. The physical and interactive rule injection unit injects physical attributes that conform to the real industrial environment into the called 3D model, including gravity, collision, and material texture. Based on the teaching task process script, it pre-sets interactive logic at key operation points in the scene, thereby generating interactive, high-fidelity immersive teaching scene instances.

5. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 1, characterized in that, The intelligent matching and precise push module includes a learner ability profiling unit, a multi-objective optimization matching unit, and a push strategy execution unit. The learner competency profiling unit continuously collects learners' historical behavioral data within the teaching platform, including course completion rate, practical training results, skills assessment results, and interactive operation records in immersive scenarios. It constructs and dynamically updates learners' multi-dimensional competency profile vectors through competency modeling algorithms. The multi-objective optimization matching unit takes the set of demand feature vectors, the teaching knowledge graph, and the learner ability profile vectors as inputs to establish a multi-objective matching optimization model. The first optimization objective of the multi-objective matching optimization model is to maximize the matching degree between the industry demand features covered by the push scenario and the gaps in the learner's ability to be improved. The second optimization objective is to minimize the estimated cognitive load of the learner in order to master the teaching elements required for the scenario. The third optimization objective is to maximize the resource adaptability between the teaching resources required for the scenario and the existing training conditions of the college. The multi-objective optimization matching unit uses a multi-objective evolutionary algorithm to solve the model and outputs a set of Pareto optimal teaching scenario generation scheme sequences. The push strategy execution unit selects the final execution plan from the optimal plan sequence according to the preset push strategy, generates a scene generation instruction containing the target teaching element set and scene complexity parameters, and sends it to the immersive scene generation engine. At the same time, the push strategy execution unit pushes the generated scene to the target learner's terminal or the designated immersive training classroom through the teaching management platform.

6. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 2, characterized in that, The word embedding model used by the demand feature vectorization unit incorporates an industry domain dictionary as prior knowledge during training. This industry domain dictionary is automatically constructed and updated by continuously analyzing high-frequency terms and co-occurrence relationships in industry policy documents and technology white papers.

7. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 4, characterized in that, The pre-built 3D teaching asset library connected to the intelligent 3D asset calling unit adopts a microservice-based distributed architecture for storage and management. In addition to geometric data and texture data, each 3D asset object is associated with a structured attribute description file. The attribute description file defines in detail the interactive components, state variables and acceptable interactive operation instruction set of the asset in the teaching scenario.

8. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 5, characterized in that, In the multi-objective matching optimization model established by the multi-objective optimization matching unit, the gap between the learner's ability profile vector and the industry demand feature vector is defined as the ability gap vector. The matching degree is measured by calculating the cosine similarity between the industry demand vector associated with the teaching elements covered by the scenario and the ability gap vector. Cognitive load prediction is calculated based on the complexity weight of teaching elements and the average time spent on related content of learners' historical learning elements. Resource suitability is calculated by comparing the list of assets required for the scenario with the list of assets in the school's inventory, and taking into account equipment reservation status and venue scheduling constraints.

9. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 5, characterized in that, The push strategy execution unit presets three push strategies: aggressive strategy, robust strategy, and balanced strategy. The aggressive strategy prioritizes the solution with the highest matching degree; the robust strategy prioritizes the solution with the lowest cognitive load prediction. The balancing strategy seeks a balance between matching degree and cognitive load; the push strategy can be set globally by teacher administrators according to the teaching plan or individually specified for specific learner groups.

10. The intelligent push system for immersive generative teaching scenarios oriented towards regional industry needs as described in claim 5, characterized in that, The system also includes a closed-loop evaluation module for teaching effectiveness, which collects real-time operation data of learners in immersive scenarios, assessment scores after scenario completion, and practical evaluations from industry mentors. By comparing the changes in learners' ability profiles before and after scenario learning, the module calculates the teaching effectiveness gain value of scenario push and feeds this gain value back to the intelligent matching and precise push module for dynamically adjusting learners' ability profile vectors and optimizing the weight parameters of the model.

Images