AI-assisted teaching path guidance methods in virtual simulation scenarios

Abstract

This invention discloses an AI-assisted teaching path guidance method in a virtual simulation scenario, relating to the field of AI teaching. First, a virtual simulation teaching scenario is constructed, including a knowledge point system and a dynamic event library. An AI data acquisition module captures multi-dimensional data such as learner operational behavior and cognitive load. Then, a two-dimensional knowledge point association model is constructed, combining current adaptation and potential prediction. A personalized fusion algorithm generates an initial teaching path containing essential basics and optional potential learning paths, intelligently resolving conflicts between custom path fragments. Real-time dynamic optimization of the teaching path is achieved by combining pre-set conditions with hierarchical priorities and dynamic event responses, adapting to differentiated teaching needs for single-person and multi-person collaborative learning. This invention achieves precise dynamic adaptation between the teaching path and learner learning progress, balancing basic achievement with potential ability development in a tiered teaching approach, significantly improving the learning effectiveness of practical courses.

Description

Technical Field

[0001] This invention relates to the field of AI teaching, specifically to a method for guiding AI-assisted teaching paths in a virtual simulation scenario. Background Technology

[0002] In the wave of digital and intelligent development in education, the deep integration of virtual simulation technology and artificial intelligence technology has become an important development direction in the education field. Especially in practical courses in vocational and higher education, virtual simulation, with its immersive, risk-free, and repeatable characteristics, effectively breaks through the limitations of offline practical scenarios. AI assistance provides technical support for personalized teaching. The teaching model combining the two has become a key means to solve the pain points of practical course teaching.

[0003] For practical courses such as mechanical design, medical simulation, and circuit simulation, traditional offline teaching suffers from high equipment costs, significant operational safety risks, difficulty in replicating complex scenarios, and uneven opportunities for learners to gain practical experience. While the application of virtual simulation scenarios has made up for these shortcomings, current virtual simulation teaching for practical courses still lacks a systematic AI-assisted teaching path guidance system. It is difficult to accurately adapt the teaching path according to the individual differences of learners, and thus cannot fully leverage the teaching advantages of virtual simulation and AI technologies.

[0004] Existing AI-assisted teaching technologies in virtual simulation scenarios have certain shortcomings. The teaching paths lack dynamic adaptability and cannot be accurately and timely adjusted according to the learners' real-time learning progress. They can only generate fixed basic teaching paths, without setting graded adjustment conditions and response priorities for changes in learning progress, nor can they optimize the difficulty and content of the path based on the response to dynamic events. This results in a low degree of matching between the teaching path and the learners' real-time learning status, making it difficult to adapt to the cognitive pace and knowledge mastery of different learners. Summary of the Invention

[0005] Based on this, the purpose of this invention is to provide an AI-assisted teaching path guidance method in a virtual simulation scenario, so as to solve the technical problems that there are certain defects in general AI-assisted teaching related technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-assisted teaching path guidance method in a virtual simulation scenario, comprising the following steps: (1) Construct a virtual simulation teaching scenario that includes a knowledge point system, an interactive task library, scene rules and a dynamic event library. The knowledge point system is associated with preset teaching objectives, cognitive levels and potential ability development directions. (2) The AI ​​data acquisition module captures multi-dimensional data of learners in real time, including operational behavior data, knowledge mastery data, cognitive load data, scene interaction data and dynamic event response data; (3) Generate a two-dimensional knowledge point association model based on the knowledge graph construction module, which integrates knowledge point dependency, difficulty coefficient, teaching objective weight and potential adaptation coefficient; (4) The AI ​​path generation module calls the personalized fusion algorithm, inputs the multi-dimensional data and the dual-dimensional knowledge point association model, generates an initial teaching path set containing basic mandatory paths and potential optional paths, and receives learners' custom path fragment requirements and integrates them into an adaptive initial teaching path. (5) During the process of learners executing the initial teaching path, the AI ​​path adjustment module receives updated multi-dimensional data in real time. When the preset adjustment conditions are met or the dynamic event response is triggered, the initial teaching path is dynamically optimized to generate a target teaching path that is adapted to the current learning situation and taps into potential abilities. (6) Output the target teaching path through the feedback interaction module of the virtual simulation scene, provide a path adjustment visualization tool at the same time, collect learners' execution data, feedback scores and path fragment modification suggestions, and send them back to the AI ​​path generation module to complete closed-loop optimization.

[0007] The dynamic event library mentioned in step (1) includes randomly triggered sudden challenge tasks, knowledge transfer application tasks, and collaborative interaction tasks. The trigger probability of the dynamic events is positively correlated with the learner's current knowledge mastery. The trigger interval is dynamically adjusted based on the learning rhythm. Each dynamic event is configured with a difficulty adaptive calibration unit. Based on the learner's historical response data, the knowledge point hiding depth and operation complexity of the event are adjusted in real time. The deviation between the calibrated event difficulty and the current potential adaptation coefficient does not exceed ±10%. The calibration result is synchronously transmitted back to the two-dimensional knowledge point association model to update the potential adaptation coefficient.

[0008] The construction of the dual-dimensional knowledge point association model in step (3) includes: establishing an initial knowledge graph based on subject standards, obtaining knowledge point association weights through learning historical data training, updating the potential adaptation coefficient in real time by combining dynamic event response data, and automatically establishing the potential migration path between a certain knowledge point and cross-chapter or cross-subject related knowledge points when the potential adaptation coefficient of a certain knowledge point is ≥0.5, thereby generating a hierarchical association network of core potential and related potential.

[0009] The personalized fusion algorithm described in step (4) is a combination of the basic adaptation algorithm, the path fragment integration algorithm, and the fragment conflict intelligent reconciliation algorithm. The path fragment requirements include the learner's self-selected order of overcoming difficulties, interest-oriented task types, and learning pace preferences. When different fragment requirements conflict, the fragment conflict intelligent reconciliation algorithm reconciles the conflict by inserting virtual demonstration tasks to aid understanding and splitting complex task nodes. The initial teaching path set contains at least one alternative path, and the optimal path is selected by weighting the learning situation matching score and interest matching score.

[0010] The preset adjustment conditions mentioned in step (5) include: ≥3 consecutive incorrect operations, cognitive load exceeding the adaptation range, knowledge mastery rate <70%, and learning progress deviating from the baseline curve by ±25%. Each condition is set with a linkage priority order, with cognitive load exceeding the standard having the highest priority and triggering immediate adjustment, and progress deviation having the lowest priority and delaying adjustment by 5 minutes. The dynamic event response includes: increasing the path difficulty when the completion rate of the sudden challenge task is ≥90%, and supplementing the basic transition task when the completion rate is <40%. The response result triggers the complementary skill triggering unit: if it is a single-person scenario, it supplements the potential enhancement task; if it is a multi-person scenario, it triggers the complementary collaborative task.

[0011] The feedback interaction module outputs information in the following formats: a visual navigation map within the virtual scene, step-by-step guidance animations, difficulty prompt windows, and voice guidance. The output formats support personalized adaptive switching. Based on a dual-dimensional learning data package, animation and voice guidance are prioritized for learners with high cognitive loads, while navigation maps and concise text guidance are prioritized for learners with a faster pace. The path adjustment visualization tool allows learners to drag knowledge point nodes, add or remove task quantities, and adjust the learning pace. Operation results are synchronized to the AI ​​path adjustment module in real time. The tool also includes a built-in function to verify the rationality of fragmented needs, providing risk warnings and alternative solutions for adjustment requests exceeding 50% of the current cognitive level.

[0012] The AI ​​data acquisition module processes multi-dimensional data: it associates operational behavior, dynamic event response, and other data with knowledge points to generate a two-dimensional learning data package containing the current level and potential tendency. At the same time, it labels the confidence level of the potential tendency data. When the confidence level is lower than 60%, it automatically triggers a supplementary detection task of the dynamic event library to collect more potential-related data to improve the confidence level.

[0013] When the virtual simulation scenario is a multi-person collaborative scenario, the AI ​​path generation module includes a complementary skill triggering unit and a skill complementarity chain generation sub-unit: the complementary skill triggering unit identifies the skill weaknesses and strengths of each learner, and the skill complementarity chain generation sub-unit generates a chain-like collaborative structure of learners with weaknesses, learners with strengths, and collaborators based on the correspondence between weaknesses and strengths. Then, combined with a two-dimensional knowledge point association model, it generates a collaborative teaching path consisting of individual improvement sub-paths, collaborative complementary nodes, and chain-like collaborative tasks. Each collaborative complementary node corresponds to a link in the skill complementarity chain.

[0014] The closed-loop optimization adopts a combination mechanism of node update, periodic iteration and dynamic allocation of feedback weights: the learner's two-dimensional profile is updated after each interactive task unit is completed, and the two-dimensional knowledge point association model is iterated after every four teaching nodes are accumulated; the dynamic allocation mechanism of feedback weights gives learners a higher weight for feedback ratings on potential exploratory tasks than on regular tasks, and a higher weight for suggestions on modifying path fragments than on ordinary feedback. The weight allocation results are used to optimize the parameters of the personalized fusion algorithm.

[0015] The difficulty adjustment of the target teaching path adopts a combination strategy of basic adaptation and potential exploration: the span of regular knowledge points at adjacent difficulty levels does not exceed 25% of the current cognitive level, and the span of potential exploration knowledge points does not exceed 40% of the current cognitive level. Each potential exploration knowledge point is accompanied by a virtual demonstration task to aid understanding. At the same time, if the mastery of potential exploration knowledge points is ≥75%, the upper limit of the span of that knowledge point will be increased by 5% in the next iteration; if the mastery is <50%, the upper limit of the span will be decreased by 5%, and the calibration results will be synchronized to the difficulty adaptive calibration unit of the dynamic event library.

[0016] In summary, the present invention has the following beneficial effects: It constructs a multi-dimensional data collection and dual-dimensional knowledge point association modeling system, generates and intelligently reconciles custom path fragment conflicts through a personalized fusion algorithm, and achieves dynamic optimization of teaching paths by combining hierarchical priority. It also adapts to the differentiated teaching needs of single-person and multi-person collaboration, and completes the continuous iteration of teaching models and paths by relying on a closed-loop optimization mechanism. This achieves precise matching between teaching paths and learners' learning situations, effectively balances the hierarchical teaching objectives of basic achievement and potential ability mining, and significantly improves the learning effect and learner satisfaction of practical courses. Detailed Implementation

[0017] This embodiment discloses an AI-assisted teaching path guidance method in a virtual simulation scenario, which is applicable to practical courses in vocational and higher education, such as mechanical design, medical simulation, and circuit simulation. By constructing dynamically adaptable virtual teaching scenarios, multi-dimensional data collection and analysis, dual-dimensional knowledge point association modeling, personalized path generation and dynamic optimization, it achieves tiered teaching objectives of basic mastery and potential exploration.

[0018] The construction of virtual simulation teaching scenarios is divided into two parts: the construction of core scenario components and the implementation of difficulty adaptive calibration units.

[0019] Scene core component construction Knowledge Point System: Based on the "Fundamentals of Mechanical Design" curriculum standard, a knowledge point system is constructed that includes three major modules: mechanical principles (cognitive levels 1-3), part design (cognitive levels 2-4), and assembly simulation (cognitive levels 3-5). Each knowledge point is associated with preset teaching objectives, such as mastering gear transmission ratio calculation, Bloom's cognitive levels, and potential ability development directions, such as engineering innovation and interdisciplinary applications.

[0020] Interactive Task Library: Includes basic operation tasks, such as gear parameter modeling, comprehensive application tasks, such as reducer assembly, and innovative design tasks, such as lightweight gear optimization. The task types cover both single-person operation and multi-person collaboration.

[0021] Scenario rules: Set virtual laboratory operation standards, such as equipment start-up and shutdown procedures, safety operation thresholds, task scoring criteria, operation accuracy rate of 60%+, completion efficiency of 30%+, innovation points of 10%, path switching permissions, learners can customize fragments, and the system retains the mandatory permissions for core knowledge points.

[0022] Dynamic event library: Stores 3 types of dynamic events: Unexpected challenges: such as an unexpected warning of insufficient material strength during gear modeling, requiring parameter adjustments and redesign; Knowledge transfer application tasks: such as transferring knowledge points about gear transmission to belt transmission design scenarios; Collaborative and interactive tasks: such as working with team members to complete the synchronous assembly of multiple parts of a speed reducer.

[0023] Dynamic event triggering rules: The triggering probability is positively correlated with the level of knowledge mastery. The triggering probability is 60% when the level of mastery is ≥80%, 40% when it is 60%-80%, and 20% when it is <60%. The default triggering interval is 20 minutes, which can be dynamically adjusted according to the learning pace: 10 minutes / time for fast pace and 30 minutes / time for slow pace.

[0024] Difficulty Adaptive Calibration Unit Implementation Each dynamic event is configured with a calibration unit that adjusts event parameters based on the learner's historical response data, including accuracy, completion time, and strategy selection in the last five similar tasks. Knowledge point hiding depth: Initially hide 2 related knowledge points, such as the relationship between module and center distance in gear design. If the historical accuracy rate is ≥80%, increase to 3; if <50%, reduce to 1. Operational complexity: Initially, it includes 5 operation steps, which increase to 7 when the accuracy is ≥85%, and are simplified to 3 when the accuracy is <55%. After calibration, the deviation between the difficulty and the current potential fit coefficient is controlled within ±8%. The calibration results are fed back to the two-dimensional knowledge point association model in real time to update the potential fit coefficient of the corresponding knowledge point. If the learner's potential is found to be higher than expected after calibration, the coefficient increases from 0.45 to 0.52.

[0025] Multi-dimensional data collection and analysis includes AI data collection module configuration and data processing workflow.

[0026] Data acquisition module configuration A three-pronged approach to data collection is adopted, combining sensors, software logs, and interactive feedback. Operational behavior data: Mouse click positions, operation sequence, number of repeated operations, and task completion time are recorded through virtual simulation software logs; Knowledge mastery data is obtained through automatically triggered in-scene quizzes and task performance scoring after each node, such as the accuracy of gear model parameters and interactive Q&A sessions with virtual tutors. Cognitive load data: Through an external eye tracker, fixation point, blink frequency, and heart rate variability can be collected. Combined with the built-in attention monitoring software, the data is converted into quantitative data of 0-100 points using the NASA-TLX scale. Scene interaction data: Records the frequency of interactions with virtual mentors, virtual parts, and team members, such as the number of questions asked, the number of parts dragged and dropped, and the number of collaborative messages sent; Dynamic event response data: Records event completion time, accuracy rate, and adjustment strategies, such as whether prompts were referenced or whether independent innovation was carried out.

[0027] Data processing flow Categorized Records: A database is established based on five categories: operations, knowledge, workload, interaction, and events. Each data entry is labeled with a timestamp and the corresponding knowledge point ID. Association Matching: Machine learning algorithms are used to associate operational behavior data with knowledge points (e.g., associating incorrect gear parameter input with module selection knowledge points), generating a two-dimensional learning data package. Example data package: Current level: Mastery of mechanical principles module 78%, mastery of parts design module 65%; Potential Tendencies: Engineering innovation potential (72% confidence level), interdisciplinary application potential (58% confidence level); Cognitive load: Current score 82 (slightly high); Learning pace: 25 minutes / session (slightly slow). Potential tendency confidence level classification: ≥70% is "high", 50%-70% is "medium", and <50% is "low". When the confidence level is below 60%, such as interdisciplinary application potential of 58%, a supplementary detection task in the dynamic event library is automatically triggered, such as transferring mechanical transmission knowledge points to aerospace component design tasks, collecting more data to improve the confidence level to above 65%. The supplementary detection task matches events with corresponding tags from the dynamic event library according to the potential tendency type with a confidence level below the threshold. Each event in the dynamic event library has a preset potential detection type tag.

[0028] Steps for building a two-dimensional knowledge point association model Step 1: Establish the initial knowledge graph. Based on the subject standards of "Fundamentals of Mechanical Design," use the Neo4j graph database to construct the node (knowledge point) and edge (dependency) structure, for example: Key components: gear drive, belt drive, reducer assembly, and material selection; Dependencies: Gear transmission → Reducer assembly (precursor dependency), Material selection → Gear transmission (constraint dependency); Step 2: Train the association weights of knowledge points. Using historical data from 1000 learners, a collaborative filtering algorithm is used to calculate the association weights (0-1). For example, the association weight between gear transmission and material selection is 0.75 (strong dependency), and the association weight between gear transmission and belt transmission is 0.3 (weak association). Step 3: Update the potential fit coefficient in real time. Combining dynamic event response data, update the coefficient using a gradient descent algorithm ranging from 0.1 to 0.8. For example: The project successfully completed the lightweight gear innovation design project (with an accuracy rate of 85%), increasing the potential fit coefficient of engineering innovation-related knowledge points from 0.4 to 0.62. For events where interdisciplinary transfer was not completed (accuracy rate 35%), the potential fit coefficient for interdisciplinary application of relevant knowledge points will be lowered from 0.35 to 0.28. Step 4: Establish cross-knowledge point potential transfer associations. When the potential fit coefficient of a certain knowledge point is ≥0.5, such as a gear transmission coefficient of 0.62, automatically retrieve related knowledge points across chapters (e.g., Mechanical Principles and Mechanical Manufacturing) and across disciplines (e.g., Mechanical Design and Mechanics of Materials), and establish a hierarchical network of core potential and related potentials: Core potential node: Innovative design of gear transmission (coefficient 0.62); Related potential nodes: Precision parts machining technology (coefficient 0.55), composite material application (coefficient 0.51).

[0029] Model output format The model output is a visualized hierarchical relationship graph, which includes knowledge point IDs, dependencies, association weights, and potential adaptation coefficients, and supports real-time invocation by the AI ​​path generation module.

[0030] Personalized fusion algorithm implementation The algorithm consists of three parts, implemented in Python: Basic Adaptation Algorithm: Input multi-dimensional data and a two-dimensional knowledge point association model, calculate the learning matching score of each knowledge point (60% weight), the formula is: Learning Matching Score = (Knowledge Mastery × 0.4 + Cognitive Load Fit × 0.3 + Operational Accuracy × 0.3) × 100 Path fragment integration algorithm: Receives learner-defined fragment requirements, inputted through an in-scene interactive interface, for example: Prioritize mastering gear module selection; Interest-driven task type: Prefers virtual simulation experiment tasks; Learning pace preference: 20 minutes / node, faster than the baseline of 25 minutes; Fragment Conflict Intelligent Reconciliation Algorithm: When fragment needs conflict, such as the contradiction between a fast-paced preference and a low learning matching score (65 points) in gear module selection, the following reconciliation method is adopted: Insert a virtual demonstration task: call up a 5-minute 3D animation demonstration of the relationship between gear module and transmission ratio; Break down the task nodes: The gear design node is broken down into 10 minutes of modular theory learning, 5 minutes of demonstration viewing, and 5 minutes of practical exercises.

[0031] The fragment conflict intelligent reconciliation algorithm has a built-in conflict judgment rule library, which detects in real time whether there are logical conflicts or teaching adaptability conflicts in the learner's custom path fragment requirements: when the learner's set learning pace is ≤20 minutes or nodes and the current knowledge point mastery is <60%, it is judged as a pace and difficulty conflict; when the association weight between the selected interest task type and the current required path knowledge point is <0.2, it is judged as an interest and required conflict; when the breakthrough order of custom knowledge points violates the pre-dependency relationship clearly defined in the knowledge graph, it is judged as an order and dependency conflict; when the number of tasks added at one time is ≥ When there are 3 or more knowledge points and the current cognitive load is ≥75 points, it is judged as a conflict between quantity and load. When the same knowledge point is studied ≥3 times and the current mastery is ≥80%, it is judged as a conflict between repetition and efficiency. Each conflict type is assigned a priority level of high, medium and low. Among them, the conflict between rhythm and difficulty, interest and necessity, and sequence and dependence is high priority, the conflict between quantity and load is medium priority, and the conflict between repetition and efficiency is low priority. High priority conflicts trigger reconciliation intervention immediately, medium priority conflicts are triggered after a 2-minute delay and a second verification, and low priority conflicts are only logged and prompts are provided to the learner without mandatory intervention.

[0032] The aforementioned fragment conflict intelligent reconciliation algorithm can identify various logical conflicts and workload risks generated by learners in the process of customizing their learning paths in real time. Based on the rule base and decision tree, it automatically generates conflict-free and highly efficient personalized learning paths. Compared with existing technologies, this algorithm not only solves the contradictions between multi-dimensional customization needs, but also achieves flexible adaptive adjustment of teaching paths through hierarchical conflict levels and differentiated reconciliation strategies, avoiding damage to the learning experience due to forced intervention.

[0033] Initial teaching path set generation and selection Generate 2-4 alternative paths, each containing a basic mandatory path and potential optional paths: The essential path, which covers core knowledge points and cannot be skipped, includes: Gear Transmission Fundamentals, Parameter Calculation, and Model Building. Potential pathways can be selected based on the potential fit coefficient: engineering innovation direction, namely gear lightweight design, or interdisciplinary application direction, namely composite material gear selection; Calculate the weighted score of learning matching (60%) and interest alignment (40%) to select the optimal path: Example: Path 1 (learning matching score 82, interest fit score 75) yields a weighted score of 82 × 0.6 + 75 × 0.4 = 79.2; Path 2 (learning matching score 78, interest alignment score 85) yields a weighted score of 78 × 0.6 + 85 × 0.4 = 80.8; Ultimately, path 2 was chosen as the initial teaching path.

[0034] Special handling for multi-person collaboration scenarios When the scenario involves collaboration among three people, such as learners A, B, and C: Complementary Skill Trigger Unit Identification: A. Strength = Theoretical Calculation (Score 90), Weakness = Practical Skills (Score 60); B. Strength = Practical Skills (Score 85), Weakness = Theory (Score 65); C. Strength = Coordination and Communication (Score 80), Weakness = Comprehensive Application (Score 70). Skill complementarity chain generation: A is responsible for parameter calculation, B is responsible for model building, and C is responsible for progress coordination and problem feedback; Collaborative teaching path generation: Individual improvement sub-path: A's practical training tasks, B's theoretical reinforcement tasks, and C's comprehensive application tasks; Collaborative and complementary nodes: parameter exchange nodes, where A submits calculation results to B; and construction feedback nodes, where B reports construction progress to C. Chain-based collaborative task: jointly complete the assembly and debugging of the reducer. A calculates the assembly parameters, B executes the assembly operation, and C coordinates the debugging process.

[0035] Dynamic optimization of teaching paths Preset adjustment condition monitoring and response The AI ​​path adjustment module monitors updated multi-dimensional data in real time, sorts them by linkage priority, and triggers adjustment when the following conditions are met: Priority 1 triggers immediate adjustment: Cognitive load exceeds the adaptation range (adaptation range 60-80 points, actual test 88 points). Adjustment method: split the current reducer debugging task into 2 small nodes and insert a 5-minute rest interaction, such as equipment maintenance demonstration in a virtual scene. Priority 2: ≥3 consecutive incorrect operations, such as entering the wrong gear module 3 times in a row. Adjustment method: pop up a knowledge review window, module selection rules, and reduce the complexity of subsequent operation steps; Priority 3: Knowledge mastery rate <70%, only 3 out of 5 required knowledge points are mastered, with a pass rate of 60%. Adjustment method: Supplement 2 basic consolidation tasks, modular calculation practice questions, and virtual tutor Q&A. Priority 4 trigger delay of 5 minutes adjustment: The learning progress deviates from the baseline curve by ±25%. The baseline is 120 minutes to complete, but the actual time is 90 minutes. Adjustment method: Add one extension task to the potential optional path, such as gear optimization design.

[0036] Dynamic event response mechanism The completion rate of sudden challenge tasks is ≥90%, such as the completion rate of emergency optimization of gear strength is 92%. Then the difficulty of the path is increased: the knowledge point hiding depth of the subsequent belt drive design task is increased from 2 to 3. If the completion rate of an unexpected challenge task is less than 40%, such as the completion rate of interdisciplinary transfer tasks being 35%, a supplementary basic transition task will be assigned: review the knowledge points related to mechanical transmission and material properties. Complementary skills trigger: In single-player scenarios, tasks that enhance potential are supplemented, such as expanding innovative gear design; in multi-player scenarios, collaborative tasks are triggered, such as A providing theoretical support and B optimizing practical solutions.

[0037] Difficulty Adjustment Strategy Basic adaptation section: The span of common knowledge points between adjacent difficulty levels is ≤25% of the current cognitive level (70% of the current cognitive level, span ≤17.5%), for example, from standard gear design (difficulty 60) to modified gear design (difficulty 72, span 12%). Potential Exploration Section: The knowledge points span ≤ 40% of the current cognitive level, such as "from modified gear design" to lightweight modified gear design, with a difficulty level of 90 and a span of 25%, accompanied by a virtual demonstration task; Effect retrospective calibration: If the mastery of potential exploration knowledge points is ≥75%, the upper limit of the span in the next iteration will be increased by 5%; If the mastery level is less than 50%, the upper limit of the span will be reduced by 5%; The calibration results are synchronized to the difficulty-adaptive calibration unit in the dynamic event library to adjust the knowledge point hiding depth of subsequent events.

[0038] Feedback Interaction and Closed-Loop Optimization Implementation of feedback interaction module Output format and personalized switching: Based on a two-dimensional learning data package, such as a cognitive load of 82 points and a slow pace, a dual-mode guidance of animation and voice is used. Visual navigation map: A distribution map of task nodes within the virtual laboratory scene, marking the current location, incomplete nodes, and potential nodes; Step-by-step guided animation: a disassembly animation of gear assembly; Voice prompt: Please confirm that the gear module is 2.5 before calculating the addendum circle diameter; If the learner's cognitive load drops to 70 points and the pace speeds up (20 minutes / node), the system will automatically switch to a navigation map and concise text guidance, with the text retaining only the core operation steps.

[0039] Path adjustment visualization tool: Uses a drag-and-drop interactive interface and supports: Drag the knowledge point node, such as moving the belt drive design node before the gear drive; Increase or decrease the number of tasks (e.g., add one troubleshooting subtask to the practical tasks). Adjust the learning pace by setting it to 15-30 minutes per session using the slider. The tool has a built-in rationality check function: If a learner attempts to add the application of quantum mechanics in mechanical design, which exceeds 60% of their current cognitive level, a risk warning will pop up: This task exceeds the current cognitive level. It is recommended to replace it with the application of materials mechanics in gear design. Are you sure?

[0040] Closed-loop optimization mechanism Node Update: Upon completion of each teaching node, such as gear module selection, the dual-dimensional profile is updated, with knowledge mastery increasing from 65% to 78% and potential fit coefficient increasing from 0.45 to 0.53; Cyclic Iteration: Every 4 teaching nodes, the two-dimensional knowledge point association model is iterated, with an iteration cycle of 18 minutes. The association weights are adjusted, such as increasing the association weight between gear design and material selection from 0.75 to 0.82. Dynamic allocation of feedback weights: The feedback score for potential exploratory tasks accounts for 40% of the evaluation. The feedback score for routine tasks has a weighting of 20%. Path fragment modification suggestions have a weight of 30%; Normal feedback has a weighting of 10%; The weighting results are used to optimize the parameters of the personalized fusion algorithm.

[0041] Taking the virtual simulation teaching of the mechanical design fundamentals course in a vocational college as an example, the subjects were 30 second-year students, divided into 5 groups (6 people in each group, including 1 group of single-person scenario control group). The implementation process is as follows: Before class, a virtual simulation teaching scenario is constructed, containing 12 core knowledge points, 3 types of dynamic event libraries, and an interactive task library; During class: 0-5 minutes: The AI ​​data acquisition module initializes, collects basic learner data, conducts pre-knowledge tests, and establishes a cognitive load baseline; 5-15 minutes: Generate a two-dimensional knowledge point association model, output the initial teaching path, 2-4 alternative paths per group, and select the best one; 15-85 minutes: Learners execute the initial path, triggering a dynamic event every 20 minutes. The AI ​​path adjustment module optimizes the path in real time. For example, 3 groups are adjusted immediately due to excessive cognitive load, and 2 groups are adjusted later due to too fast progress. 85-95 minutes: Collect execution data, feedback scores, and modification suggestions through the feedback interaction module, such as increasing the frequency of collaborative tasks; After class: The closed-loop optimization module updates the dual-dimensional profile and related models to generate an adapted path for the next lesson.

[0042] Results: Learners' mastery of knowledge points increased by an average of 32%, compared to a 20% increase in the control group; the rate of achieving the target for potential ability development reached 75%; and the learning satisfaction rate reached 88%.

[0043] Although embodiments of the present invention have been shown and described, these specific embodiments are merely explanations of the invention and are not intended to limit it. The specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. After reading this specification, those skilled in the art may make modifications, substitutions, and variations to the embodiments as needed without departing from the principles and spirit of the invention, but such modifications, substitutions, and variations are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for guiding AI-assisted teaching paths in a virtual simulation scenario, characterized in that, Includes the following steps: (1) Construct a virtual simulation teaching scenario that includes a knowledge point system, an interactive task library, scene rules and a dynamic event library. The knowledge point system is associated with preset teaching objectives, cognitive levels and potential ability development directions. (2) The AI ​​data acquisition module captures multi-dimensional data of learners in real time, including operational behavior data, knowledge mastery data, cognitive load data, scene interaction data and dynamic event response data; (3) Generate a two-dimensional knowledge point association model based on the knowledge graph construction module, which integrates knowledge point dependency, difficulty coefficient, teaching objective weight and potential adaptation coefficient; (4) The AI ​​path generation module calls the personalized fusion algorithm, inputs the multi-dimensional data and the dual-dimensional knowledge point association model, generates an initial teaching path set containing basic mandatory paths and potential optional paths, and receives learners' custom path fragment requirements and integrates them into an adaptive initial teaching path. (5) During the process of learners executing the initial teaching path, the AI ​​path adjustment module receives updated multi-dimensional data in real time. When the preset adjustment conditions are met or the dynamic event response is triggered, the initial teaching path is dynamically optimized to generate a target teaching path that is adapted to the current learning situation and taps into potential abilities. (6) Output the target teaching path through the feedback interaction module of the virtual simulation scene, provide a path adjustment visualization tool at the same time, collect learners' execution data, feedback scores and path fragment modification suggestions, and send them back to the AI ​​path generation module to complete closed-loop optimization.

2. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The dynamic event library mentioned in step (1) includes randomly triggered sudden challenge tasks, knowledge transfer application tasks, and collaborative interaction tasks. The trigger probability of the dynamic events is positively correlated with the learner's current knowledge mastery. The trigger interval is dynamically adjusted based on the learning rhythm. Each dynamic event is configured with a difficulty adaptive calibration unit. Based on the learner's historical response data, the knowledge point hiding depth and operation complexity of the event are adjusted in real time. The deviation between the calibrated event difficulty and the current potential adaptation coefficient does not exceed ±10%. The calibration result is synchronously transmitted back to the two-dimensional knowledge point association model to update the potential adaptation coefficient.

3. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The construction of the dual-dimensional knowledge point association model in step (3) includes: establishing an initial knowledge graph based on subject standards, obtaining knowledge point association weights through learning historical data training, updating the potential adaptation coefficient in real time by combining dynamic event response data, and automatically establishing the potential migration path between a certain knowledge point and cross-chapter or cross-subject related knowledge points when the potential adaptation coefficient of a certain knowledge point is ≥0.5, thereby generating a hierarchical association network of core potential and related potential.

4. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The personalized fusion algorithm described in step (4) is a combination of the basic adaptation algorithm, the path fragment integration algorithm, and the fragment conflict intelligent reconciliation algorithm. The path fragment requirements include the learner's self-selected order of overcoming difficulties, interest-oriented task types, and learning pace preferences. When different fragment requirements conflict, the fragment conflict intelligent reconciliation algorithm reconciles the conflict by inserting virtual demonstration tasks to aid understanding and splitting complex task nodes. The initial teaching path set contains at least one alternative path, and the optimal path is selected by weighting the learning situation matching score and interest matching score.

5. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The preset adjustment conditions mentioned in step (5) include: ≥3 consecutive incorrect operations, cognitive load exceeding the adaptation range, knowledge mastery rate <70%, and learning progress deviating from the baseline curve by ±25%. Each condition is set with a linkage priority order, with cognitive load exceeding the standard having the highest priority and triggering immediate adjustment, and progress deviation having the lowest priority and delaying adjustment by 5 minutes. The dynamic event response includes: increasing the path difficulty when the completion rate of the sudden challenge task is ≥90%, and supplementing the basic transition task when the completion rate is <40%. The response result triggers the complementary skill triggering unit: if it is a single-person scenario, it supplements the potential enhancement task; if it is a multi-person scenario, it triggers the complementary collaborative task.

6. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The feedback interaction module outputs information in the following formats: a visual navigation map within the virtual scene, step-by-step guidance animations, difficulty prompt windows, and voice guidance. The output formats support personalized adaptive switching. Based on a dual-dimensional learning data package, animation and voice guidance are prioritized for learners with high cognitive loads, while navigation maps and concise text guidance are prioritized for learners with a faster pace. The path adjustment visualization tool allows learners to drag knowledge point nodes, add or remove task quantities, and adjust the learning pace. Operation results are synchronized to the AI ​​path adjustment module in real time. The tool also includes a built-in function to verify the rationality of fragmented needs, providing risk warnings and alternative solutions for adjustment requests exceeding 50% of the current cognitive level.

7. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The AI ​​data acquisition module processes multi-dimensional data: it associates operational behavior, dynamic event response, and other data with knowledge points to generate a two-dimensional learning data package containing the current level and potential tendency. At the same time, it labels the confidence level of the potential tendency data. When the confidence level is lower than 60%, it automatically triggers a supplementary detection task of the dynamic event library to collect more potential-related data to improve the confidence level.

8. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: When the virtual simulation scenario is a multi-person collaborative scenario, the AI ​​path generation module includes a complementary skill triggering unit and a skill complementarity chain generation sub-unit: the complementary skill triggering unit identifies the skill weaknesses and strengths of each learner, and the skill complementarity chain generation sub-unit generates a chain-like collaborative structure of learners with weaknesses, learners with strengths, and collaborators based on the correspondence between weaknesses and strengths. Then, combined with a two-dimensional knowledge point association model, it generates a collaborative teaching path consisting of individual improvement sub-paths, collaborative complementary nodes, and chain-like collaborative tasks. Each collaborative complementary node corresponds to a link in the skill complementarity chain.

9. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 1, characterized in that: The closed-loop optimization adopts a combination mechanism of node update, periodic iteration and dynamic allocation of feedback weights: the learner's two-dimensional profile is updated after each interactive task unit is completed, and the two-dimensional knowledge point association model is iterated after every four teaching nodes are accumulated; the dynamic allocation mechanism of feedback weights gives learners a higher weight for feedback ratings on potential exploratory tasks than on regular tasks, and a higher weight for suggestions on modifying path fragments than on ordinary feedback. The weight allocation results are used to optimize the parameters of the personalized fusion algorithm.

10. The AI-assisted teaching path guidance method in a virtual simulation scenario according to claim 5, characterized in that: The difficulty adjustment of the target teaching path adopts a combination strategy of basic adaptation and potential exploration: the span of regular knowledge points at adjacent difficulty levels does not exceed 25% of the current cognitive level, and the span of potential exploration knowledge points does not exceed 40% of the current cognitive level. Each potential exploration knowledge point is accompanied by a virtual demonstration task to aid understanding. At the same time, if the mastery of potential exploration knowledge points is ≥75%, the upper limit of the span of that knowledge point will be increased by 5% in the next iteration; if the mastery is <50%, the upper limit of the span will be decreased by 5%, and the calibration results will be synchronized to the difficulty adaptive calibration unit of the dynamic event library.