A university student career scene simulation training evaluation system

Abstract

The present application relates to the technical field of professional simulation training evaluation, and discloses a kind of for university student professional scene simulation training evaluation system.The system includes task analysis, dynamic behavior modeling, multi-modal interaction management, state evolution tracking, strategy adaptation and comprehensive evaluation generation unit.The system receives initial scene description and generates structured task framework, configures behavior parameters for student role;Through real-time fusion alignment of voice, action and gaze data, a multi-modal interaction atlas is constructed;Further, key decision nodes are identified, and team state evolution paths are dynamically tracked;According to the evolution sequence, the strategy in the behavior rule base is matched and adjusted, and the adaptive intervention scheme is output;Finally, a multi-dimensional structured evaluation report is generated.The system realizes the deep insight into the complex interactive behavior in the simulation training process and the dynamic and objective intelligent evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of professional simulation training evaluation, in particular to a university student professional scene simulation training evaluation system. BACKGROUND

[0002] At present, the evaluation of university student professional scene simulation training mainly relies on on-site observation by instructors or post-event video playback. This evaluation method is highly dependent on personal experience and has strong subjectivity, and it is difficult to systematically analyze dynamic and complex interactive behaviors in team collaboration. Existing technologies usually only focus on the final task results or explicit language communication, and cannot effectively capture and analyze non-verbal cues and their role in team dynamics, resulting in single evaluation dimensions and insufficient depth and accuracy of feedback.

[0003] In terms of technical application, some systems attempt to introduce speech recognition or motion capture devices, but various types of data are usually processed independently, lacking effective integration mechanisms. Multimodal data is in isolation, and accurate temporal correlation and feature-level fusion cannot be established. This data fragmentation prevents the system from restoring the true full picture of the interaction scene, severely restricting the ability to objectively and meticulously analyze the team collaboration process.

[0004] In terms of analysis models, existing automated evaluation methods are mostly based on predetermined scoring rules to perform static evaluation on results or key nodes. This type of method cannot continuously track how individual behaviors in the training process trigger dynamic evolution of team states, and lacks identification and characterization of "decision-impact" causal paths. Due to the inability to understand the evolution logic of team internal states, evaluation results often remain at the level of phenomenon description, making it difficult to reveal the underlying reasons affecting team effectiveness, and even more difficult to provide adaptive intervention guidance based on real-time dynamics during simulation. Existing technologies have limitations in dynamic analysis and depth of insight. SUMMARY

[0005] The purpose of the present application is to provide a university student professional scene simulation training evaluation system to solve the problems raised in the background art.

[0006] To achieve the above purpose, the present application provides a university student professional scene simulation training evaluation system, which comprises:

[0007] A task analysis unit for receiving an initial professional scene description, performing semantic decomposition on the initial professional scene description, extracting scene core task elements and role constraint conditions, and generating a structured task framework;

[0008] A dynamic behavior modeling unit for importing a preset student role configuration based on the structured task framework, assigning initial behavior parameters to each student role configuration, and constructing a dynamic behavior rule library based on historical training data.

[0009] a multi-modal interaction management unit configured to capture voice interaction data, action sequence data and gaze focus data between students in real time during a simulation training process, timestamp-align and feature-fuse the voice interaction data, action sequence data and gaze focus data, and generate a multi-modal interaction graph;

[0010] a state evolution tracking unit configured to identify a decision node of an individual student according to the multi-modal interaction graph, track a team state change path triggered by the decision node, and record a state evolution sequence;

[0011] a strategy adaptation unit configured to match a candidate strategy in the dynamic behavior rule library based on the state evolution sequence, adjust a weight of the candidate strategy according to a real-time training environment parameter, and output an adapted intervention strategy;

[0012] a comprehensive evaluation generation unit configured to integrate the state evolution sequence and the adapted intervention strategy, calculate a performance consistency index of an individual and a team in multiple dimensions, and generate a structured evaluation report.

[0013] Preferably, the semantic decomposition of the initial professional scene description comprises:

[0014] adopting a hierarchical semantic parsing method to split the initial professional scene description into a scene background section, a task objective section and a constraint condition section;

[0015] performing entity recognition and relationship extraction on the scene background section to establish a scene entity relationship network;

[0016] performing action verb analysis and logical dependency parsing on the task objective section to construct a task objective tree;

[0017] performing rule clause classification and priority assignment on the constraint condition section to form a constraint rule set;

[0018] associating and mapping the scene entity relationship network, the task objective tree and the constraint rule set to generate the structured task framework.

[0019] Preferably, the construction of the dynamic behavior rule library based on historical training data comprises:

[0020] extracting a typical behavior pattern segment from historical training logs, performing cluster analysis on the typical behavior pattern segment to form a behavior pattern template;

[0021] performing condition-action pair analysis on each behavior pattern template to identify a trigger condition set and a corresponding action set;

[0022] The trigger condition set and the corresponding action set are described in a state machine representation, and a standardized behavior rule is generated;

[0023] The standardized behavior rule is indexed according to the training scene type, and the dynamic behavior rule library is constructed.

[0024] Preferably, the timestamp alignment and feature fusion of the voice interaction data, action sequence data and gaze focus data include:

[0025] Voice content features, tone features and speech rate features are extracted for the voice interaction data;

[0026] Joint motion trajectory features, action amplitude features and action frequency features are extracted for the action sequence data;

[0027] Gaze target features, gaze duration features and saccade path features are extracted for the gaze focus data;

[0028] A unified time axis is established, and the voice content features, tone features, speech rate features, joint motion trajectory features, action amplitude features, action frequency features, gaze target features, gaze duration features and saccade path features are aligned according to the time window;

[0029] The aligned multi-source features are weighted and fused using an attention mechanism to generate the multi-modal interaction graph.

[0030] Preferably, the team state change path triggered by the decision node includes:

[0031] The decision subject and the decision content are marked at the decision node;

[0032] The decision content is simulated to deduce the reaction behavior of other students at the subsequent time step;

[0033] The actual occurrence of the reaction behavior is verified according to the multi-modal interaction graph, and the deviation between the actual behavior and the expected behavior is recorded;

[0034] The causal relationship chain between the decision node, the actual reaction behavior and the deviation is recorded in the form of a graph structure to form the state evolution sequence.

[0035] Preferably, the weight adjustment of the candidate strategy according to the real-time training environment parameters includes:

[0036] Real-time monitoring of task progress indicators and team collaboration indicators in the training scene;

[0037] The task progress indicators are compared with the preset milestones to calculate the progress deviation;

[0038] The team collaboration indicator is compared with an ideal collaboration model to calculate a collaboration efficiency value;

[0039] According to the progress deviation degree and the collaboration efficiency value, the activation threshold and the execution priority of the candidate strategy are dynamically adjusted;

[0040] Based on the adjusted activation threshold and execution priority, the adaptive intervention strategy is selected from the candidate strategy.

[0041] Preferably, the performance consistency indicator of the individual and the team in multiple dimensions includes:

[0042] Define the task execution dimension, the communication coordination dimension and the emergency response dimension;

[0043] Under the task execution dimension, the matching coefficient of individual task completion degree and team task synchronization degree is calculated;

[0044] Under the communication coordination dimension, the correlation index of individual communication initiative and team information flow efficiency is analyzed;

[0045] Under the emergency response dimension, the coordination ratio of individual decision speed and team crisis recovery speed is evaluated;

[0046] The matching coefficient, the correlation index and the coordination ratio are normalized and weighted aggregated to generate the performance consistency indicator.

[0047] Preferably, the association mapping of the scene entity relationship network, the task target tree and the constraint rule set includes:

[0048] Identify the key entity nodes in the scene entity relationship network, and perform attribute matching between the key entity nodes and the leaf nodes of the task target tree;

[0049] Establish the constraint binding relationship between the constraint rule set and the branch nodes in the task target tree;

[0050] When attribute matching conflict or constraint binding conflict is detected, start conflict resolution protocol, and reassign entity attributes or adjust constraint priority;

[0051] After conflict resolution is completed, the structured task framework containing complete association relationship of entity-task-constraint is generated.

[0052] Preferably, the weighted fusion of the aligned multi-source features by using attention mechanism includes:

[0053] Calculate the relevance score of each feature dimension with the core task of the current training stage;

[0054] distribute initial attention weights according to the correlation scores;

[0055] contextually correct the initial attention weights by introducing context information in the dynamic behavior rule base;

[0056] generate a fusion feature vector by weighted sum of corresponding features using the corrected attention weights;

[0057] organize the fusion feature vector into graph nodes in time sequence and connect edges according to the time sequence dependency between features to construct the multi-modal interaction graph.

[0058] Preferably, recording the causal relationship chain between the decision nodes, the actual reaction behaviors and the deviations in the form of a graph structure comprises:

[0059] taking each decision node as a root node of a causal graph;

[0060] connecting the actual reaction behaviors as child nodes to the corresponding root nodes and marking the behavior types and time stamps;

[0061] annotating the deviations as edge attributes on the edges connecting the root nodes and the child nodes, and the annotation content includes the deviation amount and the deviation type;

[0062] when multiple decision nodes have time sequence or logical association, establishing cross-node connection edges between the corresponding root nodes to form the state evolution sequence describing the state propagation path.

[0063] Compared with the prior art, the present application has the following beneficial effects:

[0064] By accurately timestamping and feature fusion of voice, action and gaze data, a unified multi-modal interaction graph is generated, solving the problem of data fragmentation in traditional evaluation. This technology correlates behavior signals of different modalities in a unified time sequence framework, and can restore the collaborative relationship between elements such as speech content, corresponding gestures and eye contact. This deep fusion reveals the interaction patterns that isolated data cannot present. It improves the dimension and accuracy of behavior capture, enabling the evaluation of individual communication and collaboration skills to be based on objective and comprehensive data, enhancing the objectivity and subtlety of the evaluation.

[0065] By automatically identifying key decision nodes from the interaction flow and tracing the team state evolution paths triggered by them, dynamic causal analysis of team collaboration process is achieved. This technology can depict the complete chain of how a specific proposal or behavior gradually influences the team discussion direction, task progress and cooperation atmosphere. Based on this evolution sequence, the system can understand the driving mechanism of the dynamic changes within the team. This makes the evaluation mechanism change from static result judgment to process dynamic insight, which can accurately identify key behaviors that effectively promote the process or lead to low efficiency, so that the feedback information has a profound explanatory and guiding nature.

[0066] According to the state evolution sequence, the intervention strategy is dynamically matched and adjusted, realizing the closed-loop linkage of evaluation and intervention. The system can select the candidate strategy that matches the situation from the rule library according to the actual evolution trend of the team state and optimize the real-time weight, outputting a highly adaptive intervention scheme. This makes the system not only provide evaluation after training, but also provide formative intervention based on team dynamics in the process. It enhances the intelligence and real-time guidance ability of the training system, making the simulation environment become a learning scene that can adaptively respond and guide team interaction, rather than a simple examination tool. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 The working principle diagram of the university student professional scene simulation training evaluation system described in the present application;

[0068] Figure 2 The flowchart for semantic decomposition work;

[0069] Figure 3 The flowchart for constructing a dynamic behavior rule library;

[0070] Figure 4 The average deviation distribution statistical chart of each decision node;

[0071] Figure 5 The parameter adjustment and strategy effect correlation analysis chart. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0073] Please refer to Figure 1The application provides a university student professional scene simulation training evaluation system, which comprises the following steps: after receiving an initial professional scene description, a task analysis unit performs semantic decomposition on the description, extracts scene core task elements and role constraint conditions, and generates a structured task framework. A dynamic behavior modeling unit imports student role configurations according to the structured task framework, assigns initial behavior parameters to each role, and constructs a dynamic behavior rule library based on historical training data. A multi-modal interaction management unit captures student speech interaction data, action sequence data and visual focus data in real time during the simulation training process, timestamps aligns and fuses these data, and generates a multi-modal interaction graph. A state evolution tracking unit identifies individual student decision nodes according to the multi-modal interaction graph, traces the team state change path triggered by the decision nodes, and records the state evolution sequence. A strategy adaptation unit matches candidate strategies in the dynamic behavior rule library based on the state evolution sequence, adjusts the weight of the candidate strategies according to the real-time training environment parameters, and outputs an adaptive intervention strategy. A comprehensive evaluation generation unit integrates the state evolution sequence and the adaptive intervention strategy, calculates the performance consistency index of individuals and teams in multiple dimensions, and generates a structured evaluation report.

[0074] Embodiment 1: see Figure 2 In the specific embodiment, the task analysis unit performs semantic decomposition on the initial professional scene description, adopts a hierarchical semantic analysis method to split the initial professional scene description into a scene background section, a task target section and a constraint condition section, and the hierarchical semantic analysis method is divided according to the natural paragraph structure and logical connection words of the text. Entity recognition and relationship extraction are performed on the scene background section, entity recognition adopts named entity recognition technology to identify entity categories such as personnel, organization, location and tool, relationship extraction determines the action, affiliation or space-time relationship between entities through dependency syntax analysis, establishes a scene entity relationship network, and the scene entity relationship network is stored in a graph structure, the node represents an entity and is attached with an attribute label, and the edge represents a relationship and is labeled with a relationship type. Action verb analysis and logical dependency analysis are performed on the task target section, action verb analysis extracts core predicate verbs from sentences and classifies them, logical dependency analysis determines the order relationship, conditional relationship or parallel relationship between subtasks by analyzing conjunctions and word order, and a task target tree is constructed, the task target tree adopts a tree data structure, the root node represents the total task target, the child nodes represent the decomposed subtasks, and the connection edges between the nodes are labeled with dependency types such as “precondition” or “synchronous execution”. Rule clause classification and priority assignment are performed on the constraint condition section, rule clause classification is divided into time constraints, resource constraints, role constraints and other types according to the constraint content, priority assignment is numerically assigned based on the importance degree of the constraint clause and the severity of the consequences of violation, a constraint rule set is formed, and the constraint rule set is stored in a list form, each rule contains a constraint type, an applicable object, a constraint condition and a priority value.

[0075] In a specific implementation, the scene entity relationship network, the task target tree, and the constraint rule set are associated and mapped, key entity nodes in the scene entity relationship network are identified, the key entity nodes refer to entities directly related to the task target implementation or having important attributes, the key entity nodes are attribute-matched with leaf nodes of the task target tree, the attribute-matching process is to compare whether the attribute labels of the entity nodes and the task requirements of the leaf nodes are consistent, for example, the "skill level" attribute of the entity node is compared with the "required skill" requirement of the leaf node. The constraint rule set and the constraint binding relationship of the branch nodes in the task target tree are established, the constraint binding relationship is to associate each rule in the constraint rule set to the branch node in the task target tree affected by the rule, for example, a time constraint rule is bound to a branch node with a deadline requirement. When attribute matching conflicts or constraint binding conflicts are detected, a conflict resolution protocol is started, the attribute matching conflict refers to a case where the entity attribute cannot meet the task requirement, the constraint binding conflict refers to a case where multiple constraint rules produce contradictory requirements on the same branch node, the conflict resolution protocol solves the conflict by reallocating entity attributes or adjusting constraint priorities, reallocating entity attributes may involve modifying the role or ability assignment of the entity, adjusting the constraint priority is to reevaluate the importance order of the constraint rules according to the overall task target. After the conflict resolution is completed, a structured task framework containing complete association relationships of entities-tasks-constraints is generated, the structured task framework integrates all association relationships between the scene entity relationship network, the task target tree, and the constraint rule set using a unified data model.

[0076] In some embodiments, the specific operations of the hierarchical semantic parsing method include sentence segmentation and word segmentation processing of the initial professional scene description text, matching the text paragraph structure using a predefined scene template library, the scene template library contains typical paragraph structure modes of various professional scenes. The entity recognition and relationship extraction process uses a deep learning-based sequence labeling model, which is trained on a large amount of labeled text and can accurately identify specific entity types and relationship patterns in the professional scene. Action verb analysis and logical dependency parsing use semantic role labeling technology to label semantic roles such as agents, patients, and times of verbs in sentences, thereby inferring the logical relationship between actions. Rule clause classification and priority assignment rely on a predefined constraint classification system and a weight allocation algorithm based on rule importance, the weight allocation algorithm considers the authority of the rule source and the potential impact of rule violation.

[0077] In some embodiments, the association mapping process is implemented by a mapping engine that traverses all entity nodes in the scene entity relationship network, calculates semantic similarity with task target tree leaf nodes, and the semantic similarity calculation is based on a word vector model and a domain ontology library. The constraint binding relationship establishment uses a rule inference machine that dynamically establishes a binding relationship according to the condition part of the constraint rule matching the attributes of the task target tree node. The conflict resolution protocol adopts a multi-round negotiation mechanism, the first round detects the conflict type, the second round selects different resolution strategies according to the conflict type, and the third round executes the resolution operation and verifies the resolution result. The structured task framework is serialized in JSON or XML format for storage, which is convenient for other units in the system to parse and use.

[0078] Optionally, the construction of the scene entity relationship network can introduce a domain ontology to enhance the semantic expression ability of entities and relationships, and the domain ontology defines the hierarchical relationship and attribute constraints between entity types in the professional scene. The construction of the task target tree can adopt an iterative refinement method, first constructing a coarse-grained task hierarchy, and then gradually decomposing into more fine-grained leaf nodes. The management of the constraint rule set can implement version control function, record the modification history of the constraint rule and support backtracking view.

[0079] Optionally, the attribute matching process can introduce a fuzzy matching algorithm to handle incomplete attribute requirements, and the fuzzy matching algorithm calculates the similarity score between the attribute value and the required value, and sets a threshold to determine whether to match. The constraint binding relationship can support dynamic adjustment, and when the environment changes during task execution, the system can recompute the binding relationship between the constraint rule and the task node. The conflict resolution protocol can integrate a machine learning component to optimize the selection of resolution strategies by analyzing historical conflict resolution cases.

[0080] It can be understood that the hierarchical semantic parsing method can adapt to professional scene descriptions of different complexity, and the modular processing flow ensures the comprehensiveness and accuracy of semantic decomposition. The association mapping of the scene entity relationship network, the task target tree and the constraint rule set creates a unified task representation form, providing a structured basis for subsequent simulation training and evaluation. The design of the conflict resolution protocol ensures the robustness of the system in dealing with inherent uncertainties and contradictions in real-world scenarios. It can be understood that the implementation of the task parsing unit fully considers the characteristics of professional scene simulation training, and realizes the conversion from text description to computable task model by combining natural language processing technology and domain knowledge. The structured task framework not only contains the logical structure of the task itself, but also integrates the information of the execution environment and the constraint conditions.

[0081] Embodiment 2: see Figure 3The dynamic behavior modeling unit constructs a dynamic behavior rule base based on historical training data, which involves extracting typical behavior pattern segments from historical training logs, performing clustering analysis on the typical behavior pattern segments to form behavior pattern templates, performing condition-action pair analysis on each behavior pattern template to identify a set of trigger conditions and a corresponding set of actions, using state machine representation to formally describe the set of trigger conditions and the corresponding set of actions to generate standardized behavior rules, and indexing and arranging the standardized behavior rules according to training scenario types to construct the dynamic behavior rule base. The historical training logs are derived from the complete records of previous university student professional scenario simulation training, including timestamp, student identity, interaction events and behavior results, etc. The typical behavior pattern segments are extracted from the historical training logs by sliding window technology, and the window size is dynamically adjusted according to the behavior duration to ensure that the segment contains a complete behavior sequence. The clustering analysis uses unsupervised learning algorithm to group similar behavior patterns, and the behavior pattern template is stored in the form of prototype vector, each template represents a common behavior such as conflict resolution or resource negotiation. The condition-action pair analysis uses natural language processing tools to extract key elements from the template, the set of trigger conditions includes environmental variables or event thresholds, and the corresponding set of actions includes student response operations. The state machine representation describes the behavior rule in the form of finite state machine model, the state node represents the behavior stage, and the transition edge represents the state change after the condition is triggered. The indexing and arrangement are based on the classification system of training scenario types such as business negotiation or project management, and the dynamic behavior rule base is finally realized in the form of relational database table, supporting efficient query and update.

[0082] In a specific implementation, typical behavior pattern segments are extracted from historical training logs, which are stored in a structured format containing fields such as session identification, participant role, action type, timestamp and context parameters. The extraction of typical behavior pattern segments is achieved by defining behavior boundary conditions, which are based on event intervals or semantic completeness, for example, when the time difference between consecutive actions exceeds a threshold or the behavior goal switches. The extraction process uses a sequence segmentation algorithm, which scans the historical training logs to identify consecutive record segments that meet the boundary conditions. Each typical behavior pattern segment contains a start time, an end time, a list of participating students and a sequence of behavior events, each event in the sequence of behavior events is labeled with event type and associated data. The typical behavior pattern segments are then converted into feature vector representations, which include numerical features such as event frequency, duration, number of participants and categorical features such as behavior theme or interaction mode, in preparation for subsequent clustering analysis.

[0083] In specific implementation, the typical behavior pattern segments are clustered to form behavior pattern templates, the clustering analysis employs K-means algorithm, the algorithm takes the multi-dimensional feature vectors of the typical behavior pattern segments as input, calculates the similarity between vectors, and assigns the segments into multiple clusters based on the similarity. The objective function of K-means algorithm is to minimize the within-cluster sum of squares, which is expressed as:

[0084]

[0085] wherein the symbol represents the within-cluster sum of squares, the symbol represents the preset cluster number, the symbol represents the feature vector of the typical behavior pattern segment, the symbol represents the i-th cluster, and the symbol represents the centroid vector of the i-th cluster.

[0086] In specific implementation, each behavior pattern template is parsed for condition-action pairs to identify the trigger condition set and the corresponding action set, the condition-action pair parsing is based on feature analysis and context reasoning of the behavior pattern template. The trigger condition set is derived from the starting event or environmental state of the behavior pattern template, for example, when the behavior pattern template involves task allocation, the trigger condition set may include task readiness state or resource availability condition, the elements of the trigger condition set are expressed in the form of logical expressions, such as "task status = pending allocation" and "resource quantity > 0". The corresponding action set is extracted from the core behavior sequence of the behavior pattern template, the corresponding action set includes operations that the student may perform, such as sending messages, adjusting plans, or requesting assistance, each action is labeled with action type, action parameters, and expected effect. The parsing process uses a rule engine to match template features with a predefined condition-action pattern library, ensuring consistency of causal relationships between the trigger condition set and the corresponding action set. The trigger condition set and the corresponding action set are finally stored in the form of key-value pairs, facilitating subsequent formal description.

[0087] In specific implementation, the set of trigger conditions and the set of corresponding actions are formalized and described by state machine representation method to generate standardized behavior rules, and the state machine representation method is implemented by finite state machine model defined by state set, transition condition and action function. The state set includes initial state, intermediate state and terminal state, and each state corresponds to a phase of the behavior pattern template, such as “waiting for trigger” state or “executing action” state. The transition condition is derived from the set of trigger conditions, and the transition condition describes the premise of state change in Boolean expression, such as triggering transition when the environmental variable meets a specific value. The action function is mapped to the set of corresponding actions, and the action function defines the operation sequence executed when the state transitions. The standardized behavior rules are encoded in XML or JSON format, containing information such as rule identification, applicable scenario, state machine graph and rule priority. The formalization description process verifies the integrity and unambiguity of the rules through the state machine compiler to ensure that the standardized behavior rules can be correctly interpreted by the execution engine in simulation training.

[0088] In specific implementation, the standardized behavior rules are indexed and arranged according to the type of training scenario to build a dynamic behavior rule library, and the type of training scenario is based on a professional scenario classification system, such as team cooperation scenario or crisis handling scenario. The classification system adopts a hierarchical structure, with the top layer being a scenario category and the lower layer being a specific scenario variant. The indexing and arranging process groups the standardized behavior rules by scenario type, and creates an index entry for each group of rules, which contains scenario type code, rule list and matching weight. The dynamic behavior rule library is implemented in the form of a database table, with the main table storing rule metadata and the detail table storing rule content. The index structure uses B-tree or hash index to support fast retrieval. During the construction process, the rule library also integrates a version management mechanism to record the history of rule addition, modification or deletion, facilitating maintenance and update. The dynamic behavior rule library is finally deployed in the rule engine of the system to provide real-time behavior prediction and intervention support for the dynamic behavior modeling unit.

[0089] In some embodiments, the extraction of typical behavior pattern segments can introduce semantic analysis techniques to enhance the accuracy of segment boundary identification. The semantic analysis techniques analyze the text description and event context in the historical training log, identify the coherent changes in behavior topics using topic models, and thus determine more reasonable segment division points. Clustering analysis can use hierarchical clustering method instead of K-means algorithm, which merges similar segments from bottom to top to form a tree-like cluster structure, facilitating the generation of multi-granularity behavior pattern templates. The condition-action pair analysis can integrate a machine learning classifier, which learns the mapping relationship between conditions and actions from a large amount of labeled data, improving the automation degree and coverage of the analysis.

[0090] In some embodiments, the state machine representation can be extended to a timed logic state machine that introduces time constraints and concurrent states, capable of describing more complex behavior timing relationships. The index arrangement can implement a dynamic weight adjustment function that dynamically updates the priority of rules in the index according to rule usage frequency and effect feedback, optimizing rule matching efficiency. The dynamic behavior rule library can support an online learning mechanism that updates rule content in real time according to newly generated training data, enabling the rule library to have adaptive evolution capabilities.

[0091] Optionally, feature extraction of typical behavior pattern fragments can incorporate domain-specific features designed based on professional scenario knowledge, such as communication effectiveness indicators or decision quality scores, enriching the expressive power of the feature vector. The distance measure of clustering analysis can use Mahalanobis distance to handle feature correlation, considering the feature covariance matrix to improve the robustness of clustering to outliers. The rule engine for conditional-action pair analysis can configure a fallback strategy to automatically invoke default rules or request human intervention when analysis fails, ensuring system reliability. Optionally, formal description of the state machine representation can use formal verification tools for model checking, which verify the state machine for deadlock-free and reachability, ensuring the logical correctness of standardized behavior rules. The database table of index arrangement can be stored in partitions, partitioned by scenario type or time range, to improve the query performance of large-scale rule data. The access interface of the dynamic behavior rule library can be designed as a RESTful API, facilitating integration with other system units and distributed deployment.

[0092] It can be understood that the method of constructing a dynamic behavior rule library based on historical training data uses data-driven methods to capture behavior patterns, and clustering analysis induces scattered behavior patterns into reusable templates, improving the efficiency of behavior modeling. The state machine representation provides a strict formal basis, making behavior rules computable and predictable, suitable for simulation of complex professional scenarios. The index arrangement optimizes rule organization according to scenario type, supporting fast context matching and enhancing system adaptability in variable environments. The construction process of the dynamic behavior rule library focuses on the value of historical data and the manageability of rules, achieving the conversion from raw logs to structured knowledge through modular processes. Standardized behavior rules ensure the consistency and scalability of behavior models, providing reliable behavior benchmarks for simulation training.

[0093] Example 3: The multimodal interaction management unit captures real-time voice interaction data, action sequence data, and gaze focus data between students during simulated training. It then timestamps and fuses these data to generate a multimodal interaction map. Voice interaction data is acquired using a high-fidelity microphone array, action sequence data is recorded using an inertial measurement unit and an optical motion capture system, and gaze focus data is obtained using a head-mounted eye tracker. Voice content features, intonation features, and speech rate features are extracted from the voice interaction data. Voice content features are converted to text using automatic speech recognition technology, and keywords and semantic vectors are extracted. Intonation features are calculated by analyzing the fundamental frequency profile and intensity changes. Speech rate features are determined based on the number of syllables recognized per unit time. Joint motion trajectory features, motion amplitude features, and motion frequency features are extracted from the motion sequence data. Joint motion trajectory features are calculated from the three-dimensional coordinate sequence of skeletal joints, calculating the curvature and direction of the motion path. Motion amplitude features are quantified using the maximum value and variance of joint displacement. Motion frequency features are analyzed using Fourier transform to determine the repetition period of the motion. This study extracts gaze target features, gaze duration features, and saccade path features from gaze focus data. Gaze target features identify specific objects or areas focused on by students in the virtual environment. Gaze duration features statistically analyze the distribution of continuous gaze time, and saccade path features quantify changes in the angle and speed of gaze movement. A unified timeline is established to align multi-source features according to time windows, with the window size dynamically adjusted based on the rhythm of the simulated training scenario. An attention mechanism is used to weightedly fuse the aligned multi-source features to generate a multimodal interaction graph. The attention mechanism calculates the correlation between each feature dimension and the current task and assigns weights. The fused feature vectors are organized into a graph structure according to the time series.

[0094] In the specific implementation, speech content features, intonation features, and speech rate features are extracted from the voice interaction data. The speech content feature extraction process first uses a pre-trained speech recognition model to convert the audio stream into a text sequence. Then, noun entities and action verbs are extracted from the text as keywords. Simultaneously, a semantic encoder maps the entire text into a fixed-dimensional semantic vector. Intonation feature analysis targets the fundamental frequency trajectory of the audio signal. The fundamental frequency trajectory is calculated using an autocorrelation algorithm, and the standard deviation and slope of the fundamental frequency are extracted as indicators of intonation variation. Speech rate feature calculation is based on the effective speech segments segmented by speech activity detection. The number of syllables per second is counted, and the moving average and coefficient of variation of the speech rate are calculated. Feature extraction from the voice interaction data is segmented in 50-millisecond frames to ensure that the temporal resolution of the features meets real-time requirements.

[0095] In a specific implementation, joint motion trajectory features, motion amplitude features, and motion frequency features are extracted for the action sequence data, which comes from sensor nodes distributed at major joints of the student's body, each node providing three-axis acceleration and three-axis angular velocity data. Joint motion trajectory features reconstruct the three-dimensional motion path of the joint node through a sensor data fusion algorithm, and the curvature feature of the path calculates the local bending degree using the Frenet formula, and the direction feature represents the Euler angle of joint rotation through a quaternion. Motion amplitude features are extracted from the displacement distance of the joint node relative to the torso origin, and the maximum displacement value and displacement variance within the motion period are calculated as amplitude indicators. Motion frequency features are identified by performing a fast Fourier transform on the acceleration data, and the dominant frequency components in the amplitude spectrum are identified, and the motion frequency feature takes the weighted average of the top three main frequency components. The feature extraction of the action sequence data is calculated in a 100-millisecond time window to ensure the complete capture of the action dynamics.

[0096] In a specific implementation, gaze focus data is extracted to extract gaze target features, gaze duration features, and saccade path features, and the gaze focus data comes from the pupil position and corneal reflection data collected by the eye tracker. Gaze target features determine the gaze point coordinates of the student in the virtual environment through a gaze projection algorithm, and are matched with the scene object database to identify the specific target object of the gaze. Gaze duration features are based on the stability threshold of the gaze point, and the period in which the gaze moves at a speed below the threshold is combined into a gaze interval, and the duration of each gaze interval is calculated. Saccade path features analyze the rapid movement trajectory between gaze points, and extract the average angle, peak speed, and motion arc length of saccadic motion as path features. The feature extraction of the gaze focus data is synchronized with the refresh rate of the display device, and is calculated in real time at intervals of 16.7 milliseconds.

[0097] In a specific implementation, a unified time axis is established to align the speech content features, intonation features, speech rate features, joint motion trajectory features, motion amplitude features, motion frequency features, gaze target features, gaze duration features, and saccade path features according to the time window, and the unified time axis is based on a high-precision system clock to stamp each data sampling point with a microsecond-level timestamp. The size of the time window is dynamically adjusted according to the simulation training phase, and a 200-millisecond short window is used in the fast interaction phase and a 500-millisecond long window is used in the strategy decision phase. The alignment process uses an interpolation algorithm to handle data streams with different sampling rates to ensure that all features have corresponding values at the same time point. The feature data within the time window is organized in the form of a multi-dimensional matrix, with the rows corresponding to the time points and the columns corresponding to the feature dimensions.

[0098] In a specific implementation, attention mechanisms are used to weight and fuse the aligned multi-source features, and the attention mechanism first calculates the relevance score of each feature dimension to the core task of the current training phase through a query-key value matching calculation:

[0099]

[0100] wherein: symbol denotes the attention score matrix, symbol denotes the query vector of the current task state, symbol denotes the key vector matrix of all feature dimensions, symbol denotes the dimension of the feature vector, symbol denotes the matrix transpose operation. The initial attention weights are assigned according to the correlation scores, which are normalized by the softmax function. The initial attention weights are contextually corrected by introducing context information in the dynamic behavior rule library, which adjusts the weight distribution through a gating mechanism that considers the importance of the current behavior rule and the historical attention patterns. The corrected attention weights are used to weight and sum the corresponding features to generate a fused feature vector. The fused feature vectors are organized as graph nodes in a time series and connected by edges according to the temporal dependence between features to construct a multi-modal interaction graph, which adopts a directed graph structure, with node attributes containing fused feature vectors and edge weights representing feature transition probabilities.

[0101] In some embodiments, speech content feature extraction can integrate a domain-specific dictionary to enhance the recognition accuracy of professional terms, which contains professional vocabulary and expressions related to the professional scene. The kinematic trajectory features of the motion sequence data can introduce a dynamic time warping algorithm to handle the differences in different student action rhythms, which optimizes the time alignment effect of the trajectory features. The saccade path features of the gaze focus data can combine scene depth information to calculate the real movement distance of the gaze, which is obtained from the three-dimensional model of the virtual environment. The time window alignment process can use a dynamic programming algorithm to optimize the interpolated path, which minimizes the timing errors between different signal sources. The context correction of the attention mechanism can introduce a multi-head attention structure that captures feature correlations from different subspaces to improve the expressiveness of the fused features. The construction of the multi-modal interaction graph can implement an incremental update mechanism that only processes the part of the graph structure that has changed, improving the maintenance efficiency of large-scale graphs.

[0102] Optionally, the speech interaction data can include a speech rate feature, which can be supplemented by a silence segment ratio. The silence segment ratio reflects the thinking pause mode in the conversation. The motion sequence data can include a motion amplitude feature, which can be normalized to eliminate the influence of individual body size differences. The gaze focus data can include a gaze target feature, which can be integrated with target semantic information. The target semantic information describes the functional meaning of the gazed object in the task context. Optionally, the synchronization signal of the unified time axis can come from a hardware time synchronization module, which ensures the consistency of the clocks of multiple devices through GPS or network time protocol. The weight correction process of the attention mechanism can include a regularization constraint to prevent overfitting. The regularization constraint limits the range of weight changes. The storage of the multi-modal interaction graph can use graph database technology, which supports complex graph query and traversal operations.

[0103] It can be understood that the feature extraction of multi-modal data covers three key dimensions of communication, behavior, and visual attention, providing a data basis for a comprehensive understanding of interaction dynamics. The timestamp alignment technology solves the integration problem of multi-source asynchronous data, ensuring the time sequence consistency of feature analysis. The feature fusion method of the attention mechanism adapts to the task focus of different training stages, achieving focused multi-modal information integration. It can be understood that the construction of the multi-modal interaction graph converts the time sequence features into a spatial topology structure, which is convenient for subsequent state evolution and decision analysis. The entire implementation process focuses on balancing real-time processing efficiency and feature quality, meeting the dual needs of low latency and high precision for simulation training systems. The technical scheme of the multi-modal interaction management unit fully utilizes the complementary advantages of multi-source data, providing a rich basis for interaction analysis for professional scenario simulation.

[0104] Embodiment 4: The state evolution tracking unit identifies the decision nodes of individual students according to the multi-modal interaction graph, and tracks the team state change path triggered by the decision nodes to record the state evolution sequence. The decision subject and the decision content are marked at the decision node. The decision subject is determined by the binding relationship between the behavior mode in the multi-modal interaction graph and the student identity, and the decision content is analyzed from the semantic analysis of the speech interaction data and the intention recognition of the motion sequence data. The reaction behavior of other students in the subsequent time steps is deduced based on the decision content simulation. The deduction process is based on the behavior rules and state transition logic in the dynamic behavior rule library. The actual occurrence of the reaction behavior is verified according to the multi-modal interaction graph to record the deviation between the actual behavior and the expected behavior. The deviation record includes the quantitative description of the deviation amount and the deviation type. The causal relationship chain between the decision node, the actual reaction behavior, and the deviation is recorded in the form of a graph structure to form a state evolution sequence. The graph structure uses a directed acyclic graph representation, the nodes represent state events, and the edges represent causal relationships.

[0105] In specific implementation, the decision subject and the decision content are marked at the decision node, the identification of the decision node is realized by analyzing the feature mutation point in the multi-modal interaction graph, the feature mutation point includes the appearance of key instruction words in the voice interaction data, significant posture changes in the action sequence data or gaze target switching in the visual line focus data. The decision subject marking is realized by associating the decision behavior with the specific student through the student identity recognition module, the student identity recognition module is based on biological characteristics or role assignment information. The decision content marking is realized by analyzing the semantic content in the voice interaction data through natural language processing technology, and the operation intention of the decision is determined in combination with the classification result of the action sequence data. Referring to Table 1, the marking information is stored as a decision node attribute table, Table 1 includes node number, time stamp, decision subject identification and decision content description fields.

[0106] Table 1: Decision node attribute table.

[0107]

[0108] In specific implementation, the reaction behavior of other students in subsequent time steps is simulated after simulating the execution of the decision content, the simulation is executed based on the condition-action rules in the dynamic behavior rule library, and the rule engine generates a reaction behavior sequence according to the decision content and the current environment state. The simulation process adopts discrete event simulation technology, and each reaction behavior is modeled as an event with time delay, and the event delay is determined according to the behavior type and historical data statistics. The reaction behavior includes direct response behavior and indirect chain reaction behavior, and the simulation depth is set to three time steps to balance the calculation complexity and the prediction integrity. The output form of the reaction behavior is a behavior sequence with probability, and each behavior is labeled with occurrence probability and influence range.

[0109] In specific implementation, the actual occurrence of the reaction behavior is verified according to the multi-modal interaction graph, the deviation between the actual behavior and the expected behavior is recorded, and the verification process performs time sequence matching between the actual observed behavior sequence in the multi-modal interaction graph and the reaction behavior sequence obtained by simulation. The deviation amount is calculated by comparing the difference degree of the actual behavior and the expected behavior in the key features, and the difference degree is measured by using a multi-dimensional feature distance formula:

[0110]

[0111] Wherein, the symbol represents the deviation amount, the symbol represents the number of feature dimensions, the symbol represents the weight coefficient of the jth feature, the symbol represents the jth feature value of the actual behavior, and the symbol The jth feature value representing the expected behavior. The bias type is classified according to the pattern of feature difference, and the main bias types include time bias, intensity bias, and direction bias. The bias record is stored in the form of a structured bias log, which contains bias number, associated decision node, bias magnitude value, and bias type code.

[0112] In a specific implementation, the state evolution sequence is formed by recording the causal relationship chain between the decision node, the actual reaction behavior, and the bias in the form of a graph structure, which is implemented using a node-edge model. The decision node serves as the root node of the causal graph, and the actual reaction behavior is connected to the corresponding root node as a child node. The behavior type and timestamp are marked as node attributes, the behavior type is selected from a predefined behavior classification system, and the timestamp is accurate to the millisecond level. The bias is marked as an edge attribute on the edge connecting the root node and the child node, and the marked content includes the specific values of the bias magnitude and the bias type. When multiple decision nodes have a time or logical association, a cross-node connection edge is established between the corresponding root nodes, which represents the propagation path of state changes. The storage of the state evolution sequence uses graph database technology, which supports efficient path query and backtracking analysis.

[0113] In some embodiments, the identification of decision nodes can introduce a multi-modal feature fusion algorithm to improve detection accuracy, which simultaneously analyzes the collaborative patterns of speech, motion, and gaze features to reduce misjudgment. The deduction of reaction behaviors can integrate a machine learning prediction model that learns behavior response patterns from historical interaction data to supplement the coverage of the rule base. The bias verification process can implement a real-time alarm mechanism that automatically triggers system intervention when the bias magnitude exceeds the threshold.

[0114] In some embodiments, the construction of the causal relationship chain can use a probabilistic graph model to enhance expression ability, which adds conditional probability to the edge to represent the strength of the causal relationship. The analysis of the state evolution sequence can incorporate a time series pattern mining algorithm that identifies periodic or trend rules of state changes. Visualization of the graph structure can implement a dynamic display function that presents the state evolution process in chronological order. Optionally, the labeling of decision subjects can support role group-level abstraction, and when multiple students collaborate in decision-making, the decision subjects are labeled as role groups rather than individuals. The deduction of reaction behaviors can configure multiple scenario modes, which correspond to different team collaboration styles or emergency strategies. The bias record can include environmental influencing factors that describe the contribution of external condition changes to behavior bias.

[0115] Optionally, the storage of the causal chain can adopt an incremental update strategy, which only processes newly added decision nodes and associated edges, improving the processing efficiency of large-scale evolution sequences. The query interface of the state evolution sequence can support multi-condition filtering, including time range, decision type, or bias level dimensions. It can be understood that the accurate labeling of decision nodes provides a reliable starting point for state evolution analysis, ensuring the accuracy of the traceability path. The behavior deduction and verification mechanism creates an expected and actual comparison framework, providing quantitative basis for team behavior evaluation. The state evolution sequence of the graph structure intuitively presents the causal context of state changes, supporting in-depth analysis of interaction patterns. The implementation of the state evolution tracking unit realizes fine tracking of team dynamics through multi-modal data fusion and rule inference technology. The bias recording mechanism captures the detailed features of behavior differences, providing data support for personalized feedback. The entire scheme focuses on the integrity of the temporal relationship and the explainability of the causal logic, meeting the analysis needs of the professional training scene.

[0116] Referring to Figure 4 In the bias quantification statistics of the state evolution tracking unit, the average bias amount distribution of each decision node intuitively presents the difference between the actual behavior and the expected behavior caused by different decisions. Specifically, the graph takes the decision node number as the horizontal axis and the average bias amount as the vertical axis, and displays the average bias amount of DN001 to DN006 through a bar chart, and marks the 0.2 bias threshold with a dashed line. From the data distribution, the average bias amount of DN004 is the highest (0.34), which is significantly higher than the bias threshold; the average bias amounts of DN002 (0.27) and DN006 (0.21) are also higher than the threshold, which are high-risk nodes of bias; the average bias amounts of DN001 (0.11), DN003 (0.07), and DN005 (0.17) are lower than the threshold, and the behavior consistency is relatively good. This statistical result is based on the behavior feature matching of the multi-modal interaction graph and the multi-dimensional feature distance formula calculation. The quantification of bias amount integrates the weight differences of features such as speech and action, and its distribution can directly reflect the degree of deviation of team behavior corresponding to different decision nodes from the expectation, providing core quantitative basis for the subsequent intervention priority adjustment of the strategy adaptation unit and the performance consistency index calculation of the comprehensive evaluation generation unit.

[0117] In example 5, the strategy adaptation unit matches the candidate strategies in the dynamic behavior rule library based on the state evolution sequence, and outputs the adaptive intervention strategy by adjusting the weights of the candidate strategies according to the real-time training environment parameters, including the task progress indicators and team collaboration indicators. The task progress indicators and team collaboration indicators in the training scene are monitored in real time. The task progress indicators are quantified by the task completion percentage and the milestone completion time, and the team collaboration indicators are measured by the communication frequency and the conflict resolution rate. The progress deviation degree is calculated by comparing the task progress indicators with the preset milestones, which reflects the difference between the actual progress and the planned progress. The collaboration efficiency value is calculated by comparing the team collaboration indicators with the ideal collaboration model, which represents the closeness of the current team collaboration level to the ideal state. The activation threshold and the execution priority of the candidate strategies are dynamically adjusted according to the progress deviation degree and the collaboration efficiency value. The activation threshold determines the strictness of the strategy triggering conditions, and the execution priority controls the execution order of the strategies. The adaptive intervention strategy is selected from the candidate strategies based on the adjusted activation threshold and execution priority, which is the most suitable behavior guidance scheme for the current training state. The comprehensive evaluation generation unit calculates the performance consistency indicators of individuals and teams in multiple dimensions to generate a structured evaluation report. Three evaluation dimensions are defined, including the task execution dimension, the communication and coordination dimension, and the emergency response dimension. In the task execution dimension, the matching coefficient of individual task completion degree and team task synchronization degree is calculated, which measures the coordination between individual contribution and team progress. In the communication and coordination dimension, the correlation index of individual communication initiative and team information flow efficiency is analyzed, which reflects the promoting effect of individual communication behavior on team information sharing. In the emergency response dimension, the coordination ratio of individual decision speed and team crisis recovery speed is evaluated, which embodies the matching degree of individual emergency response and team recovery ability. The matching coefficient, the correlation index and the coordination ratio are normalized and weighted aggregated to generate the performance consistency indicators, which comprehensively reflect the coordination degree of individual and team performance.

[0118] In specific implementation, real-time monitoring of task progress indicators and team collaboration indicators in the training scene, task progress indicators obtain real-time data from the task management subsystem, including subtask completion status, resource consumption rate, and time utilization rate, etc. Team collaboration indicators are extracted from the multi-modal interaction graph, including dialogue turn distribution, opinion adoption rate, and response delay, etc. The monitoring process collects data at a frequency of seconds, and the index data is filtered by moving average to eliminate transient fluctuations. Real-time training environment parameters are stored as time series databases, supporting fast query and historical trend analysis. The monitoring module also includes an anomaly detection function that triggers data verification procedures when the indicator value exceeds the normal range. The team collaboration indicators are compared with the ideal collaboration model to calculate the collaboration efficiency value, and the ideal collaboration model is based on historical excellent training cases, including the optimal communication mode and behavior specification. The collaboration efficiency value is calculated by comparing the actual collaboration indicators with the ideal value, and a weighted similarity algorithm is used to measure the closeness of the current collaboration state to the ideal model. The collaboration efficiency value takes into account communication frequency, information accuracy, and decision quality, etc. Each aspect is assigned a corresponding weight coefficient. The collaboration efficiency value is standardized to a range of 0 to 1, and the closer the value is to 1, the higher the collaboration efficiency. The collaboration efficiency value is updated every five minutes, and the change trend is recorded for subsequent analysis.

[0119] In specific implementation, the activation threshold and execution priority of candidate strategies are dynamically adjusted according to the progress deviation and collaboration efficiency value, the activation threshold adjustment uses a reverse proportional mechanism, when the progress deviation increases, the activation threshold is appropriately reduced, when the collaboration efficiency value decreases, the threshold sensitivity is adjusted accordingly. The execution priority adjustment is based on a comprehensive scoring algorithm, which considers the matching degree of the strategy and the current problem, the historical success rate of the strategy, and the implementation cost of the strategy, etc. The dynamic adjustment process is executed every two minutes to ensure that the strategy parameters respond to changes in training state in a timely manner. The adjusted activation threshold and execution priority are updated to the strategy database for the strategy screening module to call. Based on the adjusted activation threshold and execution priority, the adaptive intervention strategy is selected from the candidate strategies, the selection process first filters out the candidate strategies that do not meet the trigger conditions according to the activation threshold, and then sorts the remaining strategies according to the execution priority. The adaptive intervention strategy selects the top three strategies to form a strategy set, which includes the main strategy and the backup strategy. The specific content of the adaptive intervention strategy is extracted from the dynamic behavior rule library, including strategy description, implementation steps, and expected effect, etc. The adaptive intervention strategy is pushed to the training instructor through the system interface, and is transmitted to other related units in a standardized format.

[0120] In specific implementations, the performance consistency index of the individual and the team in multiple dimensions is calculated, and the matching coefficient of the task execution dimension is obtained by calculating the Pearson correlation coefficient of the individual task completion degree and the team task synchronization degree. The correlation index of the communication coordination dimension determines the contribution of individual communication initiative to team information flow efficiency through regression analysis. The coordination ratio of the emergency response dimension is calculated by the ratio of individual decision speed to team crisis recovery speed, and is logarithmically converted to ensure symmetric distribution. The normalization processing converts the dimension indicators to the range of 0-1 using the minimum-maximum normalization method, and the weighted aggregation uses the weight coefficient determined by the analytic hierarchy process. The final value of the performance consistency index is calculated by weighted summation, and the higher the value, the more coordinated the performance of the individual and the team.

[0121] In some embodiments, the monitoring of the task progress indicator can introduce a predictive indicator that predicts the final completion based on the current progress trend, enhancing the forward-looking nature of progress management. The comparison of team collaboration indicators can use a dynamic baseline adjustment mechanism that adaptively adjusts the ideal value standard according to the characteristics of the training phase. The weight adjustment of the candidate strategy can integrate reinforcement learning algorithms that dynamically optimize and adjust parameters based on the execution effect of the strategy. The calculation of the performance consistency index can add a time decay factor that reduces the influence weight of early data on the current indicator, highlighting the importance of recent performance. The weight coefficient of weighted aggregation can be determined by expert evaluation method, which integrates the experience and judgment of experts in multiple fields. The generation of structured evaluation reports can realize automatic template customization, which automatically selects the most suitable report format according to the training type.

[0122] Optionally, the calculation of progress deviation can distinguish between positive deviation and negative deviation, with positive deviation indicating progress ahead of schedule and negative deviation indicating progress behind schedule, and different processing strategies are used respectively. The calculation of collaboration efficiency value can introduce a context correction factor that considers the impact of training difficulty changes on collaboration requirements. The dimension weight of the performance consistency index can be dynamically adjusted, dynamically adjusting the importance of each dimension according to changes in training focus. Optionally, the push of adaptive intervention strategies can use a multi-modal form, including text prompts, visual signals, and voice reminders, among other ways. The structured evaluation report can generate multi-granularity versions, including detailed versions, abstract versions, and visual versions, among other forms, to meet the needs of different use scenarios.

[0123] Referring to Figure 5In the parameter adjustment and effect evaluation of the strategy adaptation unit, the dynamic changes of the activation threshold, execution priority (normalized), strategy matching degree (%) and historical success rate (%) in each stage (task analysis, dynamic modeling, multi-modal interaction, state evolution, strategy adaptation and evaluation) are intuitively presented. Specifically, the activation threshold (blue curve) is initially 0.75 in the task analysis stage, decreases slightly in the dynamic modeling stage, and gradually rises to 0.85 in the strategy adaptation and evaluation stage; the execution priority (orange curve) presents a "first drop and then rise" fluctuation feature, decreases to 0.60 in the dynamic modeling stage, and rises to 1.00 in the strategy adaptation and evaluation stage; the strategy matching degree (green curve) and the historical success rate (red curve) are highly correlated, both of which are at a low point (78%, 70%) in the dynamic modeling stage, and both of which rise to a peak (95%, 90%) in the multi-modal interaction stage, and then fall, but rise again to near the peak level in the strategy adaptation and evaluation stage. This multi-index linkage change reflects the dynamic adjustment effect of the real-time training environment parameters (task progress, cooperation efficiency) on the candidate strategy activation threshold and execution priority, and the positive correlation between strategy parameter adjustment and strategy matching degree and historical success rate.

[0124] It should be noted that in this document, the 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. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0125] Although embodiments of the present application 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 therein without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A university student professional scene simulation training evaluation system, characterized in that, The system comprises: a task analysis unit configured to receive an initial professional scene description, perform semantic decomposition on the initial professional scene description, extract scene core task elements and role constraint conditions, and generate a structured task framework; a dynamic behavior modeling unit configured to import a preset student role configuration according to the structured task framework, assign initial behavior parameters to each student role configuration, and construct a dynamic behavior rule library based on historical training data; a multi-modal interaction management unit configured to capture voice interaction data, action sequence data, and visual focus data between students in real time during simulation training, timestamp align and feature fuse the voice interaction data, action sequence data, and visual focus data, and generate a multi-modal interaction graph; a state evolution tracking unit configured to identify decision nodes of individual students according to the multi-modal interaction graph, track team state change paths triggered by the decision nodes, and record state evolution sequences; a strategy adaptation unit configured to match candidate strategies in the dynamic behavior rule library based on the state evolution sequences, adjust weights of the candidate strategies according to real-time training environment parameters, and output an adapted intervention strategy; a comprehensive evaluation generation unit configured to integrate the state evolution sequences and the adapted intervention strategy, calculate performance consistency indicators of individuals and teams in multiple dimensions, and generate a structured evaluation report; the adjusting weights of the candidate strategies according to real-time training environment parameters comprises: monitoring task progress indicators and team cooperation indicators in a training scene in real time; comparing the task progress indicators with preset milestones to calculate a progress deviation degree; comparing the team cooperation indicators with an ideal cooperation model to calculate a cooperation efficiency value; dynamically adjusting an activation threshold and an execution priority of the candidate strategies according to the progress deviation degree and the cooperation efficiency value; selecting the adapted intervention strategy from the candidate strategies based on the adjusted activation threshold and execution priority.

2. The university student professional scene simulation training evaluation system according to claim 1, wherein, the semantic decomposition on the initial professional scene description comprises: adopting a hierarchical semantic parsing method to split the initial professional scene description into a scene background section, a task target section, and a constraint condition section; performing entity recognition and relationship extraction on the scene background section to establish a scene entity relationship network; performing action verb analysis and logical dependency parsing on the task target section to construct a task target tree; performing rule clause classification and priority assignment on the constraint condition section to form a constraint rule set; associating and mapping the scene entity relationship network, the task target tree, and the constraint rule set to generate the structured task framework.

3. The university student professional scene simulation training evaluation system according to claim 1, wherein, the constructing a dynamic behavior rule library based on historical training data comprises: extracting typical behavior mode segments from historical training logs, performing cluster analysis on the typical behavior mode segments to form behavior mode templates; performing condition-action pair analysis on each behavior mode template to identify a trigger condition set and a corresponding action set; adopting a state machine representation method to formally describe the trigger condition set and the corresponding action set to generate standardized behavior rules; and Index the standardized behavior rules according to the training scene types, and construct the dynamic behavior rule library.

4. The university student professional scene simulation training evaluation system according to claim 1, wherein, The timestamp alignment and feature fusion of the voice interaction data, action sequence data and line of sight focus data comprises: extracting voice content features, tone features and speech speed features for the voice interaction data; extracting joint motion trajectory features, action amplitude features and action frequency features for the action sequence data; extracting gaze target features, gaze duration features and saccade path features for the line of sight focus data; establishing a unified time axis to align the voice content features, tone features, speech speed features, joint motion trajectory features, action amplitude features, action frequency features, gaze target features, gaze duration features and saccade path features according to time windows; using an attention mechanism to weight and fuse the aligned multi-source features to generate the multi-modal interaction graph.

5. The university student professional scene simulation training evaluation system according to claim 1, wherein, The tracking of the team state change path triggered by the decision node comprises: labeling the decision subject and the decision content at the decision node; simulating the execution of the decision content to deduce the possible reaction behaviors of other students in the subsequent time steps; verifying the actual occurrence of the possible reaction behaviors according to the multi-modal interaction graph, and recording the deviation between the actual behavior and the expected behavior; recording the causal relationship chain between the decision node, the actual reaction behavior and the deviation in the form of a graph structure to form the state evolution sequence.

6. The university student professional scene simulation training evaluation system according to claim 1, wherein, The calculation of the performance consistency indicators of individuals and teams in multiple dimensions comprises: defining a task execution dimension, a communication and coordination dimension and an emergency response dimension; under the task execution dimension, calculating the matching coefficient of individual task completion degree and team task synchronization degree; under the communication and coordination dimension, analyzing the correlation index of individual communication initiative and team information flow efficiency; under the emergency response dimension, evaluating the coordination ratio of individual decision speed and team crisis recovery speed; normalizing and weighting the matching coefficient, correlation index and coordination ratio to generate the performance consistency indicators.

7. The university student professional scene simulation training evaluation system according to claim 2, wherein, The association mapping of the scene entity relationship network, the task target tree and the constraint rule set comprises: identifying key entity nodes in the scene entity relationship network, and performing attribute matching between the key entity nodes and leaf nodes of the task target tree; establishing a constraint binding relationship between the constraint rule set and branch nodes in the task target tree; when attribute matching conflicts or constraint binding conflicts are detected, starting a conflict resolution protocol to reassign entity attributes or adjust constraint priorities; after conflict resolution is completed, generating the structured task framework containing complete association relationships of entities, tasks and constraints.

8. The university student professional scene simulation training evaluation system according to claim 4, characterized in that, The weighted fusion of the aligned multi-source features using an attention mechanism comprises: calculating the relevance score of each feature dimension to the core task of the current training stage; allocating an initial attention weight according to the relevance score; introducing context information in the dynamic behavior rule library to contextually correct the initial attention weight; using the corrected attention weight to weight and sum the corresponding features to generate a fusion feature vector; The fusion feature vector is organized as a graph node in time sequence, and edges are connected according to the time sequence dependence between features, and a multi-modal interaction graph is constructed.

9. The university student professional scene simulation training evaluation system according to claim 5, characterized in that, The causal relationship chain between the decision node, the actual reaction behavior and the deviation recorded in the form of a graph structure comprises: Each decision node is taken as a root node of a causal graph; The actual reaction behavior is connected as a sub-node to the corresponding root node, and the behavior type and time stamp are marked; The deviation is marked as an edge attribute on the edge connecting the root node and the sub-node, and the marking content includes the deviation amount and the deviation type; When multiple decision nodes have time sequence or logical association, a cross-node connection edge is established between the corresponding root nodes to form the state evolution sequence describing the state propagation path.

Images