Personalized infection control training system and method based on generative ai

By building a personalized infection control training system using generative AI, the problems of homogeneous training content and low efficiency have been solved. This has enabled a personalized and highly interactive training experience, improving training quality and efficiency, especially in terms of rapid response capabilities during emergencies.

CN120430529BActive Publication Date: 2025-11-07HUNAN DEYAMANDA TECH CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202510934210.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-07-08
Publication Date
2025-11-07
Estimated Expiration
2045-07-08

AI Technical Summary

Technical Problem

Existing infection control training suffers from homogenized content, low coverage efficiency, and a lack of dynamic evaluation and content update mechanisms, making it impossible to achieve personalized and rapid training. The efficiency problem is particularly prominent during public health emergencies.

Method used

A personalized infection control training system is built using generative AI. Through data collection, competency profiling, knowledge generation, difficulty adjustment, assessment, and personalized learning modules, combined with VR/AR devices, it enables the generation of personalized training content and interactive feedback.

Benefits of technology

It enables personalized training pathways, enhances the relevance and effectiveness of training, strengthens the interactivity and immersion of the training process, dynamically optimizes training feedback and pathways, and improves learners' comprehensive abilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120430529B_ABST
    Figure CN120430529B_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of medical education training system, and particularly relates to a personalized infection control training system and method based on generative AI. The system comprises a data acquisition module, a capability portrait construction module, a knowledge generation module, a difficulty regulation module, an evaluation module, a personalized learning module, an output module and a virtual patient interaction module. The system integrates user behavior data through a federal learning mechanism, constructs a multi-dimensional capability portrait, generates layered training content using a generative large model, and dynamically adjusts the content difficulty by combining a Q-learning algorithm. The system supports VR / AR immersive training and voice interaction control, can simulate real infection scenarios and present 3D propagation consequences. Compared with traditional static teaching programs, the system has stronger adaptability, interactivity and training closed-loop capability, significantly improving the personalization and effectiveness of infection control training.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical education and training systems, and particularly relates to a personalized infection control training system and method based on generative AI. BACKGROUND

[0002] Infection control is one of the core links to ensure patient safety and medical staff health in modern medical institutions, especially in the context of high incidence of infectious diseases and widespread drug-resistant bacteria. Standard, timely and effective infection control training is particularly important. At present, most medical institutions still use traditional methods such as unified courseware, offline lectures or regular examinations to conduct infection control knowledge propaganda and training for medical staff. However, such methods have the following main problems:

[0003] 1. Homogenization of training content and lack of pertinence: Traditional infection control training generally uses standardized content, which cannot be adjusted according to different positions, departments, qualifications, past training records or real work scenarios, resulting in poor training effect for some practitioners and uneven actual prevention and control ability.

[0004] 2. Long training period and low coverage efficiency: Offline centralized training is limited by personnel deployment, venue arrangement and time requirements, making it difficult to achieve rapid training for a large number of medical staff, especially during public health emergencies, when efficiency is even more prominent.

[0005] 3. Lack of dynamic evaluation and content update mechanism: Existing training content is often not updated in time, cannot reflect the latest infectious disease prevention and control guidelines or hospital infection practice cases in real time, and lacks dynamic identification and intelligent feedback on students' knowledge mastery, making it difficult to achieve closed-loop improvement.

[0006] With the rapid development of generative artificial intelligence (Generative AI) technology, it has shown strong capabilities in natural language processing, semantic understanding and personalized content generation. In particular, large language models such as GPT and BERT have shown significant advantages in education, healthcare and enterprise training.

[0007] Therefore, there is an urgent need for a personalized infection control training system based on generative AI to improve the efficiency and accuracy of infection control knowledge dissemination, thereby more effectively ensuring the continuous improvement of hospital infection management level. SUMMARY

[0008] To solve the above problems, the purpose of the present application is to provide a personalized infection control training system based on generative AI, comprising:

[0009] A data acquisition module for obtaining historical training data, including operation behavior data, supervision records, course learning data and environment labels;

[0010] An ability profile construction module is configured to integrate multi-source heterogeneous data through federated learning, and construct an ability profile of a learner, including an identity label, an ability matrix, and a risk weight, to ensure data privacy compliance.

[0011] A knowledge generation module is configured to match an infection control knowledge graph according to the ability profile, and dynamically generate adaptive course content from a basic layer, a professional layer, and a case layer.

[0012] A difficulty regulation module is configured to analyze real-time learning performance of a user based on a Q-learning algorithm, and dynamically adjust course difficulty and content intensity.

[0013] An evaluation module is configured to collect user answering behavior, output scores and knowledge mastery, and generate evaluation scripts.

[0014] An individualized learning module is configured to reconstruct a learning path according to user evaluation results, and generate an individualized course package.

[0015] An output module is configured to push individualized infection control training content to a user and perform feedback loop updates, wherein the output module supports 3D interactive training scenarios of VR / AR devices, and controls training content and interactive navigation through a voice engine to recognize natural language instructions of the user.

[0016] In a preferred technical solution, the knowledge generation module is constructed based on a double-channel generation mechanism, including a natural language processing generation model and a reinforcement learning control model, the former being configured to generate text content, and the latter being configured to regulate content rhythm and complexity.

[0017] In a preferred technical solution, the system further includes a virtual patient interaction module, which is provided with a medical decision tree, embedded with pathogen transmission parameters and patient feature information, and is configured to simulate the influence of different disposal paths on infection transmission risk.

[0018] In a preferred technical solution, the virtual patient interaction module receives user voice instructions, judges operation intention through a semantic analysis engine, and performs consequence deduction by a response module, and displays infection spread results in combination with a transmission dynamics model.

[0019] In a preferred technical solution, the system further includes a 3D consequence visualization module, which is configured to build a dynamic scene based on a Unity engine, and present operation consequences in combination with a haptic feedback device.

[0020] In a preferred technical solution, the ability profile construction module supports a federated learning architecture, collects behavior data from different terminals and performs local encrypted modeling, and finally aggregates model weights at a central node to realize privacy protection.

[0021] In a preferred technical solution, the evaluation script generated by the evaluation module includes multiple knowledge dimension question groups, each question group is associated with at least one knowledge point label, and after the evaluation is completed, the system combines the evaluation results with the competency profile for path adjustment.

[0022] In a preferred technical solution, the output module is connected to a multi-platform terminal, supports parallel access of Web interface, mobile terminal and VR head-mounted display, and forms an immersive and personalized interactive experience.

[0023] The application also provides a personalized infection control training method based on generative AI, including the following steps:

[0024] S1, acquiring multi-source behavior data of learners, including operation data, supervision data, course progress and department label, and constructing a competency profile;

[0025] S2, dynamically generating training content and setting basic difficulty from the knowledge graph according to the competency profile;

[0026] S3, adjusting the content difficulty and training intensity based on the real-time performance of the learners using the Q-learning algorithm;

[0027] S4, pushing the evaluation script to the learners and collecting the answer data;

[0028] S5, calculating the scores of each dimension, analyzing the mastery degree and rearranging the knowledge point order accordingly;

[0029] S6, outputting the personalized learning path and training course, simulating the key operation situation combined with the virtual patient module, and feeding back the consequences based on the 3D visualization engine;

[0030] S7, using the learning path, answer behavior and training performance for subsequent competency profile updating to form a closed-loop learning process.

[0031] In a preferred technical solution, the competency profile construction is realized in a federated learning mode; the dynamic difficulty regulation is optimized based on the Q-learning algorithm; the operation feedback is realized through the virtual patient system and the 3D visualization engine, and is combined with the voice engine for user instruction response.

[0032] Advantages:

[0033] The personalized infection control training system based on generative AI proposed in the application integrates multi-modal behavior data acquisition, competency profile construction, dynamic content generation and interactive feedback mechanism, and can effectively solve the problems of content homogenization, evaluation lag and insufficient adaptation in the existing infection control training process, and has the following advantages:

[0034] 1. Realize personalized training path: through the federal learning mechanism, the post label, behavior data and knowledge mastery of learners are fused, the ability portrait is dynamically constructed, the training content is intelligently generated by generative AI, the precise distribution and ability adaptation of training tasks are realized, and the pertinence and effectiveness of training are improved.

[0035] 2. Improve the interactivity and immersion of the training process: the system integrates VR / AR devices and voice interaction modules, builds 3D simulation scenes based on Unity engine, realizes operational training and instant feedback function, enhances the perception experience and memory retention effect of learners.

[0036] 3. Enhance training feedback and path optimization ability: the evaluation module supports multi-dimensional behavior evaluation and knowledge point analysis, dynamically adjusts the training task difficulty and content according to the user performance combined with Q-learning reinforcement learning mechanism, and constructs a continuously optimized closed-loop learning process.

[0037] 4. Restore the real decision logic of clinical: through the virtual patient interaction module, multi-path decision simulation is constructed, combined with the propagation parameter and infection consequence deduction, the comprehensive ability of learners in emergency response, risk identification and protection decision is improved.

[0038] In summary, the present application realizes the transformation of infection control training from static teaching to intelligent adaptation and dynamic optimization by integrating generative AI and behavior data closed-loop technology, which not only improves the training efficiency and quality, but also provides a practical support scheme for digital and intelligent hospital infection management. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is a system structure schematic diagram of the present application;

[0040] Figure 2 is a method flowchart of the present application. DETAILED DESCRIPTION

[0041] In order to deepen the understanding of the present application, the present application will be further described in combination with examples, and the present embodiment is only used to explain the present application, and does not constitute a limitation on the protection scope of the present application.

[0042] Example one

[0043] According to Figure 1 , the present embodiment discloses a personalized infection control training system based on generative AI, which faces the infection prevention and control training needs of different types of medical staff in medical institutions, and comprehensively uses multi-source behavior data fusion, generative artificial intelligence model, reinforcement learning, voice interaction and 3D simulation technology to build a dynamic adaptation, closed-loop feedback personalized intelligent training platform.

[0044] The system comprises the following modules:

[0045] 1. Data acquisition module:

[0046] This module is deployed in the hospital teaching management platform and intelligent terminal, and is responsible for collecting data including but not limited to the following four types:

[0047] 1) Operation behavior data: such as hand hygiene execution rate, protective equipment wearing standard;

[0048] 2) Supervision records: such as infection control patrol scores and rectification suggestions;

[0049] 3) Course learning data: including completed courses, evaluation scores, learning frequency, etc.

[0050] 4) Environmental label information: such as department, post level, and department infection risk level, etc.

[0051] All data is preprocessed using a local encryption mechanism and transmitted to the ability portrait construction module.

[0052] 2. Ability portrait construction module:

[0053] This module uses a federated learning (Federated Learning) architecture to aggregate and process heterogeneous data from multiple terminals. The local training model weight of the first terminal is , which is transmitted to the central coordination server for global update. The update method is as follows:

[0054] ;

[0055] Where is the global model weight, is the data volume of the th device, is the total data volume of all devices; is the total number of terminals.

[0056] Through this method, the ability vector modeling under privacy protection can be realized. The ability portrait includes identity labels (such as job level, professional background), ability matrix (knowledge dimension score), and risk weight (based on work environment and historical evaluation results).

[0057] 3. Knowledge generation module:

[0058] ​This module is built on top of a hierarchical knowledge graph, including the basic layer (such as hand hygiene standards), professional layer (department operation SOP), and case layer (historical infection events and transmission chain analysis). The system embeds a dual-channel generation architecture, using large language models (such as fine-tuned large language models) to generate textual learning content, while introducing a control model based on reinforcement learning to dynamically regulate the difficulty and depth of the generated content. This dual-channel structure logically realizes the parallel generation of semantic content and structural optimization of courses.

[0059] 4. Difficulty regulation module:

[0060] The system dynamically assesses the learning state of the learner based on their behavior data and uses the Q-learning algorithm to achieve adaptive adjustment of the difficulty of training tasks. The system represents the current ability profile as state , and the pushed task as action . The Q-value update formula is as follows:

[0061] ;

[0062] where is the current state; is the current action; is the learning feedback score, representing the immediate reward (reward) obtained at the current step; is the state transition result, representing the new state (next state) reached after executing action ; is the learning rate; is the discount factor (discount factor, 0~1), used to measure the weight of future rewards; is the Q-value of the action with the maximum future expected total return value among all possible actions in the new state . By continuously updating the Q-value, the system can guide the learning path to the optimal strategy.

[0063] 5. Evaluation module:

[0064] This module generates targeted evaluation scripts based on the structure division in the knowledge graph, covering dimensions such as risk prevention and control, isolation and disinfection, and occupational protection. Each question group is tagged and mapped to specific knowledge points. After the learner completes the evaluation, the system scores and analyzes the accuracy, reaction time, and behavior preferences of the answers, and returns the ability update results to the ability profile module.

[0065] 6. Personalized learning module:

[0066] According to the performance of the learners in the evaluation module, the module reorders the learning path and prioritizes the push of knowledge points with low scores and weak mastery. Meanwhile, it can automatically generate learning packages covering wrong answer analysis, structured supplementary training content, and related knowledge review materials.

[0067] 7. Output module:

[0068] This module is responsible for pushing customized content to multiple terminal platforms. The system is compatible with Web, mobile App, and VR headset interfaces, supporting cross-platform content synchronization. Learners can enter 3D interactive scenes through VR / AR devices. The system simulates scenarios such as wearing protective clothing and entering isolation rooms to enhance immersive training experience. The scene embeds a voice recognition module, allowing learners to control the learning progress through natural language commands such as "continue to the next step" and "re-explain," improving operational convenience.

[0069] 8. Virtual patient interaction module:

[0070] This module builds a decision structure based on rule trees and causal relationships. The system embeds pathogen transmission parameters (such as R0 value and drug resistance) and patient vital signs into node logic. After learners make diagnosis and treatment decisions for virtual patients, the system evaluates the treatment effect and guides them to explore the consequences of different choices.

[0071] 9. 3D consequence visualization module:

[0072] At key training nodes, the system calls the Unity engine to build consequence scenes. Incorrect operations such as not isolating in time and incorrect hand washing procedures will trigger pathogen transmission animations, combined with haptic feedback devices such as vibration to provide risk prompts, allowing learners to establish deep memory and behavioral alertness during training.

[0073] The overall operation process of this embodiment system forms a closed-loop process of "data collection - portrait construction - knowledge generation - training push - effect evaluation - path reorganization - relearning," with real-time adaptability, high interactivity, and strong memory effect, effectively supporting the individualization and precision needs of hospital infection control personnel training.

[0074] Example Two

[0075] This embodiment provides a generative AI-based personalized infection control training system deployed in a regional medical collaboration network, suitable for multi-institution joint training management scenarios. The system supports remote data collaborative modeling, personalized path sharing, and cross-platform hybrid training, enabling unified training and differentiated adaptation of medical personnel at all levels in a medical association.

[0076] The overall structure and module division of this system are as follows:

[0077] 1. Data collection module: This module connects the online platform of hospitals, community health centers and third-party training institutions, and collects the following data:

[0078] 1) Structured training records: including course code, knowledge point index and evaluation information;

[0079] 2) Dynamic operation log: disinfection, hand washing and wearing operation behavior are collected through camera equipment and Internet of Things gesture monitor;

[0080] 3) Event response data: including the response path of medical staff involved in real hospital infection events and its evaluation;

[0081] 4) Remote supervision feedback: composed of scoring records and problem annotation data by remote infection control experts.

[0082] The collected data is encrypted by the local server of the institution and uploaded to the central node for global modeling.

[0083] 2. Ability profile construction module: This module realizes regional collaborative modeling based on federated learning mechanism, and each medical institution locally constructs model parameters , uploaded to the central server to aggregate into a unified ability profile model:

[0084] ;

[0085] wherein, is the global model weight; is the local training model weight of the th terminal; is the total number of terminals.

[0086] Finally, a multi-dimensional ability vector of each learner is obtained, including knowledge proficiency score, behavior consistency index, response bias coefficient, etc.

[0087] 3. Knowledge generation module: This module calls an AI knowledge engine that supports three-level semantic generation:

[0088] The first level of processing is fact-based question answering (e.g. hand hygiene standard procedure);

[0089] The second level of processing is situational reasoning (e.g. how to determine if immediate isolation is needed);

[0090] The third level of processing is decision simulation tasks (e.g. selecting the appropriate protection level based on case information).

[0091] The system matches each learner with a task package based on the ability profile, and the task content is generated by a large language model and synchronized with a labeled structure description to ensure traceability of evaluation.

[0092] 4. Difficulty control module: The system uses Q-learning algorithm to analyze the learning behavior of the user in the virtual training process in real time, such as whether the operation is completed within the prompt, whether the transmission path is correctly identified, etc., to adjust the task complexity. The state-action space is established as follows:

[0093] ;

[0094] wherein, is the current state; is the current action; is the learning feedback score, indicating the immediate reward (reward) obtained in the current step; is the state transition result, indicating the new state (next state) reached after executing the action ; is the learning rate; is the discount factor (discount factor, 0~1), used to measure the weight of future rewards; is the Q value of the action with the maximum future expected total return value among all possible actions in the new state .

[0095] 5. Evaluation module: The system embeds evaluation triggers at each knowledge path node, and the evaluation content includes:

[0096] Quick judgment questions (used to test reaction speed);

[0097] Video clip analysis questions (users need to identify operation errors);

[0098] Virtual patient handling questions (comprehensive judgment);

[0099] Each evaluation result is encoded as a multi-dimensional data vector and sent to the scoring model and portrait model for updating.

[0100] 6. Personalized learning module: After the user completes a training cycle, the system re-evaluates their knowledge point mastery and adjusts the path using hierarchical rules, including:

[0101] 1) Weak items are arranged in priority;

[0102] 2) Associated knowledge points are linked for supplementary training;

[0103] 3) Time sequence dynamically inserts actual event playback module.

[0104] Finally, a personalized relearning path with individual trajectory and task review functions is formed.

[0105] 7. Output module:

[0106] The output module supports multi-terminal interaction, including desktop, mobile terminal and VR headset. The system maps task content into a 3D interactive scene, renders different scenarios (such as operating room operation, ward inspection, and handling of sudden exposure events) through the Unity engine, and users can complete process control with the help of a handle and voice commands to improve immersion experience and scenario restoration. The voice recognition module supports dual-channel matching of Mandarin and standard medical terminology. Instructions such as "reset", "view consequences", and "switch patients" can be executed in real time.

[0107] 8. Virtual patient interaction module:

[0108] This module calls the decision simulation model after the user triggers a key action node (such as "select protection level") and relies on medical decision trees, infection transmission causal chains, and patient individual parameters (age, underlying disease, and drug resistance) to perform consequence deduction. If the learner does not perform protection before performing a puncture operation, the system will demonstrate a 3D animation of rapid R0 value rise and transmission chain expansion.

[0109] 9. 3D visualization module:

[0110] The system uses the Unity particle system to achieve visual simulation. The error path will generate a color-highlighted transmission path diagram, accompanied by haptic feedback (such as VR handle vibration) to build multi-sensory feedback and improve risk awareness. This module also allows administrators to inject custom case parameters through the editor to expand the trainable range of infection events.

[0111] The system described in this embodiment is applied to regional multi-institution infection control training projects, supports unified modeling and individual training, ensures privacy protection, improves training adaptability and responsiveness, and provides reliable support for building a smart medical training system.

[0112] Embodiment Three

[0113] As shown in Figure 2 The present embodiment provides a personalized infection control training method based on generative AI, which is suitable for hierarchical and classified precision training tasks carried out by infection management departments in comprehensive hospitals. This method is based on the data closed-loop principle to build a "collection-modeling-generation-evaluation-feedback-reinforcement" whole process, and through the linkage of generative AI and reinforcement learning, it realizes a highly personalized and multi-modal interactive training experience. The specific steps are as follows:

[0114] S1. Collect multi-dimensional behavior data and build a capability profile;

[0115] This method first obtains four types of data of learners through a teaching management system and on-site collection equipment:

[0116] 1) Practical operation data (such as operation time, hand hygiene specification completion rate);

[0117] 2) Historical learning records (e.g., course completion, historical assessment scores);

[0118] 3) Supervision data (e.g., monthly infection control scores for the department);

[0119] 4) Post tags and department infection risk levels.

[0120] The above data is processed locally and participates in federated learning (Federated Learning) modeling. Each terminal conducts capability profile sub-model training locally to generate local weights , and the center node conducts global aggregation:

[0121] ;

[0122] where, is the global model weight; is the local training model weight of the th terminal; is the total number of terminals.

[0123] Get the ability vector of each learner, which contains the operation behavior stability index, infection knowledge mastery, response bias score, etc.

[0124] S2, generate personalized training tasks and dynamically match the content;

[0125] According to the ability profile, the method calls the generative AI language model to dynamically generate training content against the knowledge graph structure (divided into three levels of basic knowledge, operation specification, and situational decision-making), including:

[0126] Basic knowledge module: push CDC standards, flowchart understanding questions;

[0127] Operation specification module: provide specific scene training guidance, such as glove donning and doffing sequence simulation;

[0128] Decision simulation module: generate multiple-choice questions and multi-path operation tasks combined with case texts.

[0129] At the same time, an initial training path is constructed for each learner, and the knowledge points are pushed in sequence according to the weak areas of the ability profile.

[0130] S3, introduce Q-learning algorithm for difficulty control;

[0131] When the learner performs each task, the system records his reaction time, error type, and correction efficiency, and represents the process state as , according to the performance to push the action (e.g. "intensive practice" or "enter the next level"), and calculate the reward value according to the result The Q value update formula is as follows:

[0132] ;

[0133] wherein, is the current state; is the current action; is the learning feedback score, indicating the immediate reward obtained at the current step; is the state transition result, indicating the new state reached after performing the action ; is the learning rate; is the discount factor (0~1), used to measure the weight of future rewards; is the Q value of the action with the maximum future expected total return value among all possible actions in the new state .

[0134] S4, generate evaluation scripts and implement assessment;

[0135] After completing each stage task, the system automatically generates a structured evaluation script, covering:

[0136] Multiple-choice knowledge judgment questions;

[0137] Multimedia clip identification task (identify the wrong steps in the hand washing process);

[0138] Virtual patient decision simulation (e.g., whether to immediately isolate, whether to start the exposure reporting mechanism).

[0139] All evaluation results will be fed back to the competency profile model to achieve dynamic updating.

[0140] S5, path reorganization and personalized relearning;

[0141] If the learner's score in a certain module is lower than the threshold (e.g., <60 points), the system will automatically reorder it to the knowledge points with high learning priority. Path reorganization strategies include:

[0142] Parallel training of similar knowledge points; supplementary learning of derivative knowledge points; insertion of case review modules.

[0143] At the same time, it provides targeted relearning content package, including wrong question analysis, video commentary, interactive flowchart, etc.

[0144] S6, fusion of virtual patient simulation and 3D feedback;

[0145] In some critical tasks, such as "patient sudden vomiting pollution event handling" or "blood exposure emergency response", the system calls the virtual patient module to construct the decision path and simulate the propagation consequences.

[0146] Each decision node embeds propagation parameters (R0 value, pollution area, etc.) to generate consequence simulation paths;

[0147] If the operation is wrong, the system calls the 3D module to demonstrate the virus diffusion process in the VR / AR scene and simulates the droplet trajectory with a particle system;

[0148] With the haptic feedback device (such as VR handle vibration), the risk sense is enhanced.

[0149] S7, voice interaction and closed-loop feedback mechanism;

[0150] The method supports voice command control, and the learner can complete the following operations in the virtual scene through natural language:

[0151] Control flow: "next step" "return to the previous step";

[0152] Information request: "please explain the operation" "show the consequences";

[0153] Simulated communication: "please assume I am the patient, please ask questions".

[0154] The whole process data and feedback of the training process will be recorded in the learner profile library for subsequent re-learning task calling, and finally realize the complete closed-loop operation of the infection control training task.

[0155] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A generative AI-based personalized infection control training system, characterized by, Comprise: a data collection module for obtaining historical training data, including operation behavior data, supervision records, course learning data, and environmental labels; an ability profile construction module for integrating multi-source heterogeneous data through federated learning to construct a learner's ability profile, including an identity label, an ability matrix, and a risk weight, to ensure data privacy compliance; a knowledge generation module for matching an infection control knowledge graph based on the ability profile to dynamically generate adaptive course content from the basic, professional, and case layers; the knowledge generation module is constructed based on a double-channel generation mechanism, including a natural language processing generation model and a reinforcement learning control model, the former for generating text content, and the latter for regulating content rhythm and complexity; a difficulty control module for dynamically adjusting course difficulty and content intensity based on a Q-learning algorithm to analyze user real-time learning performance; an evaluation module for generating an evaluation script and collecting user answer behavior to output scores and knowledge mastery; an individualized learning module for reconstructing a learning path based on user evaluation results to generate an individualized course package; an output module for pushing individualized infection control training content to users and performing feedback loop updates, wherein the output module supports 3D interactive training scenarios for VR / AR devices and recognizes user natural language instructions through a voice engine to achieve training content control and interactive navigation; a virtual patient interaction module; the virtual patient interaction module is provided with a medical decision tree, embedded with pathogen transmission parameters and patient feature information, for simulating the impact of different disposal paths on infection transmission risk; the virtual patient interaction module receives user voice instructions, judges operation intent through a semantic analysis engine, and performs consequence deduction by a response module, and displays infection spread results in combination with a transmission dynamics model; a 3D consequence visualization module; the 3D consequence visualization module constructs a dynamic scene based on the Unity engine and presents operation consequences in combination with a haptic feedback device.

2. The generative Al-based personalized infection control training system of claim 1, wherein: The ability profile construction module supports a federated learning architecture, collects behavior data from different terminals and performs local encrypted modeling, and finally aggregates model weights at the center node to achieve privacy protection.

3. The generative Al-based personalized infection control training system of claim 2, wherein: The evaluation script generated by the evaluation module includes a plurality of knowledge dimension question groups, each question group is associated with at least one knowledge point label, and after the evaluation is completed, the system will combine the evaluation results with the ability profile for path adjustment.

4. The generative Al-based personalized infection control training system of claim 3, wherein: The output module connects multiple platforms, supports parallel access of Web interface, mobile terminal, and VR headsets, and forms an immersive and individualized interactive experience.

5. The working method of the generative AI-based individualized infection control training system according to any one of claims 1-4, comprising the following steps: S1, obtaining multi-source behavior data of learners, including operation data, supervision data, course progress, and department labels, and constructing an ability profile; S2, dynamically generating training content from a knowledge graph based on the ability profile and setting a basic difficulty; S3, adjusting content difficulty and training intensity based on the real-time performance of learners using a Q-learning algorithm; S4, pushing an evaluation script to learners and collecting answer data; S5, calculate the dimension score, analyze the mastery degree and rearrange the knowledge point order accordingly; S6, output the personalized learning path and training course, simulate the key operation situation combined with the virtual patient module, and perform operation feedback based on the 3D visualization engine; S7, use the learning path, answering behavior and training performance for subsequent ability portrait updating to form a closed-loop learning process.

Citation Information

Patent Citations

  • Multi-crowd adaptive training evaluation system

    CN119624717A