A method for generating immersive virtual reality training content

Abstract

This invention relates to the field of virtual reality simulation technology, and more particularly to a method for generating immersive virtual reality training content. The method includes: a step of constructing a domain knowledge graph: extracting atomic knowledge elements representing operational actions, equipment entities, state parameters, and safety rules from standard documents and case libraries of the target training domain, and establishing multi-dimensional logical relationships between these atomic knowledge elements, including causal, temporal, hierarchical, and probabilistic relationships, to form and store the domain knowledge graph; a step of real-time collection of trainee data; a step of constructing and updating a trainee state model; a step of dynamic content decision-making; and a step of generating and injecting non-linear training content. By collecting trainee data in real time and matching it with the knowledge graph, the system can provide trainees with precise feedback after each training session, helping them identify and overcome their knowledge gaps.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual reality simulation, and in particular to a virtual reality immersive practical training content generation method. BACKGROUND

[0002] Virtual reality (VR) technology has been widely applied in education and training fields, especially in high-skill-required or high-risk fields such as flight simulation, medical surgery, industrial operation, etc. Existing VR practical training systems usually adopt pre-programmed fixed scenes and linear scripts. The operation flow of trainees in the virtual environment is basically limited, and the system judges the right or wrong operation according to the preset key nodes.

[0003] However, this static and linear content generation method has the following significant defects:

[0004] Lack of authenticity and adaptability: the high-risk operation environment in the real world (such as chemical production fault disposal, power system emergency repair, complex surgical operation) is full of uncertainty, and the occurrence of events is often nonlinear and dynamically evolving. Fixed scripts cannot simulate this randomness and suddenness, leading to a gap between training and actual combat.

[0005] Lack of personalized adaptation: the knowledge level, operation proficiency and psychological stress tolerance of different trainees differ greatly. Unified training content cannot focus on the weak links of students, resulting in low training efficiency. For example, for students who always make the same operation errors, the system cannot intelligently generate deep training scenes targeting the consequences of the errors.

[0006] Lack of immersion and tension: due to fixed content, students may develop a psychological expectation of backflow after repeated training, thereby weakening the immersion and tension brought by VR training in a real high-pressure environment, making it difficult to evaluate their psychological and operational performance in real emergency situations.

[0007] Therefore, there is an urgent need in the prior art for a method that can dynamically and intelligently generate nonlinear virtual reality practical training content highly adapted to the real-time state of trainees. SUMMARY

[0008] Based on the above purpose, the present application provides a virtual reality immersive practical training content generation method, comprising the following steps:

[0009] The step of constructing a domain knowledge graph: atomic knowledge elements representing operation actions, device entities, state parameters and safety rules are extracted from the specification documents and case libraries of the target training domain, and multi-dimensional logical relationships including causal relationships, temporal relationships, hierarchical relationships and probability relationships between the atomic knowledge elements are established, forming a domain knowledge graph and storing it;

[0010] Real-time collection of trainee data step: collecting the operation behavior data stream of the trainee in real time through the virtual reality interactive device, the operation behavior data stream including the identification of the virtual device object operated, the type of operation action, the time sequence of operation action and operation accuracy; collecting the physiological data stream of the trainee in real time through the biological sensor integrated in the virtual reality head-mounted device, the physiological data stream including heart rate variability, pupil diameter change rate and gaze focus staying time length;

[0011] Construction and updating of student state model step: based on the operation behavior data stream, the operation proficiency score is obtained by calculating the completion degree and the fluency degree of the operation action, and the knowledge blind spot set is obtained by identifying the operation sequence repeatedly ignored or incorrectly executed, which together constitute the ability state dimension; based on the physiological data stream, the tension index is obtained by fusing the changes of heart rate variability and pupil diameter change rate relative to the baseline, and the attention concentration index is obtained by analyzing the distribution of the gaze focus on the key device object, which together constitute the psychological state dimension; the ability state dimension and the psychological state dimension are fused into the student state model;

[0012] Dynamic content decision step: the knowledge blind spot set in the student state model is matched with the semantic similarity of the domain knowledge graph to find the associated unmastered knowledge sub-graph; the tension index and the attention concentration index are matched with the pre-defined event psychological requirements in the knowledge graph; based on the matching result, the trigger probability of the potential dynamic event is calculated and its type, trigger time and intensity are determined according to the multi-dimensional logical relationship, and the content injection instruction is generated;

[0013] Generation and injection of nonlinear practical training content step: according to the content injection instruction, the corresponding event script and resource are called from the pre-set virtual asset library, a specific dynamic event is instantiated in the virtual reality environment, and the dynamic event is coupled with the physical system, sound system and visual system of the virtual environment in real time and then injected into the current practical training process, thereby forming a self-adaptive evolving nonlinear practical training content.

[0014] Preferably, the multi-dimensional logical relationship between the atomic knowledge elements includes:

[0015] A causal strength weight factor is given to the causal relationship, which is determined by analyzing the statistical frequency of the cause event leading to the result event in the historical cases, and combining the effectiveness of the frequency to make correction and evaluation by the domain experts;

[0016] A time sequence constraint weight factor is given to the time sequence relationship, which is obtained by calculating the contribution degree of the operation violating the time sequence relationship in the historical failure cases, and combining the criticality level of the operation in the whole process to get weighted.

[0017] a probability occurrence weight factor is assigned to the probability relation, which is obtained by statistically calculating the conditional probability that the derivative event pointed by the relation actually occurs after the primary event occurs, and introducing expert knowledge to adjust the expected occurrence intensity of the conditional probability in the training environment;

[0018] The storage format of the domain knowledge graph is an attribute graph model, wherein a node represents the atomic knowledge element, an edge represents the multi-dimensional logical relation, and an attribute field of the edge is used to store the corresponding weight factor.

[0019] Preferably, the operation accuracy is obtained by the following process:

[0020] An ideal operation trajectory curve is defined in advance for each standard operation action, which is composed of a series of continuous spatial coordinate point sequences;

[0021] When the trainee performs the operation, the actual motion trajectory of the operation instrument in the virtual space is recorded, and the actual coordinate point sequence is sampled;

[0022] The average deviation distance in three-dimensional space between the actual coordinate point sequence and the ideal operation trajectory curve is calculated, and the average deviation distance is converted into a standardized accuracy score through a pre-defined mapping function, which ensures that the accuracy score is negatively correlated with the deviation distance.

[0023] Preferably, the tension index obtained by fusing the change of heart rate variability and the change rate of pupil diameter relative to the baseline includes:

[0024] During a pre-set calm period before the practical training, the heart rate variability and the change rate of pupil diameter of the trainee are measured, and the average values thereof are calculated as individual baseline levels, respectively;

[0025] During the practical training, the percentage decrease of heart rate variability relative to the individual baseline level is calculated in real time;

[0026] The percentage increase of the change rate of pupil diameter relative to the individual baseline level is calculated in real time;

[0027] The percentage decrease and the percentage increase are multiplied by a pre-set contribution coefficient, respectively, and then added to obtain a preliminary tension synthesis value;

[0028] The preliminary tension synthesis value is input into an S-shaped function for normalization processing, and the value range thereof is mapped to between zero and one, and the output is the final tension index, and the center point position of the S-shaped function is fine-tuned according to the norm data corresponding to the age and gender of the trainee.

[0029] Preferably, the obtaining the attention concentration index by analyzing the distribution of the gaze focus on the key device objects comprises:

[0030] According to the domain knowledge graph, a set of key device entities most relevant to the current practical training task stage is extracted;

[0031] Through the built-in eye movement tracking module of the virtual reality head-mounted device, the coordinates of the gaze focus of the trainee in the three-dimensional space of the virtual environment are continuously obtained;

[0032] Through a collision detection algorithm, it is judged whether the gaze focus coordinates intersect with the three-dimensional model of any entity in the set of key device entities, and the residence time is recorded;

[0033] The ratio of the total duration of the gaze focus on the key device entities to the total duration of the residence in non-key environment areas within a adjustable time window is calculated;

[0034] The ratio is compared with the baseline ratio measured during the calm period of the trainee, the relative change rate is calculated, and the change rate is mapped to the interval between zero and one through a linear normalization function to obtain the attention concentration index. The higher the index value, the more concentrated the attention.

[0035] Preferably, the obtaining the knowledge blind spot set by identifying repeatedly ignored or incorrectly executed operation sequences comprises:

[0036] The operation behavior data stream is continuously monitored and compared with the current expected correct operation sequence derived according to the domain knowledge graph;

[0037] When an operation action error or omission is detected, the corresponding atomic knowledge element identifier and its error type are recorded;

[0038] An initial error count threshold is set, and each time the same error type occurs again for the same atomic knowledge element, the error count is accumulated once;

[0039] The error count threshold is not a fixed value, but is dynamically adjusted according to the criticality of the operation in the time sequence relationship. The higher the criticality of the operation, the lower the corresponding error count threshold is set, so that key knowledge blind spots can be identified more quickly;

[0040] When the count of a certain error type exceeds its dynamically adjusted error count threshold, the atomic knowledge element associated with the error type is officially added to the knowledge blind spot set.

[0041] Preferably, the calculating the trigger probability of potential dynamic events and determining its type, trigger timing and intensity comprises:

[0042] Trigger probability calculation: starting from the unacquired knowledge subgraph, traverse all potential dynamic event nodes connected by causal and probabilistic relationships, multiply the probability occurrence weight factors of all edges on the path according to the conditional probability rule, and obtain the initial trigger probability of each potential dynamic event; then use the operation proficiency score in the trainee state model to modify the initial trigger probability, the higher the proficiency, the higher the trigger probability after modification for reinforcement training;

[0043] Type determination: select a predetermined number of potential dynamic events with the highest modified trigger probability to form a candidate event set;

[0044] Trigger timing determination: analyze the constraint relationship between each event in the candidate event set and the operation process being executed in terms of time sequence, and select the legal time point closest to the current time with the strongest time sequence constraint weight factor as the trigger timing;

[0045] Intensity determination: based on the attribute range of the candidate event itself, the intensity value is determined by linear interpolation according to the current tension index of the trainee, the high value of the intensity range is taken when the tension index is low, and the low value is taken when the tension index is high, but at the same time an incentive factor positively correlated with the operation proficiency score is introduced, so that high-level trainees will also receive higher intensity challenges even under moderate tension.

[0046] Preferably, the real-time coupling of the dynamic event with the physical system, sound system and visual system of the virtual environment comprises:

[0047] Physical system coupling: when the dynamic event is instantiated, a physical parameter disturbance value conforming to its event type is generated, and the size of the disturbance value is determined by the event intensity parameter; the disturbance value is sent to the physical engine of the virtual environment, and the physical engine calculates the chain effect of the disturbance on the state parameters of the related virtual device entities according to the built-in simulation model, and updates the physical state of the entire virtual environment;

[0048] Sound system coupling: according to the type, intensity and three-dimensional spatial position of the dynamic event, the corresponding sound effect file is called from the sound asset library, and the volume, pitch and spatial surround effect of the sound effect are adaptively adjusted based on the parameter disturbance result provided by the physical engine, and then played through the three-dimensional audio rendering engine;

[0049] Visual system coupling: according to the type and intensity of the dynamic event, the corresponding particle special effect template and material change script are called from the visual effect asset library, and the rate, range and material surface shader parameters of the particle emitter are driven combined with the updated device state parameters provided by the physical engine, to generate corresponding visual effects and submit to the rendering pipeline.

[0050] Preferably, after the step of generating and injecting nonlinear practical training content, further comprising:

[0051] The training effect evaluation and knowledge graph self-evolution step: record the complete response operation sequence of the trainee to the injected dynamic event and its final result; compare the response operation sequence with the ideal coping scheme pre-stored in the knowledge graph to generate a coping effectiveness score; extract all new operation sequence relationships, causal logic or state changes involved in the disposal process of the dynamic event and abstract them as new atomic knowledge elements or candidate objects of logical relationship edges; when the coping effectiveness score is lower than the preset level, submit the candidate objects to the knowledge graph management terminal, and after the confirmation of the field experts, integrate them as new knowledge points or fault cases into the field knowledge graph, thereby realizing the adaptive growth of the system knowledge ability.

[0052] Preferably, the target training field is a high-risk industry field that requires operators to have rapid decision-making and precise operation ability in a high-pressure environment, including high-voltage line fault elimination in power grid operation and maintenance, high-risk reaction kettle emergency shutdown disposal in chemical production, or endoscopic minimally invasive surgery in the medical field.

[0053] The beneficial effects of the present application are:

[0054] 1. The present application adopts a dynamic and nonlinear scene generation method, which can generate different training scenes according to the operation behavior of the trainee and the real-time environmental feedback during the training process, so that the trainee can operate in a more realistic environment. In this way, the system can adjust the training content in real time, simulate the occurrence and dynamic evolution of uncertain events, thereby improving the authenticity and adaptability of the training, and ensuring that the trainee's response in an emergency is more in line with actual needs.

[0055] 2. The present application analyzes the operation performance of the trainee through intelligent algorithms, automatically assesses their weak points, and generates personalized training scenarios according to the actual situation of the trainee. For example, the system can identify and strengthen students who always make the same mistakes, automatically generate corresponding deepening training scenarios, and help students make up for their shortcomings. This personalized adaptation greatly improves training efficiency and effectiveness, ensuring that each trainee receives the most suitable training within their level and ability range.

[0056] 3. The dynamic generation mechanism of the present application can continuously adjust the difficulty of the scene and task during the training process, avoiding the predictability of the students, maintaining a high level of challenge and unpredictability. Trainees operate in a constantly changing environment, not only improving their ability to respond to actual high-pressure situations, but also effectively enhancing their sense of immersion and tension in training, prompting students to perform at their best in a more challenging environment.

[0057] 4、The application generates training scenes in real time, so that the system can evaluate the operation performance of the trainee in different situations, especially when facing nonlinear and unexpected events, and more accurately assess their emergency handling ability and psychological stress resistance. This provides more comprehensive data support for the evaluation of training effect and the progress of students, and helps to optimize the subsequent training plan and improve the ability of students. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0059] Fig. 1 The step flow chart of the method of the application;

[0060] Fig. 2 The step flow chart of how the method of the application obtains operation precision;

[0061] Fig. 3 The step flow chart of how the method of the application real-time couples the dynamic event with the physical system, sound system and visual system of the virtual environment. DETAILED DESCRIPTION

[0062] The application will be described in detail below in combination with the drawings and specific embodiments. It should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement some known technologies; and the drawings are only used to more specifically describe the embodiments, and are not intended to specifically limit the application.

[0063] Please refer to Figs. 1-3 The embodiment of the application provides a virtual reality immersive practical training content generation method. The core of constructing the domain knowledge graph is to extract atomic knowledge elements representing operation actions, device entities, state parameters and safety rules from the target training field specification documents and case library. The connections between these atomic knowledge elements are established through multi-dimensional logical relationships (such as cause-effect relationship, time sequence relationship, hierarchical relationship and probability relationship), so as to form and store the domain knowledge graph. This graph serves as a structured representation of knowledge, providing a systematic knowledge base for subsequent dynamic content generation.

[0064] The domain knowledge graph provides comprehensive knowledge support for the system, which can accurately simulate and reproduce complex operations and safety rules in the target training field, so that the subsequent dynamic training content is more in line with actual needs.

[0065] The operation behavior data stream of the trainee is collected in real time through the virtual reality interactive device, including the identification, type, timing and accuracy of the virtual device operation, and the physiological data stream (such as heart rate variability, pupil diameter change rate, and gaze focus duration) is collected in real time through the biological sensor in the head-mounted device. These data provide comprehensive support for the real-time state of the trainee.

[0066] This step ensures that the system can understand the trainee's operation performance and physiological response in real time, and further provides data basis for personalized training content generation, making the training scene more suitable for the actual state of the trainee.

[0067] Based on the operation behavior data stream of the trainee, the operation proficiency and fluency of the trainee are calculated, the knowledge blind spots of the trainee are identified, the tension and attention concentration are derived from the physiological data stream, and the ability state dimension and psychological state dimension are fused to form the trainee state model.

[0068] The construction and updating of the trainee state model can dynamically adjust the training content according to the performance of each trainee, accurately identify the weak links of the trainee, and intensively train the trainee, thereby improving the training effect.

[0069] The knowledge blind spot set in the trainee state model is matched with the domain knowledge graph to find the relevant unmastered knowledge subgraph; at the same time, the psychological state (such as tension and attention concentration) of the trainee is matched with the pre-defined event psychological requirements. Based on these information, the system can infer the trigger probability, type, timing and intensity of the potential dynamic event, and generate content injection instructions.

[0070] Through accurate dynamic decision-making, the system can automatically adjust the difficulty and situation of the training content according to the actual state and psychological reaction of the trainee, ensuring that the challenge in the training process matches the bearing capacity of the trainee, and improving the training effect.

[0071] According to the generated content injection instructions, the event script and resources are called from the virtual asset library, the dynamic event is instantiated and coupled with the physical system, sound system and visual system of the virtual environment in real time, and then injected into the current practical training process to form the self-adaptive evolving nonlinear practical training content.

[0072] Through the dynamically generated nonlinear practical training content, the trainee not only can train in a changing situation, but also can experience a real and high-pressure operation environment. This flexible and evolving training method greatly enhances the immersion, authenticity and adaptability of the training.

[0073] The application solves the defects of existing VR training systems in terms of authenticity, personalized adaptation and immersion through an intelligent and dynamic training content generation method, improves the practicality of training and the emergency handling ability of students, and enhances the performance evaluation of students in real high-pressure situations, providing an innovative solution for the development of future VR training systems.

[0074] In one possible implementation, the causal relationship is used to describe the causal connection between operations. In this method, the frequency of each causal relationship is calculated by analyzing the statistical frequency of the cause event (such as operation error) leading to the result event (such as failure, accident) in historical cases, and the validity of the frequency is corrected and evaluated in combination with the field experts. Finally, based on these data, the causal strength weight factor is determined together to evaluate the importance of the causal relationship.

[0075] This method ensures the accuracy and reasonableness of the causal relationship through the combination of historical cases and expert evaluation, improves the reliability of the causal relationship in the knowledge graph, and makes the training content more in line with the actual situation.

[0076] The temporal relationship is used to describe the operation sequence and time constraints. In this method, the temporal constraint weight factor is derived by calculating the contribution of the operation that violates the temporal relationship in historical failure cases. By analyzing the criticality level of the operation (such as whether it is an important step) and combining the importance of the operation in the overall process, the final temporal constraint weight factor is obtained after weighting.

[0077] This weighted calculation method can effectively identify the key temporal constraints in operations and provide more accurate temporal arrangements for students in training, reducing operation errors and improving the smoothness of the operation process.

[0078] The probability relationship describes the likelihood of an event occurring. In this step, the system calculates the conditional probability of the derived event (such as side effects or secondary failures) actually occurring after the main event, and adjusts the expected occurrence intensity of the conditional probability in the training environment in combination with expert knowledge. In this way, the system can assign a reasonable probability occurrence weight factor to each probability relationship.

[0079] By combining historical data and expert adjustments, potential derivative events in the training environment can be more accurately simulated, providing more detailed and dynamic training scenarios for students and improving their emergency response capabilities.

[0080] In order to effectively manage and store the above multi-dimensional logical relationships, the domain knowledge graph adopts an attribute graph model format. In this model, nodes represent atomic knowledge elements, edges represent various multi-dimensional logical relationships, and the attribute field of the edge is used to store the weight factors of the corresponding causal strength, temporal constraints and probability occurrence.

[0081] The attribute graph model can clearly present knowledge elements and their relationships. The weight factors of the edges enhance the expressive power of the knowledge graph, enabling accurate modeling and real-time updates of complex operational behaviors, and ensuring that the system can flexibly adapt to different training needs.

[0082] This invention introduces weighting factors based on multi-dimensional logical relationships into a domain knowledge graph, enabling virtual reality training content to be dynamically adjusted based on historical data and expert experience, ensuring the personalization, accuracy, and efficiency of the training content. This data-driven and expert-adjusted weighting mechanism not only improves the practicality and effectiveness of the training but also enhances the system's intelligence, providing trainees with a more challenging and adaptable training environment.

[0083] In one possible implementation, an ideal operational trajectory curve is first defined for each standard operational action (such as robotic arm operation, surgical tool use, etc.). This curve consists of a series of continuous spatial coordinate points, representing the precise path of the operation under ideal conditions. The definition of the ideal operational trajectory relies on the knowledge of domain experts and standard operating procedures to ensure high precision and operational standardization.

[0084] By clearly defining the ideal trajectory, a benchmark can be provided for evaluating operational accuracy, which helps to standardize the training process and evaluation criteria.

[0085] While the trainee performs the operation, the system records the actual movement trajectory of the instrument in virtual space in real time. Using virtual reality technology, the system can accurately track the positional changes of the instrument and sample the actual coordinate sequence. This data can be collected through sensors, cameras, or motion capture technology.

[0086] By accurately capturing the trainee's operational trajectory, a comprehensive record of the actual operation process is ensured, enabling dynamic monitoring and feedback of the trainee's operational level.

[0087] Next, the average deviation distance in three-dimensional space between the coordinates of the trainee's actual operating trajectory and the ideal operating trajectory curve is calculated. The deviation distance refers to the spatial distance between the actual trajectory and the ideal trajectory, typically calculated using Euclidean distance or other suitable distance calculation methods. The calculation result shows the degree of difference between the trainee's operating trajectory and the ideal trajectory, reflecting the accuracy of the operation.

[0088] This process provides a precise numerical value by quantifying the error, which can clearly reflect the deviation in the operation and serve as an important basis for judging the accuracy of the operation.

[0089] By inputting the average deviation distance into a predefined mapping function, which converts the deviation distance into a normalized accuracy score. The mapping function is designed to make the accuracy score negatively correlated with the deviation distance, i.e. the smaller the deviation distance, the higher the accuracy score, and vice versa. This mapping function can be linear or nonlinear, adjusted according to actual needs.

[0090] The normalized accuracy score provides a quantitative index that is easy to understand and compare, which can be used for real-time feedback, evaluating the operation level of the trainee, and adjusting the training difficulty or personalized learning path according to the accuracy score.

[0091] This operation accuracy evaluation method effectively improves the precision, personalization and adaptability of virtual reality training, can significantly improve the training quality, and promote the skill improvement of the trainee.

[0092] In one possible implementation, before the training begins, in order to ensure the individualization and accuracy of the stress index, the heart rate variability and pupil diameter change rate of the trainee need to be measured first. During the pre-set calm period, usually when the trainee is completely relaxed and does not perform any training operation, the heart rate variability and pupil diameter change are recorded. By calculating the average value of these two physiological parameters, the baseline level of each trainee is obtained.

[0093] This process can provide individualized reference standards for subsequent stress evaluation based on the initial physiological state of the trainee, ensuring that the stress index has high accuracy between different trainees.

[0094] During the training process, the system monitors the changes in the trainee's heart rate variability in real time. Heart rate variability reflects the regulation ability of the autonomic nervous system, and under normal circumstances, heart rate variability will decrease as stress increases. Therefore, the system calculates the percentage decrease between the current heart rate variability and the baseline level.

[0095] This calculation can reflect the trainee's stress level in real time, thereby providing dynamic data support for subsequent stress evaluation.

[0096] The pupil diameter change rate is another important physiological indicator, and the pupil usually dilates under stress or pressure. Therefore, the system monitors the changes in the trainee's pupil diameter in real time, and calculates the percentage increase in pupil diameter relative to the baseline level.

[0097] Changes in pupil diameter reflect the physiological stress response of the trainee, so this indicator helps to complement the stress state that heart rate variability cannot fully capture, improving the comprehensiveness of stress evaluation.

[0098] Next, the percentage decrease in heart rate variability and the percentage increase in pupil diameter change rate are multiplied by a preset contribution coefficient respectively, and then added together to obtain a preliminary stress synthesis value. The contribution coefficient is set according to physiological research and practical requirements to ensure that the weights of heart rate and pupil changes are appropriate.

[0099] By synthesizing heart rate and pupil changes, the resulting stress synthesis value can comprehensively reflect the stress state of the trainee, avoiding the limitations of a single physiological indicator.

[0100] In order to convert the stress synthesis value into a standardized stress index, the system inputs the preliminary synthesis value into an S-shaped function for normalization processing. S-shaped functions are commonly used to smooth nonlinear data and can effectively map the value range to between 0 and 1, where 0 represents the lowest stress and 1 represents the highest stress. The center point position of the S-shaped function is fine-tuned according to the age and gender of the trainee, which can better adapt to the physiological differences of different trainees.

[0101] The use of S-shaped functions ensures smooth transitions in stress indices, making stress changes more consistent with human physiological and psychological actual performance, and having good interpretability.

[0102] After normalization processing, the system outputs the final stress index, which can accurately reflect the physiological stress of the trainee during the training process.

[0103] By quantifying stress, trainees and trainers can real-time understand the psychological and physiological reactions of trainees, help adjust training intensity and content, avoid excessive stress or anxiety, and improve training effectiveness and learning efficiency.

[0104] This method can effectively improve the performance evaluation and training adaptability of students in virtual reality training, helping students train in an environment with moderate stress and appropriate pressure, thereby improving learning efficiency and the quality of skill mastery.

[0105] In one possible implementation, first, using the domain knowledge graph, the system can extract the most relevant set of key device entities according to the specific stage of the current training task. The domain knowledge graph contains information about devices, tools, operations, and other information related to the training task, which can help the system determine which devices are most important in the current task.

[0106] This step ensures that the attention concentration evaluation is closely related to the actual needs of the training task, avoiding irrelevant object interference, making the evaluation results more targeted and practical.

[0107] Through the built-in eye tracking module of the virtual reality headset, the system can continuously obtain the three-dimensional coordinates of the trainee's gaze focus in the virtual environment. The eye tracking module can accurately capture the direction and position of the trainee's eyes, ensuring the accuracy and real-time nature of the gaze data.

[0108] The eye tracking technology makes the evaluation process dynamic in real-time, accurately reflecting the trainee's attention distribution at different time periods, and avoiding the errors and lag of traditional manual monitoring.

[0109] After obtaining the eye tracking data in real-time, the system determines whether the trainee's gaze focus coordinates intersect with the three-dimensional model of the key device entity through a collision detection algorithm. If there is an intersection, the system records the residence time of the gaze focus, i.e., the duration of the trainee's gaze on the device.

[0110] This technology can accurately quantify the trainee's attention time on the key device, avoiding errors in human observation and estimation, making the evaluation of attention concentration more objective and accurate.

[0111] Within a adjustable time window, the system calculates the ratio of the total duration of the gaze focus on the key device entity to the total duration of the residence in the non-key environment area. This ratio reflects the degree to which the trainee focuses on the key device during the training process. The setting of the time window can be adjusted according to the characteristics of the training task to adapt to the evaluation needs of different tasks.

[0112] Through the calculation of the ratio, the system can clearly show whether the trainee can effectively concentrate on the task-related device, helping to optimize the training content and improve the training effect.

[0113] In order to provide more personalized attention evaluation, the system compares the current calculated gaze focus ratio with the baseline ratio measured during the calm period for the trainee. The baseline ratio represents the normal attention distribution of the trainee in a relaxed state. By calculating the relative change rate between the two, the system can reflect the changes in the trainee's attention during the training process.

[0114] The introduction of the baseline ratio allows the attention concentration index to be dynamically adjusted according to individual differences of the trainee, ensuring the accuracy and personalization of the evaluation.

[0115] Finally, the relative change rate is mapped to the interval between 0 and 1 through a linear normalization function to obtain the final attention concentration index. The higher the index value, the more concentrated the trainee's attention, and vice versa, indicating that the trainee may be distracted or lack attention.

[0116] Through normalization processing, the final attention concentration index has a unified standardized range, making the evaluation results between different trainees comparable. In addition, the application of linear normalization function can also smooth the data, eliminate the influence of extreme values, and improve the stability and reliability of the evaluation results.

[0117] This method not only improves the participation and attention of trainees in virtual reality training, but also provides real-time and accurate feedback for trainers, thereby helping to optimize training strategies and improve training effectiveness.

[0118] In one possible implementation, the system first acquires the trainee's operation behavior data stream in real time through the virtual reality training platform, including information such as each device operation, operation sequence, and operation duration. Subsequently, these operation behaviors are compared in real time with the current expected correct operation sequence derived from the domain knowledge graph. The domain knowledge graph stores the association between each operation step and the corresponding atomic knowledge element and the operation timing rules.

[0119] This step realizes the whole-process monitoring and dynamic evaluation of operation behavior, ensuring that the correctness and compliance of each operation can be captured by the system in real time, providing basic data support for knowledge blind spot identification.

[0120] When the system detects that the operation performed by the trainee does not match the expected operation sequence, or that there is a missing operation, it automatically records the atomic knowledge element identifier corresponding to the operation and its error type (e.g., operation error, omission, or sequence error).

[0121] Accurate recording of each operation error and its type can clearly distinguish the mastery of different knowledge points, providing reliable basis for subsequent statistical analysis and blind spot identification.

[0122] The system sets an initial error count threshold for each atomic knowledge element, and the error count is incremented once when the same error type repeatedly occurs on the same atomic knowledge element.

[0123] By accumulating and counting repeated errors, the system can identify the difference between occasional mistakes and repeated knowledge blind spots, improving the accuracy of knowledge blind spot identification.

[0124] The error count threshold is not fixed, but is dynamically adjusted according to the criticality of the operation in the timing relationship. For critical operations, the threshold is set lower, so that a small number of errors can trigger knowledge blind spot identification; for non-critical operations, the threshold is relatively high, avoiding excessive labeling.

[0125] Dynamic threshold setting ensures that key knowledge points are identified first, improving the system's sensitivity to weak links in critical operations, while reducing the interference of non-critical operations on the blind spot set, improving the relevance of training evaluation.

[0126] When the count of a certain error type exceeds its dynamically adjusted threshold, the system officially adds the atomic knowledge element corresponding to the error type to the knowledge blind spot set. This set can be updated in real time, forming a dynamic portrait of the trainee's weak points.

[0127] The knowledge blind spot set provides precise feedback for the trainer to develop targeted remedial measures, and can be used for personalized training content generation to improve trainee learning efficiency and practical training effectiveness.

[0128] This method realizes knowledge blind spot identification driven by operation behavior data in virtual reality training, providing a reliable technical means for precise training and evaluation.

[0129] In one possible implementation, first, the system starts from the knowledge subgraph that the trainee has not mastered, and traverses all potential dynamic event nodes connected by causal and probabilistic relationships. Each potential dynamic event is composed of multiple paths, and each edge on the path has a probability-related weight factor. According to the conditional probability rule, these weight factors are multiplied to obtain the initial trigger probability of each potential dynamic event.

[0130] Next, the system will correct the trigger probability according to the trainee's proficiency score in the operation process. The higher the proficiency, the better the trainee has mastered the knowledge point, and the trigger probability will be adjusted higher to provide intensive training for the trainee. Conversely, for knowledge points that are not well mastered, the trigger probability will be correspondingly reduced, thereby avoiding excessive challenges for the trainee.

[0131] This step realizes dynamic adjustment according to the trainee's state, making the training content more suitable for the trainee's actual learning progress. By correcting according to the trainee's proficiency, the training content is precisely matched, neither too simple nor too difficult, which helps to improve the trainee's learning motivation and sense of participation.

[0132] After calculating the trigger probability, the system selects a predetermined number of potential dynamic events with the highest trigger probability based on the corrected trigger probability to form a set of candidate events. The core of this process is to identify the most challenging training tasks that the trainee may currently face through probability evaluation.

[0133] By selecting the most challenging and potentially high learning improvement dynamic events, the system can accurately guide the trainee to focus on weak links and improve learning efficiency.

[0134] After determining the set of candidate events, the system analyzes the temporal constraints between these events and the current operational process, selecting the most suitable triggering time. This timing is evaluated based on the event's temporal constraint weighting factor, and the system selects the legal time point closest to the current moment as the event's triggering time. This step takes into account the temporal requirements of the actual operational process to ensure that the event triggering is consistent with the logic of the student's operation.

[0135] Choosing the right timing for triggering actions can prevent confusion or disruption to trainees' workflow due to inappropriate timing. By selecting appropriate times, trainees can more naturally immerse themselves in the training, thereby enhancing their sense of engagement.

[0136] Once the triggering timing is determined, the system will determine the intensity of the event based on its attribute range. This intensity determination is also closely related to the learner's current stress level. When the learner's stress level is low, the system will select a higher value in the event intensity range to provide a certain level of challenge; while when the learner's stress level is high, the system will reduce the event intensity to avoid situations where the learner is overly stressed or unable to cope.

[0137] Furthermore, to incentivize high-level learners, the system introduces motivational factors that are positively correlated with their operational proficiency scores. Even under moderate stress, highly skilled learners will still face significant challenges, driving them to further improve.

[0138] By adjusting the intensity based on the trainee's level of tension and proficiency, the system can provide more personalized training content. This dynamic intensity adjustment not only maintains the trainee's interest but also effectively improves their adaptability and operational skills in real-world situations. High-level trainees can also receive appropriate challenges, further developing their comprehensive abilities.

[0139] In one possible implementation, when a dynamic event is instantiated, a physical parameter perturbation value matching the event type is first generated. The magnitude of this perturbation value is determined by the event intensity parameter; the greater the event intensity, the stronger the perturbation value. This perturbation value is then sent to the virtual environment's physics engine, which calculates the cascading effects of the perturbation on the state parameters of the relevant virtual device entities based on its built-in simulation model. For example, suppose in a virtual training scenario, a trainee needs to interact with a virtual mechanical device. When an event occurs (such as device vibration or collision), the physics engine simulates the impact of the vibration or collision on the device, updates the device's motion state, position, speed, etc., and reflects this information throughout the virtual environment.

[0140] Through physical system coupling, devices and objects in the virtual environment can accurately respond to trainees' actions and environmental events, enhancing the realism and immersion of the training. Trainees can perceive the physical changes triggered by each action or event, improving the interactivity and educational effectiveness of the training.

[0141] The key to sound system coupling is selecting appropriate sound effect files from the sound asset library based on the type, intensity, and three-dimensional spatial location of dynamic events. These sound effects are adaptively adjusted according to the perturbation results provided by the physics engine. Specifically, the volume, pitch, and spatial surround effects of the sound are adjusted based on the magnitude of the physical perturbation and the location of the event. For example, when a strong collision event occurs, the volume and pitch of the sound may increase, and the spatial surround effect of the sound will also be adjusted accordingly based on the student's position to enhance the sense of presence. Finally, the adjusted sound effects are played to the students through a 3D audio rendering engine.

[0142] Adaptive sound adjustment can greatly enhance learners' perception of events, making the virtual reality environment more immersive and realistic. By dynamically adjusting sound effects, learners can more intuitively feel the intensity and impact of events, and the spatial positioning of sound effects also helps learners locate the specific location where the event occurs, enhancing their environmental awareness.

[0143] Visual system coupling is achieved by selecting appropriate particle effect templates and material change scripts from the visual effects asset library based on the type and intensity of dynamic events. When an event occurs, the system combines updated device state parameters provided by the physics engine to drive the particle emitter's rate and range, as well as the shader parameters of the material surface, generating corresponding visual effects. For example, assuming a strong explosion event occurs, the system will select particle effects suitable for the explosion, adjust the particle emission rate and range according to the event's intensity, and update the material's color, gloss, and other properties based on the device state, ultimately generating the visual effect and submitting it to the rendering pipeline for rendering and display.

[0144] Real-time generation and adjustment of visual effects can provide trainees with more realistic and intuitive event feedback, enhancing their immersion in the virtual environment. Through particle effects and material changes, virtual events go beyond simple visual representation; they can provide more dynamic and richer visual information as events change, thereby improving training effectiveness.

[0145] Real-time coupling of physical, acoustic, and visual systems not only enriches the training content but also ensures a high degree of realism and interactivity in the virtual reality environment, helping learners to conduct effective skills training in a more natural and realistic environment.

[0146] In one possible implementation, the system first records the trainee's complete sequence of responses to injected dynamic events in the virtual environment, including each step and its final result. For example, when faced with a sudden event, the trainee might take a series of actions, such as pressing a button, rotating a controller, or operating a virtual device. The sequence and results of all these actions are recorded in detail by the system.

[0147] The system uses the event tracking mechanism of the virtual reality training platform to capture the trainee's operational data each time they interact with the virtual environment. This data is packaged into a response sequence and compared with a pre-stored ideal response plan.

[0148] The recorded response sequence is compared with pre-defined ideal responses in the knowledge graph. These ideal responses are typically constructed by domain experts based on past experience, best practices, and historical failure cases. After the comparison, the system generates a response effectiveness score, which reflects the trainee's handling of the event and efficiency. If the response effectiveness score is lower than the pre-defined standard (e.g., an incorrect sequence of operations, inefficiency, or failure to effectively resolve the event), the system initiates the next step of knowledge graph updates.

[0149] The performance score is calculated based on a quantitative assessment of pre-defined coping strategies. The comparison process uses a knowledge graph engine to automatically analyze the sequence of actions, assess whether the trainee's response meets the standards, and provide a score.

[0150] During dynamic event handling, the system extracts new operational sequences, causal logic, or state changes employed by the trainee. This new information may not be entirely consistent with the existing knowledge graph, indicating that the trainee may have proposed some innovative or effective new solutions. The system abstracts these new discoveries into candidate objects of atomic knowledge elements or logical relation edges.

[0151] This process requires the system to have strong data mining and analysis capabilities, which can identify new operational steps, causal relationships and state changes through machine learning algorithms, and transform them into standardized knowledge elements through automated tools.

[0152] When the response performance score falls below a preset level, the system submits extracted candidate objects to the knowledge graph management terminal. Domain experts then review and confirm these candidates. If approved, these new knowledge points or failure cases are added to the domain knowledge graph, thus enabling adaptive growth of the system's knowledge.

[0153] After candidates are submitted, the knowledge graph management platform will conduct manual verification through an expert review interface to ensure that the newly added knowledge points are valuable and reliable. After expert verification, the new knowledge will be integrated into the knowledge graph, enabling the graph to be continuously updated and optimized.

[0154] By evaluating training effectiveness and adaptively evolving the knowledge graph, we can not only optimize the virtual training process and improve trainees' training results, but also continuously enrich and update the system's knowledge base, thereby enhancing the overall level of intelligence.

[0155] In one possible implementation, the target training areas include high-voltage line fault troubleshooting in power grid operation and maintenance, emergency shutdown procedures for high-risk reactors in chemical production, and endoscopic minimally invasive surgery in the medical field. These areas typically involve high-risk, high-pressure environments where operators need to make rapid decisions and take precise operational measures within a limited time. Virtual reality (VR) training content generation methods are specifically designed to meet the unique needs of these industries, creating highly realistic and challenging simulation environments that allow trainees to react in simulated emergency situations, helping them improve their decision-making abilities and operational accuracy.

[0156] In the field of power grid operation and maintenance, the training system can simulate different types of high-voltage line faults, such as power equipment short circuits, grounding faults, and line breaks, and generate diverse fault scenarios. The system creates realistic fault scenarios by dynamically adjusting parameters such as voltage and current in the virtual environment, forcing trainees to quickly determine the cause of the fault and take appropriate measures (e.g., cut off the power supply, adjust the switch, or locate the fault point) within a limited time.

[0157] Virtual reality technology is used to recreate various emergencies in high-pressure environments. Combined with sensors and feedback devices, the trainee's operational behavior is monitored in real time, and performance is evaluated.

[0158] In the chemical industry, malfunctions in high-risk equipment such as reaction vessels can lead to serious safety accidents. Virtual training systems simulate emergency shutdown procedures for reaction vessels under conditions of excessively high temperature, excessively high pressure, or runaway chemical reactions. Trainees must make rapid decisions, initiate emergency shutdown procedures, and operate critical equipment such as emergency cooling and gas venting systems to prevent the accident from escalating.

[0159] The system helps trainees develop emergency response thinking by modeling various possible emergencies in chemical production processes (such as reactor explosions and gas leaks) and by monitoring the accuracy of operational behavior in the virtual environment in real time.

[0160] Endoscopic minimally invasive surgery demands precise operation; any improper operation can lead to serious injury to the patient. Virtual reality training systems can simulate complex endoscopic surgical scenarios, such as laparoscopic and thoracic surgeries. Trainees need to perform precise operations using virtual surgical instruments in a real-time feedback environment, such as removing tumors or foreign objects, while ensuring the safety and accuracy of the operation.

[0161] Through VR technology and force feedback devices, the tactile sensation of surgical instruments, the mechanical feedback during the operation, and the physiological reactions of patients are simulated, allowing trainees to experience the real surgical procedure.

[0162] In the virtual training sessions across these high-risk areas, the system dynamically adjusts the training difficulty and scenario complexity based on the trainee's real-time actions and decisions. For example, when a trainee makes a mistake, the system generates more complex emergencies or increases time pressure to test the trainee's emergency response capabilities under higher pressure. Conversely, when a trainee performs well, the system introduces more complex scenarios and more challenging tasks to enhance their ability to cope with high-risk environments.

[0163] By leveraging deep learning and data analytics, the system can record the details of each operation and automatically adjust the difficulty of the training content based on the trainee's performance, ensuring that each training session promotes the trainee's skill improvement.

[0164] The virtual reality immersive training content generation method for high-risk industries can not only provide operators with a highly immersive and challenging training environment, but also significantly improve their decision-making ability and operational accuracy under high-pressure conditions, thereby reducing safety risks in actual work and ensuring operational safety.

[0165] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0166] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A virtual reality immersive practical training content generation method, characterized in that, The method comprises the following steps: The field knowledge graph construction step: atomic knowledge elements representing operation actions, device entities, state parameters and safety rules are extracted from the specification documents and case library of the target training field, and multi-dimensional logical relationships including causal relationships, time sequence relationships, hierarchical relationships and probability relationships between the atomic knowledge elements are established to form a field knowledge graph and store it; The real-time trainee data collection step: operation behavior data streams of the trainee are collected in real time through a virtual reality interaction device, the operation behavior data streams including the identification of the virtual device object being operated, the type of operation action, the time sequence of operation action and operation accuracy; The real-time trainee physiological data stream collection step: physiological data streams of the trainee are collected in real time through a biological sensor integrated in the virtual reality head-mounted device, the physiological data streams including heart rate variability, pupil diameter change rate and gaze focus dwell time; The trainee state model construction and updating step: based on the operation behavior data streams, operation proficiency scores are obtained by calculating the completion degree and fluency degree of operation actions, and a knowledge blind spot set is obtained by identifying repeatedly ignored or incorrectly executed operation sequences, which together with the operation proficiency scores constitute the ability state dimension; Based on the physiological data streams, a tension index is obtained by fusing the changes of heart rate variability and pupil diameter change rate relative to the baseline, and an attention concentration index is obtained by analyzing the distribution of gaze focus on key device objects, which together with the tension index constitute the psychological state dimension; the ability state dimension and the psychological state dimension are fused into a trainee state model; The dynamic content decision step: the knowledge blind spot set in the trainee state model is matched with the field knowledge graph in terms of semantic similarity to find an associated unmastered knowledge subgraph; the tension index and the attention concentration index are matched with pre-defined event psychological requirements in the knowledge graph; based on the matching results, the triggering probability of potential dynamic events is calculated and their types, triggering time and intensity are determined according to the multi-dimensional logical relationships, and content injection instructions are generated; The triggering probability calculation: starting from the unmastered knowledge subgraph, all potential dynamic event nodes connected through causal relationships and probability relationships are traversed, and the probability occurrence weight factors of all edges on the path are multiplied according to the conditional probability rule to obtain the initial triggering probability of each potential dynamic event; The operation proficiency score in the trainee state model is then used to modify the initial triggering probability, and the higher the proficiency, the higher the triggering probability after modification for reinforcement training; The nonlinear practical training content generation and injection step: according to the content injection instructions, corresponding event scripts and resources are called from a pre-stored virtual asset library, a specific dynamic event is instantiated in the virtual reality environment, and the dynamic event is coupled in real time with the physical system, sound system and visual system of the virtual environment and then injected into the current practical training process, thereby forming a self-adapting evolving nonlinear practical training content. 2.The virtual reality immersive training content generation method of claim 1, wherein, The multi-dimensional logical relationship between the atomic knowledge elements comprises: a causal strength weight factor is assigned to the causal relationship, which is determined by analyzing the statistical frequency of the cause event leading to the result event in historical cases, and is modified by combining the effectiveness of the frequency evaluated by domain experts; a temporal constraint weight factor is assigned to the temporal relationship, which is obtained by calculating the contribution of the operation violating the temporal relationship in historical failure cases, and is weighted by combining the criticality level of the operation in the overall process; a probability occurrence weight factor is assigned to the probability relationship, which is obtained by counting the conditional probability of the derived event indicated by the relationship actually occurring after the main event, and introducing expert knowledge to adjust the expected occurrence strength of the conditional probability in the training environment; the storage format of the domain knowledge graph is a property graph model, in which nodes represent the atomic knowledge elements, edges represent the multi-dimensional logical relationships, and the attribute field of the edge is used to store the corresponding weight factor. 3.The virtual reality immersive training content generation method of claim 1, wherein, the operation accuracy is obtained by the following process: an ideal operation trajectory curve is defined for each standard operation action in advance, which is composed of a series of continuous spatial coordinate point sequences; when the trainee performs the operation, the actual motion trajectory of the operating instrument in the virtual space is recorded, and the actual coordinate point sequence is sampled; the average deviation distance between the actual coordinate point sequence and the ideal operation trajectory curve in three-dimensional space is calculated, and the average deviation distance is converted into a standardized accuracy score through a pre-defined mapping function, which ensures that the accuracy score is negatively correlated with the deviation distance. 4.The virtual reality immersive training content generation method of claim 1, wherein, the tension index obtained by fusing the change of heart rate variability and the change rate of pupil diameter relative to the baseline includes: during a pre-set calm period before the training, the heart rate variability and the change rate of pupil diameter of the trainee are measured, and their average values are calculated as individual baseline levels; during the training process, the percentage decrease of heart rate variability relative to its individual baseline level is calculated in real time; the percentage increase of the change rate of pupil diameter relative to its individual baseline level is calculated in real time; the decrease percentage and the increase percentage are multiplied by a pre-set contribution coefficient respectively, and then added to obtain a preliminary tension synthesis value; the preliminary tension synthesis value is input into an S-shaped function for normalization processing, and its value range is mapped to between zero and one, and the output is the final tension index, and the center point position of the S-shaped function is adjusted according to the norm data corresponding to the age and gender of the trainee.

5. The virtual reality immersive practical training content generation method of claim 1, wherein, the attention concentration index obtained by analyzing the distribution of the gaze focus on the key device object includes: according to the domain knowledge graph, the key device entity set most relevant to the current training task stage is extracted; the coordinates of the trainee's gaze focus in the three-dimensional space of the virtual environment are continuously obtained through the built-in eye tracking module of the virtual reality headset; whether the gaze focus coordinates intersect with the three-dimensional model of any entity in the key device entity set is judged by a collision detection algorithm, and the residence time is recorded. calculating the ratio of the total time length of the gaze focus on the key device entities and the total time length of the gaze focus on the non-key environment areas within an adjustable time window; comparing the ratio with the baseline ratio measured during the calm period, calculating the relative change rate, and mapping the change rate to the interval between zero and one through a linear normalization function to obtain the attention concentration index, the higher the index value, the more concentrated the attention.

6. The virtual reality immersive practical training content generation method of claim 1, wherein, the knowledge blind spot set is obtained by identifying repeatedly ignored or incorrectly executed operation sequences, including: continuously monitoring the operation behavior data stream and comparing it with the current expected correct operation sequence derived from the domain knowledge graph; when an operation action error or omission is detected, recording the corresponding atomic knowledge element identifier and its error type; setting an initial error count threshold, and each time the same error type occurs again for the same atomic knowledge element, the error count is incremented by one; the error count threshold is not a fixed value, but is dynamically adjusted according to the criticality of the operation in the time sequence, the higher the criticality of the operation, the lower the corresponding error count threshold is set, so that the critical knowledge blind spot can be identified more quickly; when the count of a certain error type exceeds its dynamically adjusted error count threshold, the atomic knowledge element associated with the error type is officially added to the knowledge blind spot set.

7. The virtual reality immersive practical training content generation method of claim 1, wherein, the triggering probability of the potential dynamic event is calculated and its type, triggering time and intensity are determined, including: type determination: selecting a predetermined number of potential dynamic events with the highest corrected triggering probability to form a candidate event set; triggering time determination: analyzing the time sequence constraint relationship between each event in the candidate event set and the currently executing operation process, and selecting the legal time point with the strongest time sequence constraint weight factor and closest to the current time as the triggering time; intensity determination: based on the attribute range of the candidate event itself, the intensity value is determined by linear interpolation according to the current tension index of the trainee, taking the high value of the intensity range when the tension index is low, and taking the low value when the tension index is high, but at the same time introducing an incentive factor positively correlated with the operation proficiency score, so that high-level trainees will also receive higher intensity challenges even under moderate tension. 8.The virtual reality immersive training content generation method of claim 1, wherein, the dynamic event is coupled with the physical system, sound system and visual system of the virtual environment in real time, including: physical system coupling: when the dynamic event is instantiated, a physical parameter disturbance value conforming to its event type is generated, and the disturbance value size is determined by the event intensity parameter; the disturbance value is sent to the physical engine of the virtual environment, and the physical engine calculates the chain effect of the disturbance on the state parameters of the related virtual device entities according to the built-in simulation model, and updates the physical state of the entire virtual environment; sound system coupling: according to the type, intensity and three-dimensional spatial position of the dynamic event, the corresponding sound effect file is called from the sound asset library, and based on the parameter disturbance result provided by the physical engine, the volume, pitch and spatial surround effect of the sound effect are adaptively adjusted, and then played through the three-dimensional audio rendering engine; Visual system coupling: according to the type and intensity of the dynamic event, the corresponding particle special effect template and material change script are called from the visual effect asset library, and combined with the updated device state parameters provided by the physical engine, the rate, range and material surface shader parameters of the particle emitter are driven to generate corresponding visual special effects and submit to the rendering pipeline. 9.The virtual reality immersive practical training content generation method of claim 1, wherein, After the step of generating and injecting nonlinear practical training content, further comprising: Training effect evaluation and knowledge graph self-evolution step: record the complete response operation sequence of the trainee to the injected dynamic event and its final result; compare the response operation sequence with the ideal coping scheme pre-stored in the knowledge graph to generate a coping efficiency score; extract all new operation sequence relationships, causal logic or state changes involved in the disposal process of the dynamic event and abstract them as new atomic knowledge elements or candidate objects of logical relationship edges; when the coping efficiency score is lower than the preset level, submit the candidate objects to the knowledge graph management terminal, and after the confirmation of the field experts, integrate them into the field knowledge graph as new knowledge points or fault cases, thereby realizing the adaptive growth of the system knowledge ability. 10.The virtual reality immersive training content generation method of claim 1, wherein, The target training field is a high-risk industry field that requires operators to have rapid decision-making and precise operation ability in a high-pressure environment, including high-voltage line fault elimination in power grid operation and maintenance, high-risk reaction kettle emergency shutdown disposal in chemical production, or endoscopic minimally invasive surgery in the medical field.

Images