A Smart Triage Training System Based on Public Emergencies

Abstract

This invention discloses an intelligent triage training system based on public emergencies, specifically in the field of triage training, to address the lack of targeted approach in existing triage training. The system utilizes a tree-building module to retrieve historical records of public emergencies and construct a time-series tree of triage strategies; an error identification module to label response error factors and build an error path library; an interference generation module to convert these error factors into interference options in a simulated training process, with an interference setting module configuring their interference induction weights and time pressure parameters; a blind spot identification module to collect response data from trainees, constructing an error trend intensity curve to identify triage error blind spots; and finally, an error training module to use these identified blind spots as the focus of training for the current training cycle. This system can intelligently identify and target weak points in triage cognition, improving the accuracy and timeliness of emergency response.

Description

Technical Field

[0001] This invention relates to the field of triage training technology, and more specifically, to an intelligent triage training system based on public emergencies. Background Technology

[0002] In contemporary society, where public health emergencies are frequent, the rapid response capability and triage efficiency of the medical system have become core aspects of public health emergency management. In emergency situations such as major epidemics, natural disasters, and mass casualty incidents, the limited allocation of medical resources, the complexity of patient conditions, and the high density of arrivals make accurately identifying priority patients and guiding them to the appropriate departments within a short period crucial to treatment effectiveness and resource utilization. However, most current emergency training systems still rely on traditional methods such as experience replication and simulation exercises, lacking structured modeling and intelligent auxiliary training mechanisms for specific triage decision-making processes. This results in fragmented training content, incomplete scenario coverage, and delayed identification of triage blind spots, making it difficult to comprehensively improve the triage strategy decision-making capabilities of medical personnel under high-pressure situations. Furthermore, existing systems generally ignore the impact of interfering and inducing factors on actual triage errors, such as cognitive errors, suggestive information prompts, or multitasking pressure. These irrational factors, which exist in real-world emergency situations, are often the direct causes of incorrect triage and waiting delays.

[0003] Therefore, there is an urgent need for an intelligent training method for triage strategies in response to public emergencies. This method should be able to reconstruct the decision-making structure of historical events, identify error-prone areas and build error-inducing models, integrate simulated interference factors for targeted error correction training, and thus form a reusable training data structure and evaluation mechanism to improve the accuracy and robustness of triage decisions. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an intelligent triage training system based on public emergencies to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A smart triage training system based on public health emergencies includes:

[0007] The tree building module is used to retrieve the historical records of emergency public events, extract the triage strategy execution records of various events, and build a triage strategy time sequence tree.

[0008] The error identification module is used to analyze the response errors of triage strategies in various events, mark error factors, and build corresponding response error paths and error path libraries.

[0009] The interference generation module is used to set the training event type, extract the corresponding error path from the error path library, convert the error factor into interference options, and insert them into the simulation training process.

[0010] The interference setting module is used to set the interference induction weight and time pressure parameters for interference options;

[0011] The blind spot identification module is used to statistically analyze the probability of training subjects hitting the interference options and the average decision time, calculate the bias tendency intensity curve under different triage strategies, and identify triage bias blind spots.

[0012] The error training module takes the triage error blind spots as the target of the current training cycle and conducts triage training.

[0013] In a preferred embodiment, the tree construction module specifically includes constructing a triage strategy time sequence tree:

[0014] Public emergencies are archived according to event type tags, and the triage decision records and recording times of the corresponding medical treatment processes are extracted.

[0015] By combining the strategy number and recording time sequence of triage decision records in events with the same type of label, event-level decision trajectory fusion is performed on triage nodes to construct a triage strategy time sequence tree for events with the same label type.

[0016] In a preferred embodiment, the error identification module, which marks error factors and constructs corresponding response error paths and an error path library, specifically includes:

[0017] In the triage strategy time sequence tree for each type of event, identify the strategy nodes where waiting delays or incorrect triage occur in the triage decision, and use the corresponding decision nodes as bias factors.

[0018] Based on the bias factor, an event decision graph is constructed, and the strategy nodes are arranged in chronological order and upstream and downstream causal relationships to form a response graph structure with unidirectional influence logic.

[0019] In the response graph, backtracking is performed starting from the node corresponding to the bias factor to extract the complete path before each bias occurs, which is taken as the strategy failure path.

[0020] Integrate all policy failure paths into a bias path library;

[0021] The strategy failure path is accompanied by actual consequence indicators, including death delay indicators, resource idleness indicators, and symptom aggravation indicators.

[0022] In a preferred embodiment, the method for identifying strategy nodes where waiting delays or incorrect triage occur in the triage decision is specifically as follows:

[0023] Traverse the triage strategy time sequence tree, obtain the time difference of each pair of adjacent tree nodes as the decision duration, preset the reaction window time for different types of events, and filter the waiting delay strategy nodes by threshold comparison method.

[0024] The triage strategy time sequence tree is decomposed into a single-chain structure using a depth-first algorithm. If a duplicate node appears in the single chain, the position of the first occurrence of the duplicate node is taken as the erroneous triage strategy node.

[0025] In a preferred embodiment, the interference generation module converts the bias factor into interference options and inserts them into the simulation training process, specifically including:

[0026] Define the target event type corresponding to the current training cycle, and establish a standardized simulated triage training process based on the historical record library of target event type tags;

[0027] Taking the policy failure path marked in the error path as input, the preset simulation training process triage behavior template library is called to generate interference operation items with policy bias, and the generated interference items are inserted into the simulation training process.

[0028] The interference operation items are generated according to two types of error logic: waiting delay logic and triage error logic.

[0029] In a preferred embodiment, the standardized simulated triage training process involves comprehensively calculating the actual consequences of each strategy failure path in all single chains after the triage strategy time sequence tree is decomposed, and selecting the single chain with the smallest calculation result value as the baseline structure.

[0030] In a preferred embodiment, the interference setting module specifically includes setting the interference induction weight and time pressure parameters of the interference options, including:

[0031] For each perturbation option, set the perturbation induction weight before inserting it into the simulation training process;

[0032] The interference-induced weight is set by the triage error frequency corresponding to the interference option. The triage error frequency is the proportion of the bias factor among the child nodes with the same parent node within the level of the strategy node corresponding to the interference option in the triage strategy time sequence tree.

[0033] Set a time pressure parameter, which is set based on the average decision time of all non-waiting delay strategy nodes to limit the effective decision time for triage;

[0034] The interference option shares the same output interface as the standard option and is activated synchronously in the simulation training process. The standard option is the correct option in the standardized simulation triage training process.

[0035] In a preferred embodiment, the blind spot identification module specifically includes identifying triage bias blind spots by:

[0036] In the simulated triage training process, the hit probability and average decision time of different trainees for all triage strategies are normalized respectively. Combined with the interference induction weight and time pressure parameter, the bias score matrix is ​​constructed according to the interference option number.

[0037] The average score matrix of all trainees is calculated according to the triage strategy dimension corresponding to the interference options to obtain the bias tendency value of each triage strategy;

[0038] By connecting all the bias tendency values, a bias tendency intensity curve is established to represent the bias distribution trend of the trainees throughout the training process. Based on the bias tendency intensity curve, bias blind spots in triage are identified.

[0039] In a preferred embodiment, the step of identifying triage bias blind spots based on the bias tendency intensity curve specifically involves extracting the peak value of the bias intensity curve during the simulation training process, obtaining the repeatability of triage strategies misjudged by the trainee under similar event labels, and marking the corresponding triage strategies as triage bias blind spots.

[0040] In a preferred embodiment, the error training module takes the triage error blind spot as the target of the current training cycle and conducts triage training specifically including:

[0041] For each type of triage error blind spot, a dedicated training process with correctly labeled options is specified, training tasks are generated, and training validation metrics are set. Validation metrics include the hit probability and average decision time of the triage strategy corresponding to the triage error blind spot.

[0042] Once both indicators meet the criteria, the triage error blind zone training and verification is deemed successful, and the blind zone task pool is automatically refreshed.

[0043] The technical effects and advantages of the intelligent triage training system based on public emergencies of this invention are as follows:

[0044] This invention achieves structured modeling and systematic identification of error causes in the triage decision-making process during public emergencies by constructing a triage strategy time sequence tree and an error path library. It can accurately reconstruct historical triage error logic, providing a highly realistic simulation foundation for training. By introducing interference options and setting induction weights and time pressure parameters, the simulation training more closely resembles real emergency situations, significantly enhancing the stress realism and strategy anti-interference capabilities of the training process. Through quantitative analysis of the trainees' response behavior, an error tendency intensity curve is constructed to accurately locate triage blind spots and dynamically generate error correction training tasks, enabling personalized identification and targeted improvement of triage capability shortcomings. The overall system possesses advantages such as clear structure, strong scalability, and a complete training feedback loop, significantly improving the triage response efficiency and decision-making accuracy of medical personnel under high-pressure scenarios, demonstrating good promotional value and practical application prospects. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the structure of an intelligent triage and training system based on public emergencies according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1

[0048] Figure 1 This invention provides an intelligent triage training system based on public health emergencies, comprising:

[0049] The tree building module is used to retrieve the historical records of emergency public events, extract the triage strategy execution records of various events, and build a triage strategy time sequence tree.

[0050] The error identification module is used to analyze the response errors of triage strategies in various events, mark error factors, and build corresponding response error paths and error path libraries.

[0051] The interference generation module is used to set the training event type, extract the corresponding error path from the error path library, convert the error factor into interference options, and insert them into the simulation training process.

[0052] The interference setting module is used to set the interference induction weight and time pressure parameters for interference options;

[0053] The blind spot identification module is used to statistically analyze the probability of training subjects hitting the interference options and the average decision time, calculate the bias tendency intensity curve under different triage strategies, and identify triage bias blind spots.

[0054] The error training module takes the triage error blind spots as the target of the current training cycle and conducts triage training.

[0055] The tree construction module constructs a triage strategy time sequence tree.

[0056] Archive historical records of public emergencies. Standardized event type labels are applied to each type of public emergency, defined based on the event's medical characteristics, occurrence scenario, and urgency. For example, labels such as "major traffic accident," "food poisoning outbreak," and "earthquake crush injury" can be set, with each archived record including this label field as an index entry. To ensure consistent label classification, a set of rules for each label must be established, such as matching keywords in event reports or registration texts to assign them their respective labels. If unclear text descriptions lead to mismatches, manual review is performed to ensure archiving accuracy. After label classification, information is extracted from the medical treatment processes involved in the archived event records. Using each event's medical treatment cycle as a unit, triage decision records are extracted. These records should include at least the following fields: strategy number, basic patient characteristics (which can be anonymized), triage options, decision-making time, and the role of the decision executor. The strategy number is a triage operation path code defined within the event, possessing stability and uniqueness; identical triage decisions are assigned the same decision number. Decision-making time is based on the standard timestamp of the operation, uniformly using a 24-hour system accurate to the second. During the extraction process, the original records undergo structural standardization cleaning. For event processing reports in paper document or natural language recording format, format standardization and data normalization are required to ensure that time and decision path data are comparable and calculable.

[0057] After obtaining a structured triage decision dataset, the trajectory fusion stage begins. Under the same event type label, common decision structures are extracted from the triage decision paths of different historical events to construct a unified triage strategy time-series tree. Specifically, the strategy numbers of each event are sorted by recording time, forming a time-series chain within each event. Each event's time-series chain consists of multiple decision nodes with a clear order, and each node corresponds to a specific triage strategy number. Next, the time-series chains of all events with the same label are aggregated, and a structure matching algorithm is used to merge similar strategy nodes. The similarity judgment is based on the consistency of strategy numbers and the relative stability of time positions. For example, if a strategy number appears in the 3rd to 5th position in most events, it is considered that trajectory fusion conditions exist. During the strategy node fusion process, an event-level trajectory clustering method is introduced. This method does not use a single time difference or path length as the standard, but rather uses "decision frequency + relative order" as the fusion basis. The fused node is set as a unified node in the strategy tree, retaining the average of its actual time across different events as the typical time identifier for that node. If some paths diverge significantly between events (e.g., some events use the surgical emergency pathway, while others use the trauma assessment pathway), the branching structure is retained as branch nodes in the strategy tree to ensure the strategy tree has complete multi-path expression capabilities. The final constructed strategy time-series tree is a multi-level structure, with the root node as the event initiation marker node, and each branch representing a different triage decision path. The node level corresponds to the phased sequence of strategy implementation in the actual triage process. Each node in the tree structure includes a time weight indicator and the proportion of events covered.

[0058] In the error identification module, error factors are marked, and corresponding response error paths and error path libraries are constructed.

[0059] After constructing the time-series trees of triage strategies for various public emergencies, the system systematically identifies problem nodes with waiting delays or incorrect triage within the time-series trees corresponding to each event type label. This identification process is based on standardized decision execution records and primarily targets two types of errors: first, situations where the decision response time corresponding to a strategy node exceeds the standard response window, i.e., waiting delays; and second, situations where there is a logical contradiction between the patient's destination corresponding to a strategy node and the subsequent treatment path, i.e., incorrect triage. Specifically:

[0060] Based on the label type of each event, the corresponding time-series tree structure is traversed, and the time label and strategy number are extracted from all strategy nodes. For the identification of waiting delays, the time difference between two adjacent strategy nodes is used as the response time of the current node, and the standard response window under the preset event type is used as the benchmark value. Timeout nodes are identified by direct comparison. This benchmark value must be set according to clinical guidelines or emergency standards and configured uniformly during the system initialization phase.

[0061] After identifying waiting delay nodes, a deep structural analysis is performed on the triage strategy time sequence tree. A depth-first traversal algorithm is used to construct single chains path by path, starting from the root node. During this process, the complete strategy path from the root node to a leaf node is recorded, forming a sequence of single-chain structures. Each single-chain structure represents a possible triage behavior flow route, used to accurately reconstruct possible multiple triage judgment paths in event processing. In each single-chain structure, the uniqueness of the node number is checked one by one. If a duplicate number appears, it indicates that the same strategy node has been executed repeatedly, usually meaning multiple invalid judgments, triage rejections, or resource scheduling errors for the same patient. Such nodes are identified as erroneous triage strategy nodes. To improve the accuracy of error identification, only the position of the first repeated node is marked as the decision starting point for the error, and no additional annotations are made for subsequent repeated segments. Once an erroneous node is identified, it is labeled with an "error factor" in the strategy time sequence tree, and an error factor list is established for subsequent construction of the response path structure.

[0062] Using each error factor as an anchor node, the system traces back to the event origin from its position in the triage strategy time-series tree, following the original record time sequence. Simultaneously, it records the strategy node number and its time label for each traversed node. The constructed paths are accumulated on an event-by-event basis, with each path existing as a chain-like sub-path in the graph. Directed edges are established between nodes based on execution order logic. This graph structure not only establishes edge relationships based on time sequence but also considers whether subsequent nodes depend on the decisions of preceding nodes as a necessary condition for edge establishment. For example, if the execution of a strategy node requires confirmation of a diagnosis from a preceding node, it will be set as a subsequent node in the graph, possessing directionality. This logical relationship mainly relies on the "execution basis" field in the actual records or rule constraints in the path template library. The completed response graph structure has the following characteristics: first, it is directed and acyclic, ensuring that each path can be traversed in a single direction without loops; second, error factors serve as marker nodes, providing entry point identification functionality; and third, all event response paths form an aggregation in the graph, supporting cross-event comparison and path commonality mining. Each path in the graph structure retains the original event number, node executor role, and execution time, laying a complete data foundation for subsequent path scoring, interference insertion, and training template construction.

[0063] After constructing the response graph, to identify the potential chain of causes leading to bias, a systematic backtracking traversal is needed for each node labeled as a bias factor in the graph. This traversal operation starts with the bias factor and traces upwards to all execution paths that could have caused the bias, ultimately forming a complete set of bias occurrence paths. Specifically, each bias factor node is first locked, and a forward tracing operation is performed based on the directed edge relationships in the response graph. A breadth-first search algorithm is used, expanding upwards from the bias node until a node without predecessors or the event starting point node is reached. Each traversal retains the node number, time label, and event number for reconstructing the path structure. During the traversal, to avoid path redundancy or infinite expansion, a maximum path length is set (e.g., 8-10 strategy nodes), and duplicate traversal paths are excluded. Each traversal path should structurally possess a closed-loop strategy logic, meaning that there must be a clear triage and deduction logic relationship between each node and subsequent nodes. If a breakpoint node appears in the traversal path that contradicts the event diagnosis logic, the path is discarded.

[0064] After traversal, the backtracking path corresponding to each bias factor is stored as a "strategy failure path," which represents a complete decision chain executed in history and ultimately producing negative consequences. All strategy failure paths are categorized and archived by event type and assigned a unique path identifier, forming a bias path library. In constructing the indicators for the consequences of strategy failure paths, three dimensions are set: mortality delay indicator, resource idleness indicator, and symptom aggravation indicator, by structurally integrating the actual consequences data caused by triage errors in historical public emergencies. The mortality delay indicator is marked by statistically analyzing the time difference between the occurrence of the bias node and the patient's death, combined with the average delay benchmark of similar events; the resource idleness indicator identifies the time spent on key resources that were not effectively used based on resource allocation records within the corresponding time period of the triage path; and the symptom aggravation indicator marks the degree of disease deterioration caused by strategy bias by comparing changes in patient symptom scores before and after triage. These consequence indicators are appended to each policy failure path as attribute fields, forming a searchable and evaluable consequence expression model. This provides an objective basis for subsequent interference generation, training priority ranking, and triage blind spot identification, enhancing the practical orientation and error correction capabilities of the training system.

[0065] In the interference generation module, the bias factor is converted into interference options and inserted into the simulation training process.

[0066] The target event type for the current training cycle is defined, which is related to the category of public emergencies relevant to the current training task, such as large-scale traffic accidents, chemical spills, or earthquake-induced surges in medical visits. Each event type corresponds to a unique tag. This tag is used to perform a precise search in the existing historical event record database, filtering out all event samples that match that type. After sample filtering, based on the triage strategy execution path and decision time corresponding to each event, the existing event type triage strategy time sequence tree is invoked, and its structure is decomposed to extract each complete single-chain structure as the raw material for constructing the simulation training process.

[0067] During the construction of the training process, the actual consequences of policy failure paths corresponding to policy nodes in each single-chain structure are extracted and quantified. Specifically, the indicators such as death delay, resource idleness, and symptom aggravation carried in each path are weighted and synthesized. A preset weighting factor (e.g., death delay 0.5, resource idleness 0.3, symptom aggravation 0.2) is used to normalize the three types of indicators before summing them to obtain the comprehensive consequence value for each single-chain path. The single chain with the smallest comprehensive consequence value among all paths is selected as the baseline structure for the simulated training process to ensure that the process has basic rationality and standard comparability in the early stages of training.

[0068] After the standard process structure is determined, the pre-built triage behavior template library is invoked, using the identified and structured error paths from previous steps as input. This template library contains various standard and error operation comparison structures. For each strategy node in the error path, an operation item corresponding to its functional objective but containing potential biases is generated. The generation process of interference operation items follows two logical paths: one is the waiting delay logic, which involves delaying or failing to initiate the triage decision operation within the effective time, with corresponding operation items including delayed confirmation and repeated judgment; the other is the triage error logic, which involves selecting the wrong referral target when faced with multiple diagnostic branches, with corresponding operation items including behavioral templates such as symptom misjudgment and referral error. After generation, these interference operation items are inserted into specific nodes of the simulation training process, replacing or existing alongside the standard operation items, so that trainees face real-world induced bias scenarios when receiving training tasks, thereby improving the adversarial nature of training and the robustness of triage decisions.

[0069] The interference setting module sets the interference induction weight and time pressure parameters for each interference option. The interference induction weight is set for each interference option before it is inserted into the simulation training process.

[0070] For each interference option inserted into the simulated triage training process, the induction intensity is quantified and set one by one. Each interference option originates from a policy failure node with a high frequency of occurrence in the historical error path, and can be located at a specific policy node in the triage policy time-series tree. For this node, its hierarchical structure information in the triage policy time-series tree is obtained, and the set of all child nodes with the same parent node within this level is extracted. In this set, the number of nodes historically marked as error factors is counted, and the proportion of this number to the total number of child nodes is used as the "triage error frequency" corresponding to the interference option. For example, if there are 5 child nodes in a certain level, and 3 of them have had incorrect triage records in historical events, then the triage error frequency is set to 3 / 5 = 0.6. This frequency value serves as the base value for the interference induction weight, directly assigned to the corresponding interference option, so that its guidance intensity in the training process has a clear reference basis and maintains consistency with the historical actual error distribution.

[0071] After setting the induction weights, a decision-time constraint, or time pressure parameter, is further set for each interference option. The setting process for this parameter is independent of waiting delay nodes to avoid distortion. Specifically, all strategy nodes previously marked as waiting delays are first excluded from the triage strategy time sequence tree, retaining only the remaining normal response nodes. The set of decision-time values ​​corresponding to these nodes is then calculated. The mean of this set is then calculated, and the resulting average decision-time is used as the standard decision-reaction time window. This time value is set as the time pressure parameter for the interference item, limiting the maximum allowable reaction time for the trainee during simulation training. Decisions exceeding this time limit are judged as failed responses and recorded as training feedback indicators.

[0072] Finally, to ensure the effectiveness of the adversarial testing mechanism between the interference options and the standard options during simulated training, a unified training feedback channel is established. In the process design, each strategy node corresponds to an operation response output interface to receive the actual operation results from the trainees. All standard options and inserted interference options are jointly registered to this interface during the execution phase, without prioritization, and are synchronously activated during the training process. This means that multiple candidate response operations under the same strategy node will be presented to the trainees simultaneously, allowing them to make their own selections. The standard options are always the operation instructions in the preset correct response path, and the corresponding behavioral model remains consistent with the training evaluation criteria. The interference options, on the other hand, challenge cognitive stability through induced weights and time pressure. The synchronous activation mechanism ensures the reproduction of real interference pressure in the triage scenario and also provides a stable data structure and statistical foundation for subsequent blind spot identification and error correction training.

[0073] The blind spot identification module identifies triage error blind spots.

[0074] Based on all trainees participating in the simulated triage training process, their response performance to each interference option during training was extracted, including two dimensions: hit probability and average decision time. Hit probability refers to the proportion of a particular interference option incorrectly selected in a trainee's historical responses, while average decision time is the average response time taken when faced with that interference option. Normalization was performed on both types of data. Hit probability was uniformly mapped to between 0 and 1, with the original proportion directly used as the normalization result. Average decision time, however, required consideration of the interference item's time pressure parameter to calculate its proportion. For example, if the time pressure for a interference item was 10 seconds and a trainee's average decision time for that item was 7 seconds, the normalization result would be 0.7. Through this process, the response indicators of all trainees under all interference options were uniformly converted into a comparable normalized data structure, ensuring the accuracy and consistency of subsequent scoring operations.

[0075] Subsequently, the aforementioned normalized indicators are weighted and combined with the previously set interference-inducing weights and time pressure parameters to form a complete bias scoring system. Specifically, each trainee's bias score for a given interference option is composed of a weighted average of the normalized hit probability value and the normalized decision time value, with the weight ratio determined by the interference-inducing weight. For example, if the interference-inducing weight is 0.6, then the hit probability weight is 0.6, and the time factor weight is 0.4. These two normalized values ​​are multiplied by their corresponding weights and then added together to form the trainee's bias score for that item. The scoring results of all trainees under all interference options are stored according to the interference option number, constructing a complete bias scoring matrix. The rows of this matrix represent the trainee's number, the columns represent the interference option number, and the elements are the specific score values.

[0076] After the scoring matrix is ​​constructed, the scores of each interference item belonging to the same training subject are averaged according to the triage strategy dimension corresponding to the interference option, based on the classification criteria, to obtain the "bias tendency value" corresponding to the strategy. This operation is performed on all strategies to obtain a summary of the bias performance of each triage strategy among all training subjects. The bias tendency values ​​of all strategy dimensions are connected sequentially to construct a continuous bias data sequence in the order of strategy number, which is the "bias tendency intensity curve" of the triage strategy in the complete training process. This curve is a visual representation of the distribution trend of bias behavior of the triage strategy in the simulation process. Its vertical axis is the bias tendency value, and the horizontal axis is the triage strategy number. The high points of the curve represent strategy nodes that are susceptible to interference, and the low points represent strategy links that have been stably mastered by all training subjects. The specific identification process is as follows:

[0077] Local extrema identification is used to screen peak points in the bias curve. Here, a peak is defined as a point within a local policy node window where the corresponding bias tendency value is significantly higher than the values ​​of adjacent policy nodes, forming a clearly prominent upward trend. To avoid misjudgments caused by random fluctuations or statistical errors, a numerical bias intensity threshold is introduced to enhance the constraint of peak extraction. This threshold is recommended to be set through statistical analysis. For example, select the set of bias tendency values ​​of all training subjects in a certain type of event, calculate its mean and standard deviation, and then set the threshold to the mean plus one standard deviation, ensuring that only bias high points that are significantly abnormal in the overall distribution are identified. For example, in a sample, if the average bias tendency value is 0.45 and the standard deviation is 0.1, then the bias intensity threshold is set to 0.55. Only when the bias value of a policy node reaches a peak value higher than this value is it considered to have potential bias risk.

[0078] After identifying bias nodes that meet both peak and threshold conditions, their repeatability under similar event labels is further verified to eliminate isolated biases caused by accidental misjudgments in a single training session. To this end, the frequency and distribution of misjudgments of the policy node are statistically analyzed in historical simulated training records belonging to the same event type label. If a node is frequently misjudged by multiple trainees across multiple training cycles and consistently shows a bias peak trend, it can be confirmed as a policy node with stable inducement or comprehension ambiguity under that type of event. Such policy nodes constitute the "triage bias blind spot" that requires close attention, representing locations where the current training group generally suffers from cognitive misconceptions or decision-making difficulties.

[0079] In the error training module, the triage error blind spot is taken as the target of the current training cycle for triage training.

[0080] For each identified triage bias blind spot, a customized training task flow will be constructed based on its associated event type label and strategy node number to ensure that trainees develop a stable and correct cognitive pattern within the identified bias areas. The complete strategy path containing the node of the bias blind spot will be retrieved from a standardized simulated triage training flow library. The correct response option in the standard path will be confirmed and marked as "labeled correct option," serving as the logical benchmark for the training flow construction. Subsequently, targeted reinforcement settings will be implemented around the biased strategy node in the training flow. Specifically, the strategy branch structure will be appropriately expanded before and after the corresponding node to include multiple upstream and downstream nodes causally related to the bias blind spot. Incorrect options will be inserted as distractors into the node's response structure to enhance the trainees' ability to identify and respond to the correct path. Correct response items will be structurally identifiable but not overtly prominent, to assess the trainees' actual triage judgment ability.

[0081] After generating the training task, corresponding training validation metrics are set for each trainee. These metrics consist of two core elements: the hit probability of a bias blind zone node and the average decision time for that node. The hit probability refers to the percentage of times the trainee selects the correct response option across all training rounds of completing that node task. This metric must reach a system-defined standard value, such as a 90% hit rate, to ensure the trainee can consistently identify and correctly respond to the policy node across multiple training rounds. The other validation metric is the average decision time, defined as the average time from the presentation of the policy node to the confirmation of the response when the trainee makes a correct response. The judgment benchmark here is usually based on a weighted adjustment of the average response time of non-delayed nodes under the event type label. For example, 0.8 times the average response time of nodes of this type is used as a control threshold, requiring the trainee to have timely responses while completing the task.

[0082] The training and validation process employs a periodic evaluation mechanism, regularly reviewing the execution of training tasks. Within each validation cycle, two validation scores are calculated based on the hit probability and average decision time, respectively, and a judgment is made as to whether both scores are simultaneously met. If both indicators meet the set standards, the training and validation of that triage error blind spot is deemed to have achieved its objective, and the subject is considered to have formed a stable understanding of the corresponding strategy node. The system will automatically remove the node from the current training subject's error blind spot task pool, completing the closed-loop update of the training task, and refreshing the status of the strategy node to "validation passed." Through this process, customized generation, scientific validation, and dynamic updating of targeted training tasks are achieved, constructing a closed-loop triage skills improvement mechanism centered on triage error blind spots.

[0083] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0084] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0085] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0086] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0087] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0088] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0089] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0090] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent triage training system based on public health emergencies, characterized in that: include: The tree building module is used to retrieve the historical records of emergency public events, extract the triage strategy execution records of various events, and build a triage strategy time sequence tree. The error identification module is used to analyze the response errors of triage strategies in various events, mark error factors, and build corresponding response error paths and error path libraries. The interference generation module is used to set the training event type, extract the corresponding error path from the error path library, convert the error factor into interference options, and insert them into the simulation training process. The interference setting module is used to set the interference induction weight and time pressure parameters for interference options; The blind spot identification module is used to statistically analyze the probability of training subjects hitting the interference options and the average decision time, calculate the bias tendency intensity curve under different triage strategies, and identify triage bias blind spots. The error training module takes the triage error blind spots as the target of the current training cycle and conducts triage training. The tree construction module specifically includes the following steps for constructing a triage strategy time sequence tree: Public emergencies are archived according to event type tags, and the triage decision records and recording times of the corresponding medical treatment processes are extracted. By using the strategy number and recording time sequence of triage decision records in similar tagged events, event-level decision trajectory fusion is performed on triage nodes to construct a triage strategy time sequence tree for similar tagged events; The error identification module, which marks error factors and constructs corresponding response error paths and an error path library, specifically includes: In the triage strategy time sequence tree for each type of event, identify the strategy nodes where waiting delays or incorrect triage occur in the triage decision, and use the corresponding decision nodes as bias factors. Based on the bias factor, an event decision graph is constructed, and the strategy nodes are arranged in chronological order and upstream and downstream causal relationships to form a response graph structure with unidirectional influence logic. In the response graph, backtracking is performed starting from the node corresponding to the bias factor to extract the complete path before each bias occurs, which is taken as the strategy failure path. Integrate all policy failure paths into a bias path library; The strategy failure path is accompanied by actual consequence indicators, including death delay indicators, resource idleness indicators, and symptom aggravation indicators; The specific method for identifying strategy nodes that result in waiting delays or incorrect triage in the triage decision-making process is as follows: Traverse the triage strategy time sequence tree, obtain the time difference of each pair of adjacent tree nodes as the decision duration, preset the reaction window time for different types of events, and filter the waiting delay strategy nodes by threshold comparison method. The triage strategy time sequence tree is decomposed into a single-chain structure using a depth-first algorithm. If a duplicate node appears in the single chain, the position of the first occurrence of the duplicate node is taken as the erroneous triage strategy node. In the interference setting module, the interference induction weight and time pressure parameters for setting interference options specifically include: For each perturbation option, set the perturbation induction weight before inserting it into the simulation training process; The interference-induced weight is set by the triage error frequency corresponding to the interference option. The triage error frequency is the proportion of the bias factor among the child nodes with the same parent node within the level of the strategy node corresponding to the interference option in the triage strategy time sequence tree. Set a time pressure parameter, which is set based on the average decision time of all non-waiting delay strategy nodes to limit the effective decision time for triage; The interference option shares the same output interface as the standard option and is activated synchronously in the simulation training process. The standard option is the correct option in the standardized simulation triage training process.

2. The intelligent triage training system based on public health emergencies according to claim 1, characterized in that, The interference generation module converts the bias factor into interference options and inserts them into the simulation training process, specifically including: Define the target event type corresponding to the current training cycle, and establish a standardized simulated triage training process based on the historical record library of target event type tags; Taking the policy failure path marked in the error path as input, the preset simulation training process triage behavior template library is called to generate interference operation items with policy bias, and the generated interference operation items are inserted into the simulation training process. The interference operation items are generated according to two types of error logic: waiting delay logic and triage error logic.

3. The intelligent triage training system based on public health emergencies according to claim 2, characterized in that, The standardized simulated triage training process involves comprehensively calculating the actual consequences of each strategy failure path in all single chains after the triage strategy time sequence tree is decomposed, and selecting the single chain with the smallest calculation result value as the benchmark structure.

4. The intelligent triage training system based on public health emergencies according to claim 1, characterized in that, The blind spot identification module specifically includes identifying triage error blind spots as follows: In the simulated triage training process, the hit probability and average decision time of different trainees for all triage strategies are normalized respectively. Combined with the interference induction weight and time pressure parameter, the bias score matrix is ​​constructed according to the interference option number. The average score matrix of all trainees is calculated according to the triage strategy dimension corresponding to the interference options to obtain the bias tendency value of each triage strategy; By connecting all the bias tendency values, a bias tendency intensity curve is established to represent the bias distribution trend of the trainees throughout the training process. Based on the bias tendency intensity curve, bias blind spots in triage are identified.

5. The intelligent triage training system based on public health emergencies according to claim 4, characterized in that, The method of identifying triage bias blind spots based on bias tendency intensity curves specifically involves extracting peak values ​​from the bias tendency intensity curves during simulated training to obtain the repeatability of triage strategies misjudged by trainers under similar event labels, and marking the corresponding triage strategies as triage bias blind spots.

6. The intelligent triage training system based on public health emergencies according to claim 1, characterized in that, In the aforementioned error training module, the triage error blind spots are taken as the target of the current training cycle, and the triage training specifically includes: For each type of triage error blind spot, a dedicated training process with correctly labeled options is specified, training tasks are generated, and training validation metrics are set. Validation metrics include the hit probability and average decision time of the triage strategy corresponding to the triage error blind spot. Once both indicators meet the criteria, the triage error blind zone training and verification is deemed successful, and the blind zone task pool is automatically refreshed.

Images