Training priority determination method and system based on aviation event data statistical analysis, electronic equipment and storage medium

Abstract

The invention relates to the technical field of aviation safety and flight training data analysis, provides a training priority determination method and system based on aviation event data statistical analysis, electronic equipment and a storage medium, and solves the problems of poor accuracy of flight training subject design and poor risk pertinence. The method comprises the following steps: extracting aviation unsafe event data of a plurality of heterogeneous data sources, extracting threat, error and unexpected aircraft states, and judging a unit management state; a group decision is combined with a distributed scoring node and block chain consensus mechanism, three-level scoring is performed on various threats from the training effect, the occurrence possibility and the consequence severity, and a threat scoring matrix is formed; the unit management rate and the actual occurrence frequency of each threat are counted, a comprehensive risk value is calculated, all threats are ranked to determine training priorities, and the threats with high priorities and related errors are embedded into the scene design of a flight training subject. According to the invention, the accuracy of flight training subject design is improved.

Description

Technical Field

[0001] This application relates to the field of aviation safety and flight training data analysis technology, and in particular to a method, system, electronic device and storage medium for determining training priorities based on statistical analysis of aviation incident data. Background Technology

[0002] In the field of aviation safety management, to enhance the emergency response and risk management capabilities of flight crews, a method is needed that can scientifically and automatically identify the types of risks and weaknesses in the current training system that should be prioritized for improvement based on historical unsafe incident data. This method aims to precisely allocate limited training resources to the areas with the greatest impact on safety and the most urgent training needs, thereby systematically improving the overall safety level.

[0003] Currently, there is a risk ranking scheme based on a two-dimensional matrix of event frequency and severity of consequences. This scheme first standardizes and classifies various types of security events, then statistically analyzes the frequency of each type of event in historical data, and combines this with a qualitative assessment of the severity of the potential consequences by an expert panel. Finally, it locates the events within a pre-defined risk matrix to determine their risk levels and suggests training priorities accordingly.

[0004] This approach has limitations in its coverage of risk dimensions, relying primarily on the objective frequency and severity of events while neglecting the impact of differences in crew performance during incident handling on risk assessment. Furthermore, the expert evaluation process is susceptible to subjective biases, and the lack of transparent and traceable mechanisms to ensure consensus formation may lead to discrepancies between risk assessment results and actual training needs. Summary of the Invention

[0005] This application provides a training priority determination method, system, electronic device, and storage medium based on statistical analysis of aviation event data, in order to solve the problems of poor accuracy and poor risk targeting in the design of flight training subjects in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for determining training priorities based on statistical analysis of aviation event data, comprising: Extracting aviation safety incident data from multiple heterogeneous data sources; The flight safety incident data is used to extract the threats, errors and undesirable aircraft states corresponding to each flight safety incident, and the flight crew’s management status of the threats, errors and undesirable aircraft states is determined, which is either a managed state or an unmanaged state. Using a group-based decision-making algorithm combined with distributed scoring nodes, and under the blockchain consensus mechanism, threats of various types are scored in three levels based on three dimensions: training effect, probability of occurrence, and severity of consequences, generating a threat scoring matrix. The crew management rate and actual occurrence frequency of each type of threat are statistically analyzed. Based on the threat scoring matrix, the crew management rate, and the actual occurrence frequency of the threat, the comprehensive risk value of each type of threat is calculated. Based on the comprehensive risk value, all threats are ranked, training priorities are determined based on the ranking results, and at least one threat with the highest training priority and the error associated with the threat are embedded into the scenario design of the flight training subject.

[0007] Optionally, the step of extracting the threats, errors, and undesirable aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and determining the crew's management status of the threats, errors, and undesirable aircraft states, includes: The aviation unsafe event data is input into a dynamic knowledge graph model constructed based on the Threat and Error Management TEM classification method for entity recognition, in order to extract the threats, errors and undesirable aircraft states corresponding to each aviation unsafe event; The flight crew's management status regarding the threats, errors, and undesirable aircraft states is determined through a temporal attention mechanism; the management status is either managed or unmanaged.

[0008] Optionally, the step of determining the crew's management status regarding the threat, error, and undesirable aircraft state through a temporal attention mechanism includes: All threats, errors, and unwanted aircraft states corresponding to aviation safety incidents are respectively designated as threat entities, error entities, and unwanted aircraft state entities. Based on the entity identifiers corresponding to each entity, an event entity sequence is constructed in timestamp order. The event entity sequence is input into a pre-configured time series analysis model based on a time series attention mechanism. The time series attention calculation module in the time series analysis model calculates the correlation strength value between each threat entity in the event entity sequence and all crew operation entities that appear within a preset time period. The crew operation entities include standard procedure response actions and pilot autonomous intervention actions. The dynamic threshold filtering module in the time series analysis model dynamically adjusts the filtering threshold according to the type attribute of the threat entity to obtain the adjusted threshold, and selects candidate unit operation entities from all unit operation entities whose association strength value with each threat entity is greater than the adjusted threshold. The action validity verification module in the time series analysis model is used to evaluate the compliance and effectiveness of each candidate unit operation entity in order to determine whether the candidate unit operation entity is an effective management action against the corresponding threat, and to obtain the validity judgment result. The state labeling logic module in the time series analysis model labels the corresponding management status for each threat entity based on the validity determination result. Based on the management status of the threat entity and the causal relationship between the threat entity and the corresponding error entity and the unwanted aircraft status entity queried from the dynamic knowledge graph, the same management status is derived and labeled for the associated error entity and unwanted aircraft status entity.

[0009] Optionally, the step of calculating the association strength value between each threat entity in the event entity sequence and all crew operation entities appearing within a preset time period through the time-series attention calculation module in the time-series analysis model includes: Each entity identifier in the event entity sequence is converted into a numerical vector, wherein the vector corresponding to the threat entity is defined as the query vector, and the vector corresponding to the crew operation entity is defined as the key vector. For each threat entity, within a fixed time window following the timestamp corresponding to the threat entity, retrieve all appearing crew operation entities; The similarity score is obtained by calculating the dot product between the query vector of the threat entity and the key vector of each candidate unit operation entity. The similarity scores are input into the nonlinear transformation layer of the temporal attention calculation module for processing to obtain the standardized association weights; The standardized association weights corresponding to each candidate unit operation entity are weighted and summed with the corresponding value vectors, where the value vectors are obtained by passing the linear transformation layer of the temporal attention calculation module through the vectors corresponding to the candidate unit operation entities. The magnitude of the weighted summation result is used as the association strength value between the threat entity and the corresponding unit operation entity.

[0010] Optionally, the method utilizes a group-based decision-making algorithm combined with distributed scoring nodes, and under a blockchain consensus mechanism, to perform a three-level scoring of various types of threats based on three dimensions: training effect, probability of occurrence, and severity of consequences, generating a threat scoring matrix, including: Based on the total number of threat types, a corresponding number of distributed scoring nodes are created. Each distributed scoring node consists of a scoring group composed of members from three roles: flight operation experts, flight instructors, and risk analysts. Each distributed scoring node uses the improved Delphi method to score the specified type of threat on three dimensions: training effect, probability of occurrence, and severity of consequences, generating a node consensus scoring vector. All distributed scoring nodes generate node consensus scoring vectors and submit them to the blockchain network. Through a practical Byzantine fault-tolerant consensus mechanism, multiple rounds of exchange and verification are performed. When more than a preset proportion of nodes reach a consensus on a specific dimension score of a specific threat, the specific dimension score is recorded as a consensus score in the distributed ledger. For each threat type, a consensus score for the threat type in three dimensions is extracted from the blockchain distributed ledger and combined to form a score vector for the corresponding threat type; Arrange the score vectors of all threat types in order of threat type identifier to construct the threat score matrix.

[0011] Optionally, the improved Delphi method is used to score the specified type of threat on three dimensions: training effectiveness, probability of occurrence, and severity of consequences, generating a node consensus score vector, including: The scoring task for a specified type of threat is initiated by the scoring team coordinator on the distributed scoring node. The scoring task includes three dimensions and a multi-round iterative rule of the improved Delphi method, which includes an improved convergence condition. During the scoring task, the original anonymous scores and reasons for the original anonymous scores submitted independently by each member of the scoring team in the first round, based on the local knowledge base and historical evaluation data, for the three dimensions. Based on the original anonymous ratings and reasons for the original anonymous ratings for each dimension, the statistical distribution values ​​of the ratings for each dimension are calculated. The statistical distribution values ​​include the interquartile range. The statistical distribution values, together with the original anonymous rating reasons of all role members, are used as feedback information and distributed to all role members so that role members can refer to the feedback information to generate new anonymous ratings and new rating reasons for each dimension. Based on the new scores and reasons for the new scores in each dimension, the statistical distribution of the scores in each dimension is recalculated. The new scores are either revised anonymous scores or original anonymous scores, and the new scores are either revised anonymous scores or original anonymous scores. Repeat the distribution, acquisition, and recalculation operations until one of the following improved convergence conditions is met: the interquartile range of each dimension's score is less than or equal to a preset convergence threshold; the preset maximum number of iterations is reached; the number of role members exceeding a preset proportion indicates that they will no longer modify the score; after the improved convergence condition is met, the arithmetic mean of each dimension's score in the last round is used as the consensus score for the threat type of the corresponding distributed scoring node, and the consensus scores of all dimensions are combined in a fixed order of training effect, probability of occurrence, and severity to generate the node consensus score vector.

[0012] Optionally, the reference feedback information generates new anonymous ratings and new rating reasons for each dimension, including: The anonymous ratings of the role members in the previous round across the three dimensions are compared with the statistical distribution values ​​of the previous round in the feedback information to identify whether there are target dimensions based on the comparison results. Target dimensions include: dimensions where the anonymous rating is outside the consensus interval and dimensions where the interquartile range is greater than the preset divergence threshold. When a target dimension exists, the ratings are adaptively corrected based on the original anonymous rating reasons of all role members in the reference feedback information to obtain new anonymous ratings and new rating reasons for each dimension.

[0013] Secondly, this application provides a training priority determination system based on statistical analysis of aviation event data, comprising: The extraction module is used to extract aviation safety incident data from multiple heterogeneous data sources; The determination module is used to extract the threats, errors and undesirable aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and to determine the management status of the crew on the threats, errors and undesirable aircraft states, wherein the management status is a managed status or an unmanaged status. The scoring module is used to score various types of threats in three dimensions—training effect, probability of occurrence, and severity of consequences—using a group decision-making algorithm combined with distributed scoring nodes and a blockchain consensus mechanism, generating a threat scoring matrix. The calculation module is used to statistically analyze the crew management rate and actual occurrence frequency of various types of threats, and calculate the comprehensive risk value of each type of threat by combining the threat scoring matrix, the crew management rate and the actual occurrence frequency of the threat. The determination module is used to sort all threats according to the comprehensive risk value, determine the training priority based on the sorting result, and embed at least one threat with the highest training priority and the error associated with the threat into the scenario design of the flight training subject.

[0014] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the training priority determination method based on statistical analysis of aviation event data as described in the first aspect above.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the training priority determination method based on aviation event data statistical analysis as described in the first aspect above.

[0016] The technical solution provided in this application has the following beneficial effects: This application achieves unified aggregation of security data scattered across different systems and formats, laying a data foundation for comprehensive analysis. It transforms unstructured text event reports into structured, quantifiable security elements, accurately identifying the crew's handling performance within the event chain. By integrating the wisdom of multiple experts and employing tamper-proof mechanisms, it yields objective and consistent multi-dimensional threat assessment results, reducing subjective bias. It combines the objective occurrence of threats, the crew's subjective management capabilities, and expert scores to form a more comprehensive and multi-dimensional risk assessment index. The analysis results are directly translated into specific training improvement measures, ensuring that training resources are concentrated on the most critical security weaknesses, enhancing the relevance and effectiveness of training.

[0017] Furthermore, this application first inputs aviation unsafe event data into a dynamic knowledge graph model constructed based on a threat and error management classification method to identify and extract threats, errors, and undesirable aircraft state entities in the event; then, it analyzes the correlation between crew operations and threats in the event sequence through a temporal attention mechanism to determine the crew's management status of these safety elements.

[0018] Furthermore, this technical solution enables the automated and structured extraction of safety elements from incident reports and can accurately assess the management effectiveness of the unit in a dynamic event sequence, providing a reliable and refined analytical basis for subsequent risk quantification and training priority determination. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1A flowchart illustrating a training priority determination method based on statistical analysis of aviation event data, provided for embodiments of this application; Figure 2 A schematic diagram illustrating a specific implementation of a training priority determination method based on statistical analysis of aviation event data, provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a training priority determination system based on statistical analysis of aviation event data, provided in an embodiment of this application. Detailed Implementation

[0021] Existing risk prioritization schemes based on event frequency and severity of consequences have limitations in guiding flight training priority setting. These schemes primarily rely on objective statistics of historical events and qualitative assessments of consequences by experts, failing to incorporate the actual management performance of the crew during incident handling. Furthermore, the formation of expert consensus depends on traditional discussions, which have room for improvement in terms of subjectivity and transparency. This may lead to discrepancies between the final risk prioritization results and actual training needs.

[0022] To address the aforementioned issues, this application proposes a training priority determination method based on statistical analysis of aviation incident data. It introduces the key dimension of crew management effectiveness and employs a transparent and tamper-proof consensus mechanism for multi-dimensional evaluation, thereby constructing a more comprehensive and objective risk assessment model. This ensures that the training priority ranking results can truly reflect the safety shortcomings and training needs of on-site operations, effectively overcoming the shortcomings of existing solutions in terms of evaluation dimensions and process reliability.

[0023] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] The core of this application is to provide a training priority determination method based on statistical analysis of aviation event data. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes: Step 101: Extract aviation safety incident data from multiple heterogeneous data sources.

[0025] In step 101, multiple heterogeneous data sources refer to aviation safety data sources with different formats and structures, such as text-formatted event reports, database-formatted flight data records, and tabular-formatted maintenance records. Aviation unsafe event data refers to all relevant event record information that may affect or has already affected flight safety. For any event, embodiments of this application can provide event number, event occurrence date, aircraft type, departure and arrival airports, flight phase, event tag, and event classification.

[0026] In this embodiment, data files from different sources with different formats are first accessed and read. Then, data parsing technology is used to convert the format and align the fields of these raw data, and they are uniformly processed into a standardized intermediate data format. Next, the converted data is cleaned to remove errors or duplicate entries. Finally, all the cleaned valid data is merged and aggregated into a complete and unified aviation unsafe event dataset to prepare for subsequent analysis steps.

[0027] For example, the system automatically extracts descriptions of all flight operation-related events from three heterogeneous data sources over the past year: Airline A's text-based event report database, Regulatory Agency B's database-based safety reports, and Airport C's tabular ground support records. The system first converts all this data, which varies in format, into a unified JSON intermediate format. Then, it automatically filters out incomplete or obviously contradictory entries, such as a report lacking a specific timestamp. Finally, it integrates these data to generate a standardized and complete dataset of aviation safety incidents for further analysis.

[0028] Step 102: Extract the threats, errors and undesirable aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and determine the management status of the crew for the threats, errors and undesirable aircraft states, wherein the management status is a managed status or an unmanaged status.

[0029] In step 102, a threat refers to an external condition or event in the aviation operating environment that exists before the pilot's operation and may pose a potential risk to flight safety. An error refers to an operational or judgmental mistake by the pilot during the execution of a mission that may cause the system state to deviate from the expected state. An undesirable aircraft state refers to an unplanned, non-standard flight or ground state that the aircraft enters due to the influence of a threat or error. The managed state is used to identify whether the crew has successfully prevented the threat or error from escalating into more serious unsafe consequences through effective operational intervention; its state is divided into managed and unmanaged.

[0030] In this embodiment, a unified event dataset is first input into a dynamic knowledge graph model built on a threat and error management classification method. This model can automatically identify threat entities, error entities, and unwanted aircraft state entities described in the text. Subsequently, the system employs a temporal attention mechanism to analyze a series of actions performed by the crew within a specific time window after a threat occurs. By calculating the correlation strength between these actions and the threat, and combining this with a pre-set standard operating procedure library for compliance and effectiveness verification, the system intelligently determines whether the crew has effectively managed the current threat. Finally, based on the threat management status determination result and the causal relationships between entities in the knowledge graph, the system automatically derives and labels the management status of associated errors and unwanted aircraft states.

[0031] For example, using the dataset obtained in the previous step, the system utilizes a dynamic knowledge graph model to accurately identify the threat entity "low-level wind shear," the error entity "initial energy management was slightly conservative," and the undesired aircraft state entity "aircraft below the glide slope" from an event report describing "the flight encountered low-level wind shear during the approach phase." Next, the time-series analysis model finds that the crew executed the standard procedure of "immediately performing the go-around procedure" after the occurrence of "low-level wind shear." Verification shows that this procedure was fully compliant and effectively addressed the threat; therefore, the management status of this threat is determined to be "managed." Based on the causal chain in the knowledge graph—"low-level wind shear led to conservative initial energy management, which in turn caused the aircraft to fall below the glide slope"—the system automatically marks the management status of the related error and undesired aircraft state as "managed."

[0032] Step 103: Using a group-based decision-making algorithm combined with distributed scoring nodes, and under the blockchain consensus mechanism, the threat types are scored in three levels based on three dimensions: training effect, probability of occurrence, and severity of consequences, to generate a threat scoring matrix.

[0033] In step 103, the group decision-making algorithm is a decision-making method that gathers the opinions of multiple experts to reach a collective consensus. Distributed scoring nodes refer to units composed of different expert groups that independently perform scoring work within the network. The blockchain consensus mechanism is a technical protocol that ensures all participants in a distributed network can reach an agreement on a certain data value and prevents data tampering. The training effectiveness dimension is used to evaluate the potential value of improving the crew's ability to manage the threat through targeted training. The probability of occurrence dimension is used to assess the probability of this type of threat occurring in actual operation. The severity of consequences dimension is used to assess the severity of the security consequences that may result if the threat is not effectively managed. Three-level scoring refers to setting multiple levels (e.g., high, medium, low) for each dimension for quantitative evaluation. The threat scoring matrix is ​​a mathematical table where rows represent different types of threats, columns represent the three scoring dimensions of training effectiveness, probability of occurrence, and severity of consequences, and each element in the matrix is ​​the consensus score value for the corresponding threat on the corresponding dimension.

[0034] In this embodiment, firstly, based on the total number of identified threat types, a corresponding number of distributed scoring nodes are formed. Each node consists of a scoring group comprising flight operation experts, flight instructors, and risk analysts. For each specific threat type assigned to it, each node employs a modified Delphi method. Group members anonymously and independently submit scores and justifications across three dimensions: training effectiveness, probability of occurrence, and severity, in multiple rounds. The scores are continuously revised based on statistical feedback from previous rounds until a consensus is reached within the group, forming the node's consensus score vector. Subsequently, the consensus score vectors of all nodes are submitted to a blockchain network. Through its built-in practical Byzantine fault-tolerant consensus mechanism, multiple rounds of exchange and verification are performed. When a sufficient proportion of nodes reach agreement on the specific dimension scores of a specific threat, the consensus score is permanently recorded on the distributed ledger. Finally, the system extracts the final consensus scores for all threat types across the three dimensions from the blockchain ledger and arranges them in order of threat type identifier to construct a threat score matrix.

[0035] For example, suppose the system has identified multiple threats such as "low-level wind shear" and "communication disruption." To address this, multiple distributed scoring nodes are established. For the scoring node targeting the "low-level wind shear" threat, the expert group conducts multiple rounds of anonymous scoring and explanation, ultimately reaching an internal consensus on the threat's scores in three dimensions: "training effectiveness" (high value), "probability of occurrence" (medium), and "severity" (high), forming a vector. The consensus vectors for all threat types are uploaded to the blockchain network. After multiple rounds of verification among nodes, a global consensus is reached on the scores for all threats across all dimensions. The system then extracts these consensus scores from the blockchain, forming a complete threat scoring matrix. The row containing "low-level wind shear" records the consensus results for each of its three dimensions.

[0036] Step 104: Calculate the crew management rate and actual occurrence frequency of each type of threat, and combine the threat scoring matrix, the crew management rate and the actual occurrence frequency of the threat to calculate the comprehensive risk value of each type of threat.

[0037] In step 104, the crew management rate is the proportion of events for a specific type of threat that the crew has managed as "managed" out of all relevant unsafe events. Actual frequency refers to the number of times a specific type of threat occurs among all unsafe events within the statistical period. The comprehensive risk value is a numerical value obtained through a specific formula. This value comprehensively reflects the expert consensus score of the threat, its manageability (indirectly reflected through the management rate), and its frequency of occurrence, and is used for the final risk ranking.

[0038] In this embodiment, firstly, based on the event data marked with management status in step 102, the crew management rate for each threat type is statistically calculated. Simultaneously, the actual occurrence frequency of each threat type in historical data is statistically analyzed. Next, from the threat scoring matrix generated in step 103, the scores for each threat type in three dimensions—training effectiveness, probability of occurrence, and severity of consequences—are read. Finally, these data are substituted into a preset comprehensive risk value calculation formula. This formula typically combines the management rate (reflecting management effectiveness), occurrence frequency (reflecting exposure probability), and the scores of the three dimensions (reflecting inherent threat attributes and training value) through weighted summation or multiplication, thereby calculating a comprehensive risk value representing the overall risk level of each threat.

[0039] For example, following the steps above, the system statistically analyzes and finds that among all recorded events involving the threat of "low-level wind shear," the proportion of events where the crew successfully managed the threat (i.e., the management status was "managed") (i.e., the crew management rate) is a certain value, and the actual frequency of the threat during the statistical period is also a certain value. Simultaneously, consensus scores for "low-level wind shear" across three dimensions—training effectiveness, probability of occurrence, and severity—are obtained from the threat scoring matrix. Assume the formula for calculating the comprehensive risk value is... ,in Represents the overall risk value. Severity rating Represents the probability score of occurrence. Represents the unit management rate, Represents the actual frequency of occurrence. This represents the training effectiveness score. Substituting the obtained values ​​into this formula, the overall risk value R of the "low-level wind shear" threat can be calculated.

[0040] Step 105: Based on the comprehensive risk value, sort all threats, determine the training priority based on the sorting results, and embed at least one threat with the highest training priority and the error associated with the threat into the scenario design of the flight training subject.

[0041] In step 105, sorting refers to arranging all threat types in descending (or ascending) order of their overall risk value. Training priority is determined based on the sorting results, identifying the threat sequences that require priority attention and reinforcement in subsequent flight training. Threat-related errors refer to pilot errors identified by the knowledge graph model in step 102 that have a causal relationship with a specific threat. Scenario design for flight training subjects refers to constructing training scripts, scenarios, or simulator missions to simulate real flight environments.

[0042] In this embodiment, the comprehensive risk values ​​of all threat types calculated in step 104 are first sorted from highest to lowest. Then, based on this sorted list, the order of training resource allocation is determined; generally, threats with higher comprehensive risk values ​​have higher training priority. Next, the system selects the top-ranked threats and, based on the dynamic knowledge graph constructed in step 102, automatically identifies the pilot errors most frequently associated with these high-priority threats. Finally, these high-priority threats and their typical associated errors are designed as core elements into specific subject scenarios in the flight training simulator. For example, the threat conditions are deliberately set in simulated flight missions, and the trainees are assessed on their ability to avoid or correctly handle the associated errors.

[0043] For example, after calculating the comprehensive risk value of all threats (including "low-level wind shear," "communication interruption," and "engine failure"), the system sorts them in descending order. Assuming that "low-level wind shear" has the highest comprehensive risk value and ranks first, it is determined as the highest training priority. The system then uses a knowledge graph to find that the most frequently associated error with "low-level wind shear" is "initial energy management is slightly conservative." Therefore, in the next cycle of pilot refresher training simulator courses, the training management department specifically designed a training scenario of "encountering strong wind shear during the approach phase." The core assessment point of this scenario is whether pilots can effectively avoid or promptly correct the error of "initial energy management being slightly conservative" when facing the threat of "low-level wind shear," thereby safely completing the go-around or approach. When embedding high-priority threats and associated errors into scenario design, these key competency elements must be considered simultaneously to ensure that the designed training scenarios not only reproduce specific threats and errors but also specifically train and evaluate pilots' performance in these core capabilities. Key competencies include knowledge application, procedural application, automated flight path management, manual flight path management, communication, leadership and teamwork, decision-making, situational awareness, and workload management. Simulator training element design includes threat-related design, error-related design, and design related to undesirable aircraft states. When stating that "at least one high-priority threat and its associated errors are embedded into the scenario design of flight training subjects," it is further clarified that this "scenario design" specifically includes three core elements: threat-related design, error-related design, and design related to undesirable aircraft states, thereby making the construction of training scenarios more structured and targeted.

[0044] This method systematically collects and analyzes multi-source aviation safety data, automatically identifying key safety threats and crew shortcomings in operation. It integrates expert wisdom with objective operational data to conduct multi-dimensional and reliable risk assessments. Ultimately, this method accurately identifies training needs and directly transforms the analysis results into highly targeted training content, thereby improving the efficiency of flight training resource allocation and pilots' risk management capabilities, effectively supporting the continuous improvement of aviation safety levels.

[0045] To address the insufficient accuracy of existing methods in extracting safety elements from aviation unsafe events and determining crew management status, some embodiments include step 102: extracting the threats, errors, and undesirable aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and determining the crew's management status regarding the threats, errors, and undesirable aircraft states, such as... Figure 2 As shown, it includes: Step 201: Input the aviation unsafe event data into a dynamic knowledge graph model constructed based on the Threat and Error Management TEM classification method for entity recognition, so as to extract the threats, errors and undesired aircraft states corresponding to each aviation unsafe event.

[0046] In step 201, the Threat and Error Management (TEM) taxonomy is a framework theory for systematically classifying various risks in aviation operations. A dynamic knowledge graph model is a semantic network capable of representing entities and their relationships. In this application, its nodes represent specific entities identified from event data, such as threat entities, error entities, and undesirable aircraft state entities, while edges represent causal or temporal relationships between these entities. This model can dynamically update the structure of its nodes and edges based on new input data. Entity recognition refers to the process of automatically identifying and extracting named entities that conform to predefined types from unstructured text data.

[0047] In this embodiment, a knowledge framework is first pre-constructed based on threat and error management theory. This framework defines entity types such as threats, errors, and undesirable aircraft states, as well as their interrelationships. Then, pre-processed text data of aviation safety incidents is input into the model. The model automatically scans the text using its internal entity recognition component, identifying and extracting threat entities, error entities, and undesirable aircraft state entities that conform to the predefined types. During the recognition process, the model dynamically learns new relationships between entities and updates the graph structure. The final output is a structured set of entities corresponding to each safety incident, clearly defining what the threat, the error, and the undesirable state caused by the incident are.

[0048] Step 202: Determine the management status of the crew regarding the threats, errors, and undesirable aircraft states through a temporal attention mechanism, wherein the management status is either managed or unmanaged.

[0049] In step 202, the temporal attention mechanism is a computational model capable of focusing on the importance of information at different locations within a time series. In this application, it is used to analyze which of the subsequent crew operation entities are truly relevant to managing the threat after the emergence of a threat entity. There is a close synergy between the temporal attention mechanism and the "dynamic knowledge graph model": the dynamic knowledge graph model is responsible for structurally extracting entities such as threats, errors, and undesirable aircraft states and their causal relationships from aviation unsafe event data, constructing a knowledge representation of the event; while the temporal attention mechanism, based on this, analyzes the temporal correlation strength between the threat and subsequent crew operations according to the temporal order of entity emergence, thereby determining whether the crew effectively managed the threat. The combination of the two realizes a complete analytical chain from static event structure to dynamic behavior management. A managed state indicates that the crew successfully mitigated or eliminated the risk through effective operations, while an unmanaged state indicates that effective intervention was not achieved.

[0050] In this embodiment, the entities extracted in step 201 are first arranged in chronological order of events to form an event entity sequence. The core of the temporal attention mechanism is to analyze the degree of association between all crew operation entities and the threat entity within a specific time window after the threat entity appears. The mechanism calculates the association strength score between each threat entity and each crew operation entity within the window. Then, the system sets a dynamically changing screening threshold based on the type of threat to filter out candidate crew operations whose association strength exceeds this threshold. Next, these candidate operations are checked in two ways: first, whether they comply with the prescribed operating procedures; and second, whether they are effective in mitigating or eliminating the current threat. Only operations that are both compliant and effective are judged as effective management actions against the threat. If a threat finds a corresponding effective management action, its management status is marked as managed; otherwise, it is marked as unmanaged. Finally, the system automatically derives the same management status for errors and undesirable aircraft states associated with the identified threats based on the causal relationships between threats, errors, and undesirable aircraft states in the knowledge graph.

[0051] Here is a specific example: After the system prioritized "low-altitude wind shear" as the highest training priority based on comprehensive risk values, the training management department designed flight training scenarios targeting this threat and its associated error, "initial energy management being slightly conservative," based on the system's analysis results. In the next cycle of pilot refresher training for Airlines A, a high-fidelity simulation scenario called "encountering strong wind shear during approach" was added to the simulator training curriculum. The core design elements of this scenario are entirely derived from the analysis conclusions of the proposed method: the simulator dynamically simulates the threat entity of low-altitude wind shear during aircraft approach and focuses on monitoring the pilot's energy management strategy in the early stages of the threat, examining whether they can effectively avoid or promptly correct the specific error entity of "initial energy management being slightly conservative." The training evaluation criteria are directly linked to the judgment logic of the management status. If the pilot can execute standard operations such as the go-around procedure in a timely, compliant, and effective manner, as in successful historical cases, thereby successfully managing the threat and preventing the aircraft from entering an undesirable aircraft state, they are deemed to have passed the scenario assessment. This design ensures that the training content closely matches the real safety risks and crew capability shortcomings identified through data analysis, thereby enhancing the relevance and effectiveness of the training.

[0052] In this embodiment of the application, by combining dynamic knowledge graphs and temporal attention mechanisms, the key safety elements in aviation safety incidents are accurately and automatically extracted, and the management efficiency of the crew is objectively judged in accordance with the temporal logic of the incident. This provides a high-quality and structured analytical foundation for subsequent accurate risk assessment and training priority ranking.

[0053] To further improve the automation and accuracy of determining the crew management status, in some embodiments, step 202: determining the crew's management status regarding the threats, errors, and undesirable aircraft states through a temporal attention mechanism includes: Step 301: Treat the threats, errors, and unwanted aircraft states corresponding to all aviation safety incidents as threat entities, error entities, and unwanted aircraft state entities, respectively. Based on the entity identifiers corresponding to each entity, construct an event entity sequence in timestamp order.

[0054] In step 301, the entity identifier is a unique identification code that represents an entity object. The timestamp records the time information when an entity occurred during an event. The event entity sequence refers to the sequence formed by arranging all entities identified in an insecure event in chronological order of their occurrence.

[0055] In this embodiment, the system first retrieves all identified entities from the dynamic knowledge graph model, including threat entities, error entities, and unwanted aircraft state entities. Then, the system reads the timestamp information associated with each entity. Next, based on this timestamp information, the system arranges all entities in the same insecure event in chronological order from earliest to latest, ultimately forming an organized sequence of event entities.

[0056] Step 302: Input the event entity sequence into a pre-configured time series analysis model based on a time series attention mechanism. Through the time series attention calculation module in the time series analysis model, calculate the correlation strength value between each threat entity in the event entity sequence and all crew operation entities that appear within a preset time period. The crew operation entities include standard procedure response actions and pilot autonomous intervention actions.

[0057] In step 302, the temporal attention calculation module is a component specifically designed to calculate the strength of correlation between different elements in a time series. The preset time period refers to a time range set to search for crew operations that may be related to a certain threat entity, typically starting from the time the threat entity appears. The correlation strength value is a numerical value used to quantify the closeness of the correlation between a threat entity and a crew operation entity in the time series.

[0058] In this embodiment, the system sends the event entity sequence generated in step 301 into the time series analysis model. The time series attention calculation module in the model begins to work. For each threat entity in the sequence, it searches for all appearing crew operation entities within a fixed time period after its appearance. This module calculates the degree of association between the threat entity and each found crew operation entity using a specific mathematical method, and calculates a numerical value representing the degree of association for each pair of relationships, i.e., the association strength value.

[0059] Step 303: Using the dynamic threshold screening module in the time series analysis model, the screening threshold is dynamically adjusted according to the type attribute of the threat entity to obtain the adjusted threshold, and candidate unit operation entities with an association strength value greater than the adjusted threshold are selected from all unit operation entities.

[0060] In step 303, the dynamic threshold screening module is a component that can automatically adjust the screening criteria according to different conditions. The screening threshold is a critical value used to determine whether the association strength value is large enough; only crew operations with an association strength value exceeding this critical value will be further considered. The threat entity type attribute refers to the specific category to which the threat belongs, such as whether it is a weather threat or a mechanical threat. Candidate crew operation entities refer to those crew operations that have a sufficiently strong association strength with the threat entity and are thus initially screened out for subsequent detailed examination.

[0061] In this embodiment, the dynamic threshold filtering module intelligently selects a suitable filtering threshold value based on the type of threat entity currently being processed. This threshold value may differ for different types of threats. Then, the module compares all the association strength values ​​calculated in step 302 with this dynamically determined threshold value, retaining only those crew operation entities whose association strength values ​​are greater than the threshold value. These retained operation entities are marked as candidate crew operation entities.

[0062] Step 304: Through the action validity verification module in the time series analysis model, the compliance and utility of each candidate unit operation entity are evaluated to determine whether the candidate unit operation entity is an effective management action against the corresponding threat, and to obtain the validity judgment result.

[0063] In step 304, the action validity verification module is responsible for evaluating whether a unit operation is truly effective. Compliance assessment checks whether the operation conforms to established, written operating procedures and standards. Utility assessment determines whether the operation can effectively address or mitigate its associated threat in real-world scenarios. The validity determination result is a conclusion indicating whether the candidate unit operation entity is ultimately recognized as an effective management action against its corresponding threat.

[0064] In this embodiment, the action validity verification module performs a two-step verification on each candidate aircraft operation entity selected in step 303. The first step is to verify whether the operation is within the scope of standard procedures, i.e., whether it complies with regulations. The second step is to determine whether the execution of the operation can indeed manage the threat and reduce the risk, based on the specific context of the threat. Only operations that pass both steps of verification are judged as valid management actions. The specific implementation process is as follows: First, based on the preset standard operating procedure library and expert rule library, the compliance of the candidate aircraft operation entities is matched and checked to determine whether they comply with standard procedures; second, based on the threat type and context, a utility evaluation model (such as a statistical model based on historical data or lightweight simulation) is used to determine whether the operation can effectively mitigate or eliminate the threat; if the operation is both compliant and valid, it is judged as a valid management action. Example: When a "low-altitude wind shear" threat occurs, the candidate operation is "immediately execute the go-around procedure". After verification, the operation complies with standard procedures, and historical data shows that the operation can reduce the risk of loss of control caused by wind shear, so it is judged as a valid management action.

[0065] Step 305: Using the state labeling logic module in the time series analysis model, label the corresponding management status for each threat entity based on the validity determination result, and based on the management status of the threat entity and the causal relationship between the threat entity and the corresponding error entity and the unwanted aircraft status entity queried from the dynamic knowledge graph, derive and label the same management status for the associated error entity and unwanted aircraft status entity.

[0066] In step 305, the state labeling logic module is the component responsible for ultimately assigning management state labels to entities. Causal relationships refer to the sequential causal connections recorded in the dynamic knowledge graph, where a threat entity causes an error entity to appear, and the error entity, in turn, causes an undesirable aircraft state entity to appear.

[0067] In this embodiment, the state labeling logic module operates based on the validity determination result obtained in step 304. If a threat entity possesses at least one management action deemed valid, its management status is labeled as managed; otherwise, it is labeled as unmanaged. Subsequently, the module queries the dynamic knowledge graph to find all error entities and unwanted aircraft state entities directly or indirectly caused by this threat entity. Finally, the system sets the management status of these associated error entities and unwanted aircraft state entities to the same management status as their root threat entity. After labeling the management status, the method further includes: writing the labeled management status information back to the dynamic knowledge graph model as entity attributes through the knowledge graph update interface, and storing it in association with the corresponding threat entity.

[0068] Here is a specific example: The identification of the high-priority threat low-level wind shear and its associated error, which exhibits slightly conservative initial energy management, stems from the meticulous analysis of a specific unsafe event in this step. In this event, the system first constructs a sequence of event entities based on entity identifiers and timestamps. The sequence is as follows: threat entity low-level wind shear, error entity slightly conservative initial energy management, undesirable aircraft state entity aircraft below the glide slope, and crew operation entity performing a go-around procedure. The temporal attention calculation module in the time series analysis model then processes this sequence. For the threat entity low-level wind shear, it retrieves the crew operation entity performing a go-around procedure within a preset time period after its occurrence and calculates a high correlation strength value to quantify the temporal closeness between the two. Next, the dynamic threshold filtering module dynamically selects a high filtering threshold based on the attribute that low-level wind shear belongs to the meteorological threat category. Because the calculated correlation strength value exceeds this threshold, the crew operation entity performing a go-around procedure is successfully filtered as a candidate crew operation entity. Subsequently, the action validity verification module checked the candidate crew operation entity, confirming that it fully complied with the standard operating procedures and was proven highly effective in responding to this type of wind shear threat by historical data and expert knowledge. Therefore, it was determined to be an effective management action against the low-altitude wind shear threat. Finally, based on this validity determination, the status labeling logic module labeled the management status of the threat entity, low-altitude wind shear, as "managed." Based on the clear causal relationship found in the dynamic knowledge graph—that low-altitude wind shear caused initial energy management to be slightly conservative, leading to the aircraft falling below the glide slope—it automatically derived and labeled the related error entity, "initial energy management slightly conservative," and the undesired aircraft status entity, "aircraft falling below the glide slope," with the same "managed" status. This complete and automated judgment chain provides crucial data support for accurately calculating the crew management rate of this threat and ultimately positioning it as the highest training priority.

[0069] In this embodiment of the application, the above-mentioned refined steps realize the automated and accurate determination of the unit management status. The determination process fully considers the time sequence of events, the logical relationship between operations and threats, and the actual effectiveness of actions, and automatically transmits the determination results to relevant entities based on causal relationships, thereby providing a highly reliable management status data foundation for subsequent risk assessment.

[0070] To further improve the accuracy of calculating the association strength between threat entities and crew operation entities, in some embodiments, step 302: calculating the association strength value between each threat entity in the event entity sequence and all crew operation entities appearing within a preset time period using the time series attention calculation module in the time series analysis model, includes: Step 401: Convert each entity identifier in the event entity sequence into a numerical vector, wherein the vector corresponding to the threat entity is defined as the query vector, and the vector corresponding to the crew operation entity is defined as the key vector.

[0071] In step 401, the entity identifier is a code that uniquely identifies an entity. The numeric vector is an ordered list of numbers used to mathematically represent an entity. The query vector is a numeric vector specifically representing a threat entity, used in computation to proactively initiate a match. The key vector is a numeric vector specifically representing a crew operation entity, used in computation to match the query vector.

[0072] In this embodiment, the temporal attention calculation module first converts the identifier of each entity in the event entity sequence into a fixed-length numerical vector through an embedding layer. During this process, the numerical vectors converted from all threat entities are specifically defined as query vectors. The numerical vectors converted from all crew operation entities are specifically defined as key vectors. Simultaneously, the vector corresponding to each crew operation entity is also transformed through an independent linear transformation layer to generate another numerical vector called a value vector, for use in subsequent steps.

[0073] Step 402: For each threat entity, within a fixed time window following the timestamp corresponding to the threat entity, retrieve all appearing crew operation entities.

[0074] In step 402, the fixed time window refers to a pre-defined duration calculated from the moment the threat entity appears. The search refers to finding and identifying all eligible crew operation entities within this time frame.

[0075] In this embodiment, the system iterates through each threat entity. For a threat entity currently being processed, the system defines a fixed-length time range based on its recorded timestamp. Then, within this time range, the system scans the sequence of event entities to identify all crew operation entities that appeared within this time period. These identified crew operation entities are referred to as candidate crew operation entities.

[0076] Step 403: Calculate the dot product between the query vector of the threat entity and the key vector of each candidate unit operation entity to obtain the similarity score.

[0077] In step 403, the dot product is a mathematical operation used to calculate the alignment of two vectors in a direction, and the result is a numerical value. The similarity score is the result of this dot product operation, and its numerical value represents the semantic similarity between the query vector of the threat entity and the key vector of the crew operation entity.

[0078] In this embodiment, for each candidate aircraft operation entity found in step 402, the temporal attention calculation module extracts its key vector. Simultaneously, it extracts the query vector of the current threat entity. Then, the module calculates the dot product between this query vector and the key vector of the candidate aircraft operation entity. This calculated dot product value is the initial similarity score between the current threat entity and the candidate aircraft operation entity.

[0079] Step 404: Input the similarity score into the nonlinear transformation layer of the temporal attention calculation module for processing to obtain the standardized association weight.

[0080] In step 404, the nonlinear transformation layer is a component that performs nonlinear mathematical transformations on the input data. Standardization is the process of scaling a set of numerical values ​​until their sum is one. The standardized association weight is a standardized similarity score, which represents the proportion of importance of a particular operation relative to the current threat among all candidate operations.

[0081] In this embodiment, the temporal attention calculation module collects the similarity scores between all candidate aircraft operation entities and the current threat entity obtained in step 403, and inputs them into a nonlinear function for transformation. The purpose of this transformation is to introduce complexity and stabilize the values. Then, these transformed scores are standardized to ensure that the sum of all scores is one. After processing, each candidate aircraft operation entity receives an association weight between zero and one.

[0082] Step 405: The standardized association weights and corresponding value vectors of each candidate unit operation entity are weighted and summed, where the value vectors are obtained by the linear transformation layer of the temporal attention calculation module from the vectors corresponding to the candidate unit operation entities.

[0083] In step 405, weighted summation is a mathematical operation that multiplies each value by a weight and then sums all the results.

[0084] In this embodiment, the temporal attention calculation module obtains the association weight of each candidate unit operation entity from step 404. Simultaneously, it obtains the value vector pre-generated for each candidate unit operation entity in step 401. Then, the module multiplies the association weight of each candidate unit operation entity with its corresponding value vector, and sums these weighted value vectors of all candidate unit operation entities to obtain a comprehensive context vector.

[0085] Step 406: Use the magnitude of the weighted summation result as the association strength value between the threat entity and the corresponding unit operation entity.

[0086] In step 406, the modulus is the length of a vector, which mathematically represents the size of the vector.

[0087] In this embodiment, the temporal attention calculation module calculates the magnitude of the synthesized context vector obtained in step 405. This calculated magnitude is ultimately defined as the association strength value between the current threat entity and all candidate crew operation entities appearing within a fixed time window.

[0088] Here is a specific example: The key computational details for obtaining a high association strength value from the temporal attention calculation module are as follows: First, the system converts the entity identifier of the threat entity low-altitude wind shear into a query vector, resembling a list of values. Simultaneously, it converts the entity identifier of the crew operation entity performing the go-around procedure into a key vector and a value vector. Next, within a fixed time window following the timestamp corresponding to the threat entity low-altitude wind shear, the module retrieves the crew operation entity performing the go-around procedure as a candidate. Then, the module calculates the dot product between the query vector of the threat entity low-altitude wind shear and the key vector of this candidate crew operation entity performing the go-around procedure, obtaining a similarity score indicating a high initial similarity between the two. This similarity score is then input into the nonlinear transformation layer of the temporal attention calculation module for processing, such as using the Softmax function for standardization, so that the sum of the weights of all candidate operations is one, thus obtaining a standardized association weight for the operation of performing the go-around procedure, which has a high weight value. Finally, the module performs a weighted summation operation on this high association weight and the value vector corresponding to the candidate crew operation entity performing the go-around procedure. Finally, the magnitude of the vector obtained from the weighted summation was calculated, and its value was used as the final correlation strength value between the threat entity low-altitude wind shear and the crew operation entity's execution of the go-around procedure. This high correlation strength value accurately quantifies the strong temporal response relationship between the crew operation and the threat in this event, laying a solid computational foundation for successfully screening it as a candidate crew operation entity and ultimately determining its management status.

[0089] In this embodiment of the application, the temporal correlation between threats and operations is transformed into quantifiable intensity values ​​through the above-described vectorized representation and attention calculation process. This process can capture complex contextual semantic information, thereby providing reliable and accurate data for subsequent precise screening of effective management actions.

[0090] To further improve the objectivity, credibility, and tamper-proof nature of threat scoring, in some embodiments, step 103: Utilizing a group decision-making algorithm combined with distributed scoring nodes, and under a blockchain consensus mechanism, a three-level scoring system is used to assess each type of threat from three dimensions: training effectiveness, probability of occurrence, and severity of consequences, generating a threat scoring matrix, including: Step 501: Create a corresponding number of distributed scoring nodes based on the total number of threat types. Each distributed scoring node consists of a scoring group composed of members from three roles: flight operation experts, flight instructors, and risk analysts.

[0091] In step 501, a distributed scoring node is an independent logical unit in the network responsible for scoring specific types of threats. The scoring team is a group of people with different professional backgrounds who make scoring decisions within the distributed scoring node.

[0092] In this embodiment, the system first counts the total number of threat types that need to be assessed. Then, based on this total, an equal number of distributed scoring nodes are created. Each distributed scoring node is assigned a specific type of threat as its scoring object. Within each node, a scoring team is formed, which must include members from three different professional fields: flight operations experts, flight instructors, and risk analysts, to ensure a comprehensive assessment perspective.

[0093] Step 502: Each distributed scoring node scores the specified type of threat on three dimensions: training effect, probability of occurrence, and severity of consequences, using the improved Delphi method, and generates a node consensus scoring vector.

[0094] In step 502, the improved Delphi method is a specific type of group decision-making algorithm that guides an expert group to gradually reach a consensus through multiple rounds of anonymous opinion solicitation and feedback. The node consensus score vector is an ordered list of three numbers, which represent the consensus scores reached by the scoring node on the three dimensions of training effectiveness, probability of occurrence, and severity of consequences of the threat it is responsible for.

[0095] In this embodiment, within each distributed scoring node, a coordinator initiates a scoring task for a specified threat type. Team members first independently and anonymously submit initial scores and reasons for each of the three dimensions. After collecting these scores, the system calculates the statistical distribution of each dimension and feeds it back to all members along with the anonymous reasons. Members can revise their scores based on the feedback. This process is repeated multiple times until a preset convergence condition is met, such as consensus in scoring opinions or reaching the maximum number of rounds. Finally, the average score of each dimension in the last round is taken as the consensus score for that node and combined in a fixed order to form a node consensus score vector.

[0096] Step 503: Submit the node consensus score vector generated by all distributed scoring nodes to the blockchain network, and perform multiple rounds of exchange and verification through a practical Byzantine fault-tolerant consensus mechanism. When more than a preset proportion of nodes reach a consensus on the specific dimension score of a specific threat, the specific dimension score is recorded as a consensus score in the distributed ledger.

[0097] In step 503, the blockchain network is a distributed database maintained by multiple independent nodes, where data is difficult to tamper with once recorded. The Practical Byzantine Fault Tolerance consensus mechanism is an algorithm that ensures all honest nodes reach consensus on the data even if a small number of malicious or faulty nodes exist in the network. The consensus score is an authoritative score for a specific threat on a specific dimension, globally verified and ultimately confirmed by the blockchain network; it represents the consensus score for the same dimension by members of three roles: flight operation experts, flight instructors, and risk analysts. The distributed ledger is an immutable database in the blockchain network used to record all consensus results.

[0098] In this embodiment, after achieving internal consensus, all distributed scoring nodes submit their generated node consensus scoring vectors to the blockchain network. Nodes in the network communicate and verify these scoring vectors through multiple rounds of communication using a Byzantine fault-tolerant consensus mechanism. During this process, nodes exchange and verify each other's scores for the same dimension of the same threat. When the network finds that more than a preset proportion of nodes have given the same or very similar scores for a certain dimension of a threat, this score is recognized as a globally valid consensus score and is permanently and immutably recorded on the blockchain's distributed ledger.

[0099] Step 504: For each threat type, extract the consensus score of the threat type in three dimensions from the blockchain distributed ledger, and combine them to form the corresponding threat type score vector.

[0100] In step 504, the threat type score vector is an ordered list of three numbers, which are the final consensus scores of the threat type extracted from the blockchain distributed ledger on three dimensions: training effectiveness, probability of occurrence, and severity of consequences.

[0101] In this embodiment, after the blockchain network completes the recording of consensus scores for all threats across all dimensions, the system queries and reads the consensus scores confirmed for each threat type across the three scoring dimensions from the blockchain's distributed ledger, according to the threat type identifier. Then, the system combines these three consensus scores in a fixed order: training effectiveness, probability of occurrence, and severity of consequences, to form a score vector representing the final assessment result for that threat type.

[0102] Step 505: Arrange the score vectors of all threat types in order of threat type identifier to construct the threat score matrix.

[0103] In this embodiment, the system collects the score vectors generated for all threat types in step 504. Then, the system arranges these score vectors sequentially according to the natural order of the threat type identifiers or a specific sorting rule. Finally, these ordered score vectors together form a complete threat score matrix, which serves as a key input for subsequent calculation of the overall risk value. After constructing the threat score matrix, this embodiment can store the Merklegen hash value of the threat score matrix in the blockchain block header to ensure immutability.

[0104] Here is a specific example: After identifying various threat entities such as low-altitude wind shear and communication disruption and completing management status labeling, the system initiates a group consensus scoring process to generate a threat scoring matrix. Specifically, the system first creates a corresponding number of distributed scoring nodes based on the total number of identified threat types. The distributed scoring nodes for the low-altitude wind shear threat include a scoring group composed of three types of members: flight operation experts, flight instructors, and risk analysts. This group uses an improved Delphi method to conduct multiple rounds of anonymous independent scoring and reasoning for low-altitude wind shear, and continuously revises its opinions based on the statistical feedback of each round of scoring. Finally, the group reaches an internal consensus on three dimensions: the probability of training effect occurring and the severity of the consequences, generating a node consensus scoring vector, which can be represented as: training effect dimension score, probability of occurrence dimension score, severity dimension score. Subsequently, all distributed scoring nodes, including those handling threats such as communication disruptions caused by low-level wind shear, submit their generated node consensus score vectors to the blockchain network. Nodes in the network exchange and verify these vectors through a practical Byzantine fault-tolerant consensus mechanism in multiple rounds. When more than a preset proportion of nodes agree on the scores for specific threats such as low-level wind shear across all dimensions, these scores are recorded as immutable consensus scores in the blockchain's distributed ledger. The system then extracts the consensus scores for each threat type across the three dimensions from the distributed ledger, combining them to form a corresponding score vector for low-level wind shear. Finally, the system arranges the score vectors for all threat types according to their threat type identifiers, thereby constructing a threat score matrix for subsequent comprehensive risk value calculation. The row data corresponding to low-level wind shear in this matrix represents its final consensus score across the three dimensions.

[0105] In this embodiment, a decentralized, transparent, and tamper-proof threat scoring system is constructed by combining distributed node organization, improved group decision-making methods, and blockchain consensus technology. This system effectively gathers the wisdom of experts from multiple fields, ensuring the fairness, authority, and reliability of the scoring results, and laying a solid foundation for subsequent accurate risk assessment.

[0106] To further improve the efficiency and quality of consensus reached within distributed scoring nodes, in some embodiments, step 502 involves using an improved Delphi method to score the specified type of threat across three dimensions: training effectiveness, probability of occurrence, and severity of consequences, generating a node consensus scoring vector, including: Step 601: Obtain a scoring task initiated by the scoring team coordinator on the distributed scoring node for a specified type of threat. The scoring task includes three dimensions and a multi-round iterative rule of the improved Delphi method, which includes an improved convergence condition.

[0107] In step 601, the scoring team coordinator is responsible for initiating and managing the scoring process within the distributed scoring nodes. A scoring task refers to a work instruction requiring the scoring team to evaluate a specific threat type based on three dimensions: training effectiveness, likelihood of occurrence, and severity of consequences. The multi-round iteration rules define the number of rounds the scoring process needs to be repeated and the operational specifications for each round. The improved convergence criteria are a set of more flexible standards than traditional methods for determining whether the scoring process can terminate.

[0108] In this embodiment, within the distributed scoring node, the scoring team coordinator first creates a new scoring task. This task explicitly identifies the threat type to be assessed and specifies the three dimensions upon which the assessment will be based. Simultaneously, the task details the process of the improved Delphi method to be used, including how to conduct multiple rounds of scoring and feedback, and the improved convergence condition for determining when scoring can stop. Once the system receives and confirms this task, it initiates the subsequent scoring process.

[0109] Step 602: During the scoring task, obtain the original anonymous scores and original anonymous scoring reasons submitted independently by each member of the scoring team in the first round, based on the local knowledge base and historical evaluation data, for the three dimensions.

[0110] In step 602, the local knowledge base is a database stored locally by the rating members, containing their expertise and experience. Historical evaluation data consists of rating data and results accumulated from similar past evaluation tasks. The original anonymous ratings are the initial ratings submitted anonymously by each member in the first round for the three dimensions. The original anonymous rating rationale is the explanation written by the members for their initial ratings.

[0111] In this embodiment, after the scoring task is initiated, the system sends scoring invitations to all members of the scoring team. Each member, based on their local expertise and historical evaluation data, independently assesses the designated threat across three dimensions: training effectiveness, likelihood of occurrence, and severity of consequences, forming their initial score. They also need to write reasons supporting their score. All these initial scores and reasons are submitted anonymously to the system to ensure the independence of viewpoints and avoid influence from others.

[0112] Step 603: Based on the original anonymous ratings and reasons for the original anonymous ratings for each dimension, calculate the statistical distribution value of the ratings for each dimension. The statistical distribution value includes the interquartile range. The statistical distribution value and the original anonymous rating reasons of all role members are used together as feedback information and distributed to all role members so that role members can refer to the feedback information to generate new anonymous ratings and new rating reasons for each dimension.

[0113] In step 603, the statistical distribution value is a numerical value used to describe the distribution characteristics of a set of data. The interquartile range (IQR) is one such value, reflecting the dispersion of the middle 50% of the data. The feedback information is a data package distributed to each member after system integration, containing the statistical distribution of scores for each dimension and the reasons for anonymous scoring by all members.

[0114] In this embodiment, after collecting the original anonymous ratings submitted by all members in the first round, the system calculates the statistical distribution of all ratings for each rating dimension, specifically calculating the interquartile range (IVR) to understand the central tendency and dispersion of the ratings. Then, the system integrates these statistical distribution values ​​with the original anonymous rating reasons submitted by all members (maintaining anonymity) to form a complete feedback message. This feedback message is then distributed to each member of the rating panel for their reference in the next round of rating.

[0115] Step 604: Based on the new scores and new scoring reasons for each dimension, recalculate the statistical distribution value of the scores for each dimension. The new scores are either revised anonymous scores or original anonymous scores, and the new scores are either revised anonymous scoring reasons or original anonymous scoring reasons.

[0116] In step 604, the new anonymous rating is a rating that a member may have modified and resubmitted anonymously after referring to the feedback information. The new rating reason is an explanation rewritten by the member for the modified rating, or the reason for maintaining the original rating.

[0117] In this embodiment, after receiving feedback, members of the scoring panel review their previous scores by referring to the statistical data and reasons given by others. They can decide to modify their scores and elaborate on new reasons, or they can stick to their original scores and reasons. Members anonymously submit these new scores and reasons back to the system. After collecting all the new scores for this round, the system, as in step 603, recalculates new statistical distribution values ​​for each dimension, including new interquartile ranges.

[0118] Step 605: Repeat the distribution, acquisition, and recalculation operations until one of the following improved convergence conditions is met: the interquartile range of each dimension's score is less than or equal to a preset convergence threshold. The preset maximum number of iterations is reached. Role members exceeding a preset proportion indicate they will no longer modify the scores. After meeting the improved convergence conditions, the arithmetic mean of the scores for each dimension in the last round is used as the consensus score for the threat type of the corresponding distributed scoring node. The consensus scores for all dimensions are then combined in a fixed order of training effect, probability of occurrence, and severity to generate the node consensus score vector.

[0119] In step 605, the improved convergence condition is a flexible criterion for determining whether the iteration can terminate. Dimensional scoring can be understood as anonymous scoring.

[0120] In this embodiment, the system repeatedly executes steps 603 and 604, namely, distributing feedback information, obtaining new ratings and reasons, and recalculating statistical distribution values. This iterative process continues until at least one improved convergence condition is met: for example, the interquartile range of all dimension ratings is small enough to indicate sufficient consensus; or the maximum number of iterations set by the system is reached; or a sufficiently high proportion of members explicitly state that they will no longer modify their ratings. Once any convergence condition is met, the iteration stops. The system uses the arithmetic mean of all member ratings for each dimension in the last round as the node's final consensus rating for that threat in that dimension. Finally, the consensus ratings for the three dimensions are combined into a vector according to a fixed order of training effect, probability of occurrence, and severity, thus generating the node consensus rating vector for that node.

[0121] Here is a specific example: The scoring team coordinator first initiated a scoring task for low-altitude wind shear within the node, specifying that evaluation should be based on three dimensions: training effectiveness, probability of occurrence, and severity of consequences. The coordinator also explained the multi-round iteration rules and improved convergence conditions. In the first round of scoring, members from three roles—flight operation experts, flight instructors, and risk analysts—independently and anonymously submitted original anonymous scores for each of the three dimensions, along with corresponding reasons for the anonymity, based on their respective local knowledge bases and historical assessment data. The system then calculated the statistical distribution values, including the interquartile range, for each dimension based on these original anonymous scores and integrated these statistical distribution values ​​with all anonymity reasons into feedback information, which was distributed to each member. After referring to this feedback information, some members revised their views and generated new anonymous scores and reasons for each dimension, while others maintained their original scores and reasons. The system recalculated the statistical distribution values ​​based on the new scores submitted in the next round. This process of distributing feedback, obtaining new scores and reasons, and recalculating statistical distribution values ​​was repeated several times until the improved convergence condition was met: the interquartile range of the scores in each dimension was less than or equal to the preset convergence threshold, indicating that expert opinions had been sufficiently concentrated. Subsequently, the system used the arithmetic mean of all scores in each dimension in the last round as the node's consensus score for the low-level wind shear threat in this dimension. Following a strict fixed order of training effect, probability of occurrence, and severity, the consensus scores of the three dimensions were combined to generate the node's consensus score vector. This vector forms the basis for subsequent submissions to the blockchain network for global consensus.

[0122] In this embodiment, by introducing an improved Delphi method and setting flexible convergence conditions, distributed scoring nodes can efficiently and orderly achieve high-quality group consensus based on full communication and information transparency, thereby ensuring the scientific nature and representativeness of the node consensus scoring vector.

[0123] To further improve the targeting and efficiency of rating correction, in some embodiments, step 603: the reference feedback information generates new anonymous ratings and new rating reasons for each dimension, including: Step 701: Compare the anonymous ratings of the role members in the previous round across the three dimensions with the statistical distribution values ​​of the previous round in the feedback information to identify whether there is a target dimension based on the comparison results. The target dimensions include: dimensions where the anonymous rating is outside the consensus interval and dimensions where the interquartile range is greater than the preset divergence threshold.

[0124] In step 701, the consensus interval is a numerical range calculated based on the statistical distribution of all member ratings from the previous round, typically representing the area where most members' opinions are concentrated. The preset disagreement threshold is a pre-defined critical value used to determine whether the rating disagreement is too large. The target dimension refers to the specific dimensions that rating members need to pay special attention to and consider whether to modify their ratings.

[0125] In this embodiment, the system first calculates a consensus interval for each rating dimension based on all anonymous ratings collected in the previous round. Then, the system compares each member's anonymous rating for each dimension in the previous round with the corresponding consensus interval. Simultaneously, the system checks whether the interquartile range of each dimension's rating exceeds a preset disagreement threshold. Through this comparison and checking process, the system identifies two types of dimensions requiring focused attention for each member: dimensions where the member's own rating falls outside the consensus interval, and dimensions where overall opinions remain highly divergent, i.e., the interquartile range exceeds the preset disagreement threshold. These identified dimensions are the member's target dimensions for the current round.

[0126] Step 702: When a target dimension exists, the rating is adaptively corrected based on the original anonymous rating reasons of all role members in the reference feedback information to obtain new anonymous ratings and new rating reasons for each dimension.

[0127] In step 702, adaptive rating correction refers to the process by which rating members autonomously adjust their ratings based on feedback information, especially the reasons given by other members.

[0128] In this embodiment, when the system identifies one or more target dimensions for a member, that member will focus on reading the feedback information regarding these target dimensions, especially the original anonymous rating reasons submitted by all other role members (anonymous). The member carefully analyzes the rationality and persuasiveness of these reasons, reflecting on whether their previous ratings were biased or inadequate. Based on this analysis and reflection, the member decides whether to modify their rating on the target dimensions and how to modify it. If the rating is modified, a new rating reason needs to be written to explain the reason for the modification; if the decision is not to modify, the member can also choose to reiterate or slightly adjust the original rating reason to strengthen their position. Finally, the member forms a new round of anonymous ratings and rating reasons for all dimensions (including target and non-target dimensions) and submits them.

[0129] Here is a specific example: The system first compares the anonymous ratings of the expert in the previous round on the three dimensions of the probability and severity of the training effect with the corresponding statistical distribution values ​​in the feedback information. The comparison results identify the training effect dimension as its target dimension. This is because the anonymous rating of the expert in the previous round on the training effect is outside the consensus interval determined by the statistical distribution value. At the same time, the interquartile range of the rating in this dimension is greater than the preset disagreement threshold, indicating that the experts have a large disagreement on this dimension. When this target dimension exists, the expert adaptively corrects the score based on the original anonymous rating reasons of all role members included in the feedback information. He focuses on reading the rating reasons of other anonymous experts regarding the training effectiveness dimension. These reasons generally emphasize the key role of targeted training in improving the crew's wind shear change-out decision-making and control quality. After careful consideration of these reasons, the expert believes that he may have previously underestimated the value of training. Therefore, he corrects the rating of the training effectiveness dimension towards the consensus range and generates a new anonymous rating. At the same time, the new rating reason elaborates in detail that his change of view is based on the persuasiveness of peer reasons and his new understanding of the value of training. As for the two non-target dimensions of probability and severity, he believes that his original rating and reason are already reasonable and therefore maintains them. Thus, he finally obtains his new anonymous rating and new rating reason for each dimension in this round.

[0130] In this embodiment of the application, by accurately identifying the dimensions that need attention and providing targeted reference information, the scoring members are guided to conduct efficient reflection and correction, thereby accelerating the consensus formation process and improving the scientificity and rationality of the final node consensus scoring vector.

[0131] Figure 3 A schematic diagram of a training priority determination system based on statistical analysis of aviation event data is provided for an embodiment of this application. The specific implementation method is described below: Extraction module 31 is used to extract aviation unsafe event data from multiple heterogeneous data sources.

[0132] The determination module 32 is used to extract the threats, errors and undesirable aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and to determine the management status of the crew on the threats, errors and undesirable aircraft states, wherein the management status is a managed status or an unmanaged status.

[0133] The scoring module 33 is used to generate a threat scoring matrix by using a group decision-making algorithm, combined with distributed scoring nodes, and under the blockchain consensus mechanism to score each type of threat in three dimensions: training effect, probability of occurrence, and severity of consequences.

[0134] The calculation module 34 is used to statistically analyze the crew management rate and actual occurrence frequency of each type of threat, and calculate the comprehensive risk value of each type of threat by combining the threat scoring matrix, the crew management rate and the actual occurrence frequency of the threat.

[0135] The determination module 35 is used to sort all threats according to the comprehensive risk value, determine the training priority based on the sorting result, and embed at least one threat with the highest training priority and the error associated with the threat into the scenario design of the flight training subject.

[0136] The system in this application is used to implement the aforementioned method. Therefore, the specific implementation of the training priority determination system based on aviation event data statistical analysis can be found in the embodiment section of the training priority determination method based on aviation event data statistical analysis above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0137] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the training priority determination method based on statistical analysis of aviation event data described above.

[0138] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for determining training priorities based on statistical analysis of aviation event data.

[0139] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0140] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the training priority determination method based on aviation event data statistical analysis.

[0141] Furthermore, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. 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 implementations should not be considered beyond the scope of this application.

[0142] The foregoing has provided a detailed description of the training priority determination method, system, electronic device, and storage medium based on statistical analysis of aviation event data provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A training priority determination method based on statistical analysis of aviation event data, characterized in that, The method comprises the following steps: extracting aviation unsafe event data from multiple heterogeneous data sources; extracting threats, errors and undesired aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and determining the management state of the crew to the threats, errors and undesired aircraft states, which is a managed state or an unmanaged state; using a group decision algorithm based on distributed scoring nodes, under a blockchain consensus mechanism, to score each type of threat from three dimensions of training effect, occurrence probability and severity of resulting consequences, and generate a threat scoring matrix; statistically analyzing the crew management rate and actual occurrence frequency of each type of threat, combining the threat scoring matrix, the crew management rate and the actual occurrence frequency of the threat, and calculating the comprehensive risk value of each type of threat; According to the comprehensive risk value, all threats are sorted, and the training priority is determined based on the sorting result, and at least one threat with high training priority and the error associated with the threat is embedded into the scene design of the flight training subject.

2. The training priority determination method based on statistical analysis of aviation event data according to claim 1, characterized in that, The method for extracting threats, errors and undesired aircraft states corresponding to each aviation unsafe event from the aviation unsafe event data, and determining the management state of the crew to the threats, errors and undesired aircraft states, comprises: inputting the aviation unsafe event data into a dynamic knowledge graph model constructed based on the threat and error management (TEM) classification method for entity recognition, to extract the threats, errors and undesired aircraft states corresponding to each aviation unsafe event; determining the management state of the crew to the threats, errors and undesired aircraft states through a time series attention mechanism, which is a managed state or an unmanaged state.

3. The training priority determination method based on statistical analysis of aviation event data according to claim 2, characterized in that, The method for determining the management state of the crew to the threats, errors and undesired aircraft states through a time series attention mechanism comprises: Taking the threats, errors and undesired aircraft states corresponding to all aviation unsafe events as threat entities, error entities and undesired aircraft state entities respectively, constructing an event entity sequence in chronological order based on the entity identifiers corresponding to each entity; inputting the event entity sequence into a time series analysis model based on a pre-configured time series attention mechanism, calculating the correlation strength value between each threat entity and all crew operation entities appearing in a preset time period through a time series attention calculation module in the time series analysis model, wherein the crew operation entities include standard procedure response actions and pilot autonomous intervention actions; Through a dynamic threshold screening module in the time series analysis model, dynamically adjusting the screening threshold according to the type attribute of the threat entity to obtain an adjusted threshold, and screening candidate crew operation entities from all crew operation entities whose correlation strength value with each threat entity is greater than the adjusted threshold; Through an action effectiveness verification module in the time series analysis model, performing compliance and utility evaluation on each candidate crew operation entity to determine whether the candidate crew operation entity is an effective management action for the corresponding threat, and obtaining an effectiveness determination result; The state labeling logic module in the time series analysis model labels a corresponding management state for each threat entity based on the validity determination result, and derives and labels the same management state for the associated error entity and the undesired aircraft state entity based on the management state of the threat entity and the causal relationship between the threat entity and the corresponding error entity and undesired aircraft state entity queried from the dynamic knowledge graph.

4. The training priority determination method based on statistical analysis of aviation event data according to claim 3, characterized in that, The time series attention calculation module in the time series analysis model calculates the association strength value between each threat entity in the event entity sequence and all crew operation entities appearing within a preset time period, including: Converting each entity identifier in the event entity sequence into a numerical vector, wherein the vector corresponding to the threat entity is defined as a query vector, and the vector corresponding to the crew operation entity is defined as a key vector; For each threat entity, all appearing crew operation entities within a fixed time window after the corresponding timestamp of the threat entity are retrieved; The dot product between the query vector of the threat entity and the key vector of each candidate crew operation entity is calculated to obtain a similarity score; The similarity score is input into the nonlinear transformation layer of the time series attention calculation module for processing to obtain a standardized association weight; The standardized association weight corresponding to each candidate crew operation entity is weighted and summed with the value vector corresponding to the candidate crew operation entity, wherein the value vector is obtained by linear transformation of the vector corresponding to the candidate crew operation entity through the time series attention calculation module; The length of the weighted sum result is taken as the association strength value between the threat entity and the corresponding crew operation entity.

5. The training priority determination method based on statistical analysis of aviation event data according to claim 1, characterized in that, The threat scoring matrix is generated by using a group decision algorithm based on the training effect, occurrence probability, and severity of the resulting consequences of each type of threat under the blockchain consensus mechanism, including: According to the total number of threat types, a corresponding number of distributed scoring nodes are created, each of which is composed of a scoring group consisting of flight operation experts, flight instructors, and risk analysts; Each distributed scoring node scores the specified type of threat in the three scoring dimensions of training effect, occurrence probability, and severity of resulting consequences by an improved Delphi method, generating a node consensus score vector; The node consensus score vectors generated by all distributed scoring nodes are submitted to the blockchain network, and the practical Byzantine fault tolerance consensus mechanism is used for multiple rounds of exchange and verification. When more than a preset proportion of nodes agree on the specific dimension score of a specific threat, the specific dimension score is recorded as a consensus score to the distributed ledger; For each threat type, the consensus scores of the threat type in the three dimensions are extracted from the blockchain distributed ledger to form a scoring vector for the corresponding threat type; All threat type scoring vectors are arranged in order of threat type identifiers to construct a threat scoring matrix.

6. The training priority determination method based on statistical analysis of aviation event data according to claim 5, characterized in that, The three dimensions of the specified type of threat are scored by the improved Delphi method, a node consensus score vector is generated, including: Obtaining a scoring task for a specified type of threat initiated by a scoring group coordinator at the distributed scoring node, the scoring task containing three dimensions and the multi-round iteration rules of the improved Delphi method, the multi-round iteration rules containing an improved convergence condition; During the execution of the scoring task, the raw anonymous scores and raw anonymous score reasons for the three dimensions independently submitted by each role member of the scoring group based on the local knowledge base and historical evaluation data in the first round are obtained; Based on the raw anonymous scores and raw anonymous score reasons of each dimension, the statistical distribution value of each dimension score is calculated, including the interquartile range, and the statistical distribution value and all role members' raw anonymous score reasons are collectively used as feedback information and distributed to all role members for reference by role members to generate new anonymous scores and new score reasons for each dimension; Based on the new scores and new score reasons of each dimension, the statistical distribution value of each dimension score is recalculated, and the new score is a revised anonymous score or an original anonymous score, and the new score reason is a revised anonymous score reason or an original anonymous score reason; The distribution operation, the acquisition operation and the recalculation operation are repeated until one of the following improved convergence conditions is met: the interquartile range of each dimension score is less than or equal to the preset convergence threshold; the preset maximum iteration round is reached; more than a preset proportion of role members indicate that they no longer modify the score; After meeting the improved convergence condition, the arithmetic mean of the last round of each dimension score is taken as the consensus score of the corresponding distributed scoring node for the threat type, and the consensus scores of all dimensions are combined in a fixed order of training effect, occurrence possibility and severity to generate the node consensus score vector.

7. The training priority determination method based on statistical analysis of aviation event data according to claim 6, characterized in that, The reference feedback information generates new anonymous scores and new score reasons for each dimension, including: Comparing the anonymous scores of the role members in the last three dimensions with the statistical distribution value in the last round of feedback information to identify whether there is a target dimension based on the comparison result, the target dimension including: the dimension whose anonymous score is outside the consensus interval and the dimension whose interquartile range is greater than a preset divergence threshold; When there is a target dimension, the original anonymous score reasons of all role members in the reference feedback information are used for adaptive correction to obtain new anonymous scores and new score reasons for each dimension.

8. A training priority determination system based on statistical analysis of aviation event data, characterized by, It includes: An extraction module is used to extract aviation safety event data from a plurality of heterogeneous data sources; A determination module is used to extract threats, errors and undesirable aircraft states corresponding to each aviation safety event from the aviation safety event data, and determine the management state of the crew to the threats, errors and undesirable aircraft states, the management state being a managed state or an unmanaged state; The scoring module is configured to score each type of threat from three dimensions of training effect, occurrence probability and severity of resulting consequences by using a group decision algorithm in combination with distributed scoring nodes under a blockchain consensus mechanism, and generate a threat scoring matrix. The calculating module is configured to count the unit management rate and actual occurrence frequency of each type of threat, and calculate a comprehensive risk value of each type of threat in combination with the threat scoring matrix, the unit management rate and the actual occurrence frequency of the threat. The determining module is configured to sort all threats according to the comprehensive risk value, determine a training priority based on a sorting result, and embed at least one threat with a high training priority and an error associated with the threat into a scene design of a flight training subject.

9. An electronic device, comprising: The memory is configured to store a computer program. The processor is configured to implement the steps of the training priority determination method based on statistical analysis of aviation event data when the computer program is executed. The computer program stored in the computer readable storage medium can implement the training priority determination method based on statistical analysis of aviation event data when executed by the processor.

10. A computer-readable storage medium, characterized in that, ​

Images