Air safety data and training data integration analysis method, system, electronic device and storage medium

Abstract

The application relates to the technical field of aviation safety data analysis and pilot training optimization, and provides an aviation safety data and training data integrated analysis method and system, an electronic device and a storage medium, which solve the problems of poor pertinence of pilot training and poor foresight of safety guarantee. The method comprises the following steps: automatically extracting threats, errors and unexpected aircraft states in aviation unsafe events from multi-source heterogeneous data, and determining a management state; based on an aviation safety knowledge graph, graph calculation is performed to identify positive and negative competence and behavior indexes corresponding to the managed and unmanaged states; priority lists on the safety data side are generated by analyzing and counting frequencies; combined with the statistical analysis results of the training data in the same period, a targeted training list is obtained by weighted fusion; combined with the threat priority, a training scene configuration scheme containing threat elements and capability items is generated, and a course content, evaluation standard and resource optimization configuration scheme are formed. The application improves the pertinence of pilot training.

Description

Technical Field

[0001] This application relates to the field of aviation safety data analysis and pilot training optimization technology, and in particular to a method, system, electronic device and storage medium for the integrated analysis of aviation safety data and training data. Background Technology

[0002] In the field of aviation safety and pilot training, there is an urgent need to effectively integrate safety data generated in actual operations with simulator training data. By analyzing the crew's response performance in unsafe incidents, key capability gaps and behavioral tendencies can be identified and accurately fed back into the training process. This allows for the design of more targeted training scenarios and courses to improve pilots' practical skills and safety performance.

[0003] Currently, there is a solution that combines data analysis and a rule engine. This solution first extracts specific event elements and operational evaluations from structured event reports and training records using pre-defined keyword matching and rule logic. Then, based on fixed weights set by expert experience, the system scores and ranks data from different sources, ultimately generating a list of recommended capabilities for enhanced training to guide adjustments to subsequent training content.

[0004] This scheme relies on fixed keywords and rules for identifying event elements, which limits its depth of identification when it comes to the deeper information hidden in complex event descriptions and the relationships between elements. Furthermore, the relatively fixed weighting system used for data integration and sorting restricts the dynamic adaptability of the analysis results. Ultimately, the accuracy of the generated training suggestions in matching the core security issues exposed in real-world operating environments has room for improvement. Summary of the Invention

[0005] This application provides a method, system, electronic device, and storage medium for the integrated analysis of aviation safety data and training data, in order to solve the problems of poor targeting of pilot training and poor foresight in safety assurance in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for integrating and analyzing aviation safety data and training data, comprising:

[0007] Using a deep learning-based semantic analysis model, threats, errors, and undesirable aircraft states corresponding to various aviation safety incidents are extracted from multiple heterogeneous data sources, and the management status of the crew regarding the threats, errors, and undesirable aircraft states is determined, wherein the management status is either managed or unmanaged.

[0008] Based on the mapping relationship between management status, competence and behavioral indicators provided by the aviation safety knowledge graph, the positive competence and first behavioral indicator corresponding to the managed status, and the negative competence and second behavioral indicator corresponding to the unmanaged status are identified by graph computing.

[0009] The frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period is statistically analyzed using a time series analysis model to generate a first analysis result. The first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side.

[0010] Acquire training data from the same period, perform statistical analysis on the performance data of trainees in various training scenarios, and generate a second analysis result. The second analysis result includes a ranking list of competency scores and a ranking list of negative evaluations of behavioral indicators from the training data side.

[0011] The first analysis result and the second analysis result are fused and analyzed, and a targeted training list is calculated based on a preset weighting algorithm. The targeted training list includes competency items and behavioral indicator items that need to be trained first.

[0012] By combining the existing threat training priority list, at least one threat with the highest priority is matched and optimized with the target training list to generate a training scenario configuration scheme. The training scenario configuration scheme includes the threat elements to be embedded in the training scenario and the corresponding competency items and behavioral indicators.

[0013] Based on the training scenario configuration scheme and combined with resource constraints, training course content, evaluation criteria, and training resource optimization configuration scheme are generated. The training resource optimization configuration scheme includes instructor resource configuration, simulator usage plan, and training time arrangement.

[0014] Optionally, the step of using a deep learning-based semantic analysis model to extract the threats, errors, and undesirable aircraft states corresponding to each aviation safety incident from multiple heterogeneous data sources, and determining the crew's management status of the threats, errors, and undesirable aircraft states, wherein the management status is a managed state or an unmanaged state, includes:

[0015] Data cleaning and format standardization are performed on multi-source heterogeneous data to unify the timestamp format and flight number format. The multi-source heterogeneous data comes from: QAR data list, LOSA report, company unsafe incident report and aircraft maintenance data record.

[0016] The multi-source heterogeneous data after unification of format is input into a semantic analysis model based on deep learning. The semantic analysis model includes an aviation-enhanced BERT encoder, an entity feature extractor, and a management state decision network. The entity feature extractor consists of a BiLSTM layer and an attention mechanism.

[0017] By using an aviation-enhanced BERT encoder, domain-adaptive encoding is performed on text-based event descriptions to convert aviation terminology into fixed-dimensional vector representations containing contextual semantics. Simultaneously, numerical flight parameters are embedded to generate feature vectors that fuse multimodal information.

[0018] The entity feature extractor performs sequence modeling and key information enhancement on the feature vector to identify different types of threats, including wind shear, turbulence and thunderstorms, errors including program execution errors and standard announcement errors, and undesirable aircraft states including unnecessary weather crossings, vertical deviation and horizontal deviation.

[0019] Based on a pre-defined TEM classification rule base, the threats, errors, and undesirable aircraft states are automatically classified and coded to obtain a structured event coding sequence. Based on the structured event coding sequence and event handling result data, the management status of the crew for each aviation safety incident is determined through a management status determination network.

[0020] Secondly, this application provides an integrated analysis system for aviation safety data and training data, including:

[0021] The extraction module is used to extract the threats, errors and undesirable aircraft states corresponding to each aviation safety incident from multiple heterogeneous data sources using a semantic analysis model based on deep learning, and to 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.

[0022] The identification module is used to identify the positive competence and first behavioral indicator corresponding to the managed state, and the negative competence and second behavioral indicator corresponding to the unmanaged state, based on the mapping relationship between management status, competence and behavioral indicators provided by the aviation safety knowledge graph and graph computing.

[0023] The first analysis module is used to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period using a time series analysis model, and generate a first analysis result. The first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side.

[0024] The second analysis module is used to acquire training data from the same period, perform statistical analysis on the performance data of trainees in various training scenarios, and generate a second analysis result. The second analysis result includes a ranking list of competency scores and a ranking list of negative evaluations of behavioral indicators from the training data side.

[0025] The fusion analysis module is used to fuse the first analysis result and the second analysis result, and calculate the target training list based on the preset weighting algorithm. The target training list includes competency items and behavioral indicator items that need to be trained first.

[0026] The matching optimization module is used to combine the existing threat training priority list, match and optimize at least one threat with the target training list, and generate a training scenario configuration scheme. The training scenario configuration scheme includes the threat elements to be embedded in the training scenario and the corresponding competency items and behavioral indicators.

[0027] The generation module is used to generate training course content, evaluation criteria, and training resource optimization configuration scheme based on the training scenario configuration scheme and resource constraints. The training resource optimization configuration scheme includes instructor resource configuration, simulator usage plan, and training time arrangement.

[0028] Thirdly, this application provides an electronic device, comprising:

[0029] Memory, used to store computer programs;

[0030] A processor, used to execute the computer program to implement the steps of the method for integrating and analyzing aviation safety data and training data as described in the first aspect above.

[0031] 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 integrated analysis method for aviation safety data and training data as described in the first aspect above.

[0032] The beneficial effects of this application are as follows:

[0033] This application provides an integrated analysis method for aviation safety data and training data. The technical solution offered by this method has the following beneficial effects: It enables the automatic and accurate identification of core elements (threats, errors, and undesirable aircraft states) of unsafe events from complex heterogeneous data and objectively assesses the crew's handling effectiveness, laying the foundation for data-driven analysis. It links event elements with pilot capabilities, automatically identifying effective (positive) and need-to-improve (negative) competency and behavioral indicators, allowing the analysis to delve from the event's surface appearance to the root cause of the capability. It identifies frequently occurring and impactful capability gaps within a certain period from a dynamic perspective, providing data support based on objective trends for prioritizing training resource allocation. It systematically assesses the common weaknesses of trainees in simulated training, clarifying the areas requiring focused improvement from a training perspective. By integrating real safety performance with simulated training feedback, it calculates the most urgently needed competencies and behaviors for training, ensuring that training plans simultaneously align with the dual goals of safe operation and skills development, thus enhancing the targeted nature of interventions. By combining and matching the skills to be trained with high-priority threats, highly realistic and clearly defined training scenarios are constructed, ensuring that the training content is closely aligned with real-world security risks. The analysis results are then transformed into executable and specific training plans and resource scheduling schemes, guaranteeing a closed-loop implementation from data analysis to training deployment, thus improving the scientific nature and overall efficiency of training management.

[0034] Furthermore, this application cleans and standardizes heterogeneous data from multiple sources, including fast access recorders, flight operation safety audits, company unsafe incident reports, and aircraft maintenance records. The processed data is then input into a specialized deep learning semantic analysis model. This model uses an aviation-enhanced encoder to process textual and numerical information, generating fused features. Next, an entity feature extractor composed of a bidirectional long short-term memory network and an attention mechanism performs sequence modeling and key information enhancement to identify specific threats, errors, and undesirable aircraft states such as wind shear, procedural errors, and horizontal deviations. Finally, automatic classification and coding are performed based on a pre-defined threat and error management classification rule base. Based on this coded sequence and the event handling results, a management status determination network automatically determines the crew's management status for each event.

[0035] Furthermore, this technical solution enables automated, in-depth information extraction and understanding of multi-source heterogeneous aviation safety data. It can accurately identify various key elements contained in unsafe events and make objective and consistent judgments on the management effectiveness of the crew based on professional rules and handling results. This provides a high-quality, structured data foundation for subsequent capability analysis and training planning, improving the efficiency and accuracy of event analysis. Attached Figure Description

[0036] 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.

[0037] Figure 1 A flowchart illustrating a method for integrating and analyzing aviation safety data and training data, provided in an embodiment of this application;

[0038] Figure 2 A schematic diagram illustrating a specific implementation of an integrated analysis method for aviation safety data and training data provided in this application embodiment;

[0039] Figure 3 This is a schematic diagram of the structure of an integrated analysis system for aviation safety data and training data provided in an embodiment of this application. Detailed Implementation

[0040] Existing integrated aviation safety and training solutions face challenges in terms of analytical depth and flexibility when dealing with complex, multi-source data. These solutions primarily rely on pre-defined rules and static weights for data processing, and their ability to analyze unstructured textual descriptions in incident reports, the deep semantics of technical terms, and dynamically changing risk patterns is relatively limited. This results in the generated analytical conclusions and training recommendations lagging behind the evolving safety risks and personnel skill gaps in the actual operational environment, leaving room for improvement in the relevance of training content and its alignment with actual safety needs.

[0041] To address the aforementioned limitations, this application proposes an integrated analysis method for aviation safety data and training data. The core of this approach lies in firstly, using a deep learning model to automatically interpret safety reports and flight data from multiple sources, accurately identifying key elements and their handling outcomes in unsafe incidents; then, using knowledge graph technology to intelligently associate these elements with pilots' core competency indicators; and finally, integrating real-time training performance data to dynamically calculate the most critical training areas and automatically generate matching scenario plans and resource plans. This method transforms the traditional rule-based analysis model, enabling training design to more closely respond to the dynamic risks and capability gaps reflected in real safety data through intelligent modeling and dynamic association, thereby effectively improving the accuracy and timeliness from safety insights to training implementation.

[0042] 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.

[0043] The core of this application is to provide a method for integrating and analyzing aviation safety data and training data. A flowchart of one specific implementation method is shown below. Figure 1 As shown, the method includes:

[0044] Step 101: Using a deep learning-based semantic analysis model, extract the threats, errors, and undesirable aircraft states corresponding to each aviation safety incident from multiple heterogeneous data sources, and determine the management status of the crew regarding the threats, errors, and undesirable aircraft states, wherein the management status is either managed or unmanaged.

[0045] In step 101, a threat refers to external or internal conditions that may pose a potential safety risk during flight, such as severe weather or equipment malfunction. An error refers to a mistake made by the crew during operation or decision-making, such as a procedural omission or communication error. An undesirable aircraft condition refers to an unplanned flight attitude or trajectory that occurs during flight, such as unplanned deviation from the heading or abnormal altitude. The managed status refers to the crew's effectiveness in handling the aforementioned threats, errors, and undesirable aircraft conditions. A managed status indicates that the crew successfully identified and effectively handled the situation, while an unmanaged status indicates that the crew failed to respond effectively.

[0046] In this embodiment, data is first collected from various sources, including fast access recorders, route operation safety audits, company unsafe incident reports, and aircraft maintenance records. This data is then cleaned and formatted to ensure consistency in time information and flight identification. Next, the processed data is input into a specially designed deep learning model. This model includes an encoder optimized for the aviation field to understand technical terms and contextual meaning in textual descriptions, while also processing numerical flight parameters and converting them into unified feature representations. Then, a component combining sequence modeling and attention mechanisms identifies specific threat types (e.g., wind shear), error types (e.g., procedural errors), and undesirable aircraft states (e.g., horizontal deviation) from these features. Finally, based on a pre-defined classification rule base, the identified event elements are automatically classified and coded. Combined with the actual handling results of the events, a decision network evaluates the crew's management status for each event, determining whether it is managed or not.

[0047] For example, an airline collected recent Fast Access Recorder (FAR) data, route operation safety audit reports, and internal unsafe incident reports. The system first cleaned this data, standardizing timestamps and flight number formats. Then, a deep learning model analyzed a report describing "a flight encountering wind shear during approach, and the crew executing a go-around procedure." The model identified "wind shear" as a threat and "executing the go-around procedure" as a threat management action. Combining numerical data such as aircraft attitude parameters, the model determined that the crew's threat management status in this incident was "managed." Another report described "the crew failing to execute standard announcements, resulting in altitude deviation." The model identified "standard announcement error" as an error and "altitude deviation" as an undesirable aircraft condition, and combined this with the handling records to determine its management status as "unmanaged."

[0048] Step 102: Based on the mapping relationship between management status, competence and behavioral indicators provided by the aviation safety knowledge graph, identify the positive competence and first behavioral indicator corresponding to the managed status, and the negative competence and second behavioral indicator corresponding to the unmanaged status through graph calculation.

[0049] In step 102, the aviation safety knowledge graph is a data model that represents knowledge in the aviation safety field using a graph structure. Nodes represent entities such as threats, errors, competencies, and behavioral indicators, while edges represent relationships between entities such as causing, corresponding to, and manifesting. Competency refers to the core capabilities required for flight crews to complete flight missions, such as situational awareness and decision-making ability. Behavioral indicators are specific, observable behavioral manifestations that measure competency levels. Positive competency refers to effective capabilities demonstrated under managed conditions, while negative competency refers to capability gaps exposed under unmanaged conditions. The first behavioral indicator is the specific effective behavior corresponding to positive competency, and the second behavioral indicator is the specific behavior requiring improvement corresponding to negative competency.

[0050] In this embodiment, an aviation safety knowledge graph is first constructed, using threats, errors, and undesirable aircraft states as event nodes and various competencies and behavioral indicators as capability nodes. Two relationship types, "managed" and "unmanaged," are used to connect event nodes and capability nodes. For each unsafe event that has been determined to be managed, the graph is traversed and queried along the corresponding relationship path based on whether it is managed or unmanaged. For example, for a managed event, the system searches for all competency nodes and their corresponding behavioral indicator nodes connected to it through the "managed" relationship; these are identified as positive competencies and first behavioral indicators. Similarly, for an unmanaged event, the system searches for competency nodes and behavioral indicator nodes connected through the "unmanaged" relationship, identifying them as negative competencies and second behavioral indicators.

[0051] For example, continuing from the previous example, for the managed "wind shear go-around" event, the system traverses the "managed" relationship path in the knowledge graph to identify the competency nodes "procedure compliance" and "decision-making ability" associated with the event, as well as the corresponding first behavioral indicator nodes "strictly implement go-around procedures" and "timely decision-making". For the unmanaged "failure to follow standard announcements leading to altitude deviation" event, the system traverses the "unmanaged" relationship path to identify the negative competency "communication and coordination" and the second behavioral indicator "lack of standard announcements".

[0052] Step 103: Utilize a time series analysis model to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period, and generate a first analysis result. The first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side.

[0053] In step 103, time series analysis is a statistical method used to analyze data sequences that change over time to discover trends and patterns. Frequency of occurrence refers to the number of times a competency or behavioral indicator is identified within a specific time frame. Preset period refers to the time period set for the analysis, such as one month or one quarter. The first analysis result is a priority ranking based on safety data analysis. The aviation safety data-side competency priority list is a list that ranks the competencies requiring attention based on frequency of occurrence and trend analysis. The behavioral indicator priority list is a list that ranks specific behavioral indicators by importance.

[0054] In this embodiment, firstly, the frequency of occurrence of each identified positive competency, first behavioral indicator, negative competency, and second behavioral indicator is statistically analyzed within a preset period using a set time unit, such as monthly, forming a frequency matrix that changes over time. Then, an exponential smoothing model considering trends and seasonality is used to analyze this frequency matrix and predict the occurrence trends of each competency and behavioral indicator over a future period. Next, the system sets a dynamically changing frequency threshold and calculates a comprehensive priority score by combining the occurrence frequency of each item, the predicted trend, and their respective importance weights. Finally, based on the scores, the competency items and behavioral indicator items are sorted separately to generate the final competency priority list and behavioral indicator priority list.

[0055] For example, the system statistically analyzed the frequency of occurrence of various competency and behavioral indicators identified in the most recent quarter. Time series model analysis revealed that "procedural compliance" and "communication and coordination" occurred frequently and showed an upward trend. Combined with dynamic threshold and weight calculations, "communication and coordination," due to its high frequency and upward trend, received a high priority score and was placed at the top of the competency priority list. Consequently, the corresponding behavioral indicator "lack of standard communication" was also placed in an important position on the behavioral indicator priority list.

[0056] Step 104: Obtain training data from the same period, perform statistical analysis on the performance data of trainees in various training scenarios, and generate a second analysis result. The second analysis result includes a ranking list of competency scores and a ranking list of negative evaluations of behavioral indicators from the training data side.

[0057] In step 104, training data refers to data recorded regarding trainee performance during simulator training. Trainee refers to the pilot receiving training. Training scenario refers to a specific flight situation set up in the simulator. Performance data refers to the instructor's scores and evaluations of the trainee's various abilities in the training scenario. The competency score ranking list is a list ranking various competencies based on the trainee's average score. The behavioral indicator negative feedback ranking list is a list ranking behavioral indicators based on the number of negative feedback received by the trainee during training.

[0058] In this embodiment, firstly, simulator training data is collected within the same time period as the safety data analysis cycle. This data includes instructors' scores for each trainee's competencies in different training scenarios, as well as negative evaluation records for specific behavioral indicators. Then, the system calculates the average score for each competency across all trainees and all relevant scenarios, and sorts them from lowest to highest average score to generate a competency score ranking list. Simultaneously, the number of negative evaluations received for each behavioral indicator is counted and sorted from highest to lowest number to generate a behavioral indicator negative evaluation ranking list.

[0059] For example, the airline obtained simulator training records from the same period. The system calculated the average scores of all trainees for various competencies in scenarios such as "landing decision" and "manual flight," and found that "situational awareness" had the lowest average score. Simultaneously, the number of negative feedbacks for behavioral indicators was analyzed, revealing that "insufficient cross-checking" received the most negative feedback. Therefore, a competency score ranking list was generated with "situational awareness" ranked first, and a list of negative feedbacks for behavioral indicators was generated with "insufficient cross-checking" ranked first.

[0060] Step 105: Combine the first analysis result with the second analysis result and calculate the target training list based on the preset weighting algorithm. The target training list includes competency items and behavioral indicator items that need to be trained first.

[0061] In step 105, fusion analysis refers to the comprehensive consideration of analytical results from different sources. A weighted algorithm is a method that calculates a comprehensive score by assigning specific weights to different factors. The targeted training list, derived from the fusion analysis, clearly identifies the competency and behavioral indicators that most require priority training.

[0062] In this embodiment, firstly, the priority scores in the first analysis result are normalized to a uniform scale, resulting in a weight vector for the security data side. Simultaneously, the score ranking and the number of negative reviews in the second analysis result are standardized and converted into a weight vector for the training data side. Then, a weighted summation algorithm is used to combine these two weight vectors according to weight coefficients pre-set by domain experts to calculate the comprehensive priority score for each competency item and behavioral indicator item. Finally, the items are sorted from high to low based on this comprehensive priority score to generate the final targeted training list.

[0063] For example, the system integrates the high-priority item "communication and coordination" and its behavioral indicator "lack of standard communication" identified from the security data side with the low-scoring item "situational awareness" and its high-frequency negative behavior "insufficient cross-checking" identified from the training data side. By calculating a comprehensive priority score using a weighted algorithm, both "communication and coordination" and "situational awareness" receive high scores and are therefore placed in a priority position on the targeted training list. Their corresponding behavioral indicators "lack of standard communication" and "insufficient cross-checking" are also included.

[0064] Step 106: Combine the existing threat training priority list, match and optimize at least one threat with the target training list to generate a training scenario configuration scheme. The training scenario configuration scheme includes the threat elements to be embedded in the training scenario and the corresponding competency items and behavioral indicators.

[0065] In step 106, the threat training priority list is derived from historical data analysis and ranks the threat events that require focused training. Matching optimization refers to combining high-priority threats with high-priority training needs to design the most effective training scenarios. The training scenario configuration plan is a guideline document that specifically describes how to set up scenarios in simulation training, clarifying the types of threats to be included and the competencies and behaviors to be examined and trained.

[0066] In this embodiment, the system reads an existing threat training priority list and selects one or more top-ranked threats. Then, these threats are matched with top-ranked competency and behavioral indicator items in the targeted training list. Through an optimization algorithm, it determines which threat combinations and competency items can most effectively expose and train the target's weaknesses, thereby generating one or more training scenario configuration schemes. Each scheme details the threat elements to be simulated in the simulated training scenario, such as wind shear, and the competencies to be focused on and evaluated, such as decision-making ability, and the behavioral indicators, such as rigorous procedure execution.

[0067] For example, based on the threat training priority list, the system selects the high-priority threat "wind shear." It then matches this threat with the high-priority competencies "decision-making ability" and "procedure compliance" and their behavioral indicators from the targeted training list. A training scenario configuration scheme is generated, requiring the embedding of the "wind shear" threat element into the simulator training scenario, and focusing on examining and training the trainees' "decision-making ability" competency and behavioral indicators such as "strict adherence to go-around procedures."

[0068] Step 107: Based on the training scenario configuration scheme and combined with resource constraints, generate training course content, evaluation criteria, and training resource optimization configuration scheme. The training resource optimization configuration scheme includes instructor resource configuration, simulator usage plan, and training time arrangement.

[0069] In step 107, resource constraints refer to the actual limiting factors that need to be considered when formulating the plan, such as the number of instructors, available simulator time, and available training time for trainees. Training course content is the specific teaching and training syllabus developed based on the training scenario configuration plan. Evaluation criteria are the standards used to measure trainees' performance in training. The training resource optimization configuration plan is a plan for rationally allocating limited training resources. Instructor resource allocation refers to how instructors are assigned to different training courses. The simulator usage plan is the schedule for using the simulation training equipment. The training time arrangement is the specific training time set for trainees and instructors.

[0070] In this embodiment, firstly, based on the generated training scenario configuration scheme, the specific training course content is designed in detail, including theoretical learning modules, simulator operation practice modules, etc., and corresponding evaluation criteria are formulated to clarify how to score the students' performance. Then, considering existing resource constraints, such as the number and qualifications of available instructors, the number of simulators and available time slots, and the students' class schedules, an optimization algorithm is used to perform resource scheduling calculations. Finally, a feasible training resource optimization configuration scheme is generated, clearly specifying which instructor will use which simulator at which time to train which students in a specific course, and arranging the entire training process schedule.

[0071] For example, based on a configuration scheme that includes training on "wind shear" threats and "decision-making ability," the system generates specific training course content and evaluation criteria for "Approach and Go-around Decisions under Complex Weather Conditions." Combining the company's currently available number of senior instructors, simulator idle time slots, and trainees' available time, the system generates an optimized resource allocation scheme through calculation. This scheme specifies that Instructor A will use simulator number three to train trainee group B on a specific date and time period, and clearly defines the training duration and evaluation points.

[0072] This method intelligently analyzes multi-source safety data to automatically identify key risk points and crew capability gaps in flight operations. It then deeply integrates these with simulated training data to precisely pinpoint the most critical training areas requiring improvement. Furthermore, by intelligently matching high-priority threats with high-demand capabilities, it constructs highly realistic and clearly defined training scenarios. Ultimately, this method automatically generates executable, customized training courses and resource scheduling plans closely linked to safety requirements, forming a complete closed loop from safety insight to training implementation. This enhances the relevance and effectiveness of pilot training and the sophistication of aviation safety management.

[0073] To address the shortcomings of existing methods in extracting information on aviation safety incidents due to insufficient depth and relying on fixed rules for determining management status, some embodiments include step 101: utilizing a deep learning-based semantic analysis model to extract the threats, errors, and undesirable aircraft states corresponding to each aviation safety incident from multiple heterogeneous data sources, and determining the crew's management status of these threats, errors, and undesirable aircraft states, whereby the management status is either managed or unmanaged. Figure 2 As shown, it includes:

[0074] Step 201: Perform data cleaning and format standardization on the multi-source heterogeneous data to unify the timestamp format and flight number format. The multi-source heterogeneous data comes from: QAR data list, LOSA report, company unsafe incident report and aircraft maintenance data record.

[0075] In step 201, data cleaning refers to identifying and correcting errors, omissions, or inconsistencies in the data. Format standardization involves converting data from different sources according to a unified standard. A timestamp is a numerical sequence that identifies the time an event occurred. A flight number is a unique code used to identify a flight mission. A QuickAccess Recorder (QAR) is a structured flight parameter record exported from the aircraft's quickaccess recorder. It records a large amount of real-time data during flight, such as altitude, airspeed, attitude, and engine parameters, for subsequent flight quality monitoring and safety analysis. A Line Operations Safety Audit (LOSA) is a document generated through the Line Operations Safety Audit procedure. Specially trained observers non-intrusively collect data during normal flight operations, systematically recording crew operational behavior, threats faced, and management performance to assess the organization's safety risks. Company unsafe incident reports are documents collected by the airline's internal mandatory or voluntary reporting system. They record various safety deviations, incidents, and potential hazards reported by employees, typically including an incident description, possible causes, and appropriate actions. Aircraft maintenance data records refer to various records generated during aircraft maintenance and support activities, including fault reports, scheduled maintenance records, and component replacement records. They reflect the technical condition of the aircraft and its possible correlation with operational safety.

[0076] In this embodiment, the system first collects raw data from multiple channels, including fast access recorders, route operation safety audits, company security departments, and aircraft maintenance departments. These data may differ in time representation and flight numbering rules. The system checks the integrity of the data, handles missing fields, and converts the time information in all data into a unified format. Simultaneously, it converts various forms of flight numbers into standard codes, establishing a consistent data foundation for subsequent analysis.

[0077] Step 202: Input the formatted multi-source heterogeneous data into a deep learning-based semantic analysis model. The semantic analysis model includes an aviation-enhanced BERT encoder, an entity feature extractor, and a management state decision network. The entity feature extractor consists of a BiLSTM layer and an attention mechanism.

[0078] In step 202, the deep learning-based semantic analysis model is a complex computational model capable of understanding the meaning of text and data. The aviation-specific enhanced encoder is a component of the model specifically optimized for aviation terminology and context, enabling a better understanding of textual descriptions within the aviation domain. The entity feature extractor is a component of the model dedicated to identifying and extracting key information elements. The management status determination network is a component of the model dedicated to assessing the effectiveness of crew actions. The attention mechanism is a technique that allows the model to focus on key parts when processing information.

[0079] Step 203: Using an aviation-enhanced BERT encoder, the text-based event descriptions are domain-adaptive encoded to convert aviation terminology into fixed-dimensional vector representations containing contextual semantics. Simultaneously, numerical flight parameters are embedded to generate feature vectors that fuse multimodal information.

[0080] In step 203, domain-adaptive encoding refers to converting textual information into a numerical form that can be processed by computers while retaining its domain-specific semantics. Aviation terminology consists of vocabulary and expressions unique to the aviation field. Vector representation is a numerical representation that preserves the meaning of words in a specific context while maintaining a uniform length. Numerical flight parameters refer to digital data that records flight status. Embedding processing is the process of converting discrete numerical data into continuous, semantically rich vector representations. Feature vectors integrate information from different types of data (such as text and numerical values) into a unified numerical representation.

[0081] In this embodiment, the model's aviation-enhanced encoder begins operation. It reads event descriptions in text form, understanding the meaning of technical terms such as "wind shear" and "go-around" in their specific context, and converts this text into meaningful numerical sequences. Simultaneously, it processes numerical flight parameters, converting these parameters into information-rich numerical sequences as well. Finally, it fuses the processed text and numerical information together to form a unified feature vector containing rich information.

[0082] Step 204: Perform sequence modeling and key information enhancement on the feature vector using the entity feature extractor to identify different types of threats. These different types of threats include wind shear, turbulence, and thunderstorms. Errors include program execution errors, omissions, and standard announcement errors. Undesired aircraft states include unnecessary weather crossings, vertical deviations, and horizontal deviations.

[0083] In step 204, sequence modeling refers to the process of analyzing the order and dependencies of elements in the data. Key information enhancement refers to highlighting the information most important to the task objective during processing. Identification refers to finding and classifying specific types of entities from the data. For different types of threats, exemplary threat types include wind shear, turbulence, and thunderstorms; however, in practical applications, there can be over 100 threat types. For errors, exemplary error handling includes program execution errors and standard announcement errors; however, in practical applications, there can be dozens of error handling methods.

[0084] Step 205: Based on the preset TEM classification rule base, the threats, errors and undesirable aircraft states are automatically classified and coded to obtain a structured event coding sequence. Based on the structured event coding sequence and event handling result data, the management status of the crew for each aviation unsafe event is determined through the management status determination network.

[0085] In step 205, the TEM classification rule base is a predefined system of rules for classifying and coding threats, errors, and undesirable aircraft states. Automatic classification coding refers to automatically assigning category labels and codes to identified entities according to the rule base. A structured event coding sequence is a sequence formed by encoding all elements of an unsafe event according to a predetermined format. Event handling outcome data describes how the crew responded to the event and its final result.

[0086] In this embodiment, the system automatically categorizes identified threats, errors, and undesirable aircraft states according to preset TEM classification rules and assigns them standard codes, forming a clearly structured event coding sequence. Subsequently, the management status determination network combines this coding sequence with data on the final handling outcome of the event to comprehensively analyze the crew's performance throughout the event, and finally outputs a determination of the management status of the event, i.e., whether it is managed or not.

[0087] In the embodiments of this application, the above complete steps achieve accurate and automated extraction of key elements of aviation safety incidents through deep semantic understanding and multi-source information fusion. Based on professional rules and objective handling results, a consistent and reliable judgment is made on the crew management efficiency, providing a high-quality and structured data foundation for subsequent capability analysis and training improvement, and enhancing the depth and objectivity of incident analysis.

[0088] To further improve the accuracy and interpretability of management status determination, in some embodiments, step 205: determining the crew's management status for each aviation unsafe event through the management status determination network based on the structured event coding sequence and event handling result data, includes:

[0089] Step 301: Perform feature association processing on the structured event encoding sequence and the event handling result data to generate an event handling association feature set containing event type, severity, handling duration, and number of operation corrections.

[0090] In step 301, feature association processing is an analytical process that integrates and correlates data from different sources. The event handling association feature set is a multi-dimensional feature set formed after association processing. Event type describes the classification attributes of the unsafe event. Severity characterizes the severity level of the consequences that the event may cause. Handling time reflects the time spent by the crew from discovering the event to completing the handling. Operation correction count records the number of times the crew adjusted operations during the handling process.

[0091] Step 302: The event handling related feature set is dimensionally mapped and fused with spatiotemporal features through the feature transformation layer of the management status determination network to obtain the handling efficiency feature vector, which includes event handling timeliness features, accuracy features and compliance features.

[0092] In step 302, the feature transformation layer is the network layer in the management status determination network responsible for feature transformation and fusion. Dimension mapping is a processing method that transforms features from the original space to a new feature space. Spatiotemporal feature fusion is a comprehensive analysis technique that considers both time and space dimensions. The handling effectiveness feature vector is a comprehensive feature representation characterizing the unit's handling effect. The event handling timeliness feature reflects the time efficiency index of the unit's response speed. The accuracy feature reflects the degree to which the unit's operation conforms to standard procedure requirements. The compliance feature measures the degree to which the handling process complies with safety regulations and standards.

[0093] Step 303: The multilayer perceptron of the management status determination network performs a nonlinear transformation on the handling effectiveness feature vector, and outputs the probability distribution value of the aviation unsafe event belonging to the managed state and the unmanaged state. The probability distribution value reflects the probability of the aviation unsafe event belonging to the managed state and the unmanaged state.

[0094] In step 303, a multilayer perceptron is a deep learning model structure composed of multiple network layers. The nonlinear transformation is a complex feature transformation process implemented through activation functions. The probability distribution value is the numerical result of the model output representing the likelihood of an event belonging to each category.

[0095] In this embodiment of the application, the handling effectiveness feature vector is fed into a multilayer perceptron for processing. This structure, composed of multiple network layers, performs in-depth analysis of the input features through complex nonlinear transformations, and finally outputs two numerical results, which respectively represent the probability that the current aviation unsafe event belongs to the managed state and the probability that it belongs to the unmanaged state. These two values ​​together constitute the probability distribution value.

[0096] Step 304: Based on a preset judgment threshold, perform threshold judgment on the probability distribution value to obtain the management status corresponding to each aviation safety event.

[0097] In step 304, the decision threshold is a pre-set critical value used to classify state categories, for example, 0.7. Threshold determination is a decision-making process that classifies probability results based on the set value.

[0098] In this embodiment, the system compares and analyzes the probability distribution value output by the multilayer sensor with a pre-set judgment threshold, and determines the final management status classification of the aviation safety incident based on the comparison result. If the probability value of the incident belongs to the managed state exceeds the threshold, it is judged as the managed state; otherwise, it is judged as the unmanaged state.

[0099] In this embodiment of the application, the above complete steps achieve an objective and accurate determination of the crew management status through systematic feature association and in-depth analysis, providing a reliable decision-making basis for subsequent competency analysis and training optimization, and improving the scientificity and accuracy of aviation safety management.

[0100] To more precisely characterize the unit's handling efficiency, in some embodiments, step 302 involves: mapping the event handling-related feature set through the feature transformation layer of the management status determination network and fusing spatiotemporal features to obtain a handling efficiency feature vector. This handling efficiency feature vector includes event handling timeliness features, accuracy features, and compliance features, including:

[0101] Step 401: Perform time dimension alignment processing on the event handling associated feature set, align the event type, severity, handling duration, and number of operation corrections according to the event occurrence timestamp, and generate a time alignment feature matrix.

[0102] In step 401, the time dimension alignment process is the process of arranging and matching different features according to their chronological order of occurrence. An event timestamp is a time marker that records the specific moment an unsafe event occurs. Time alignment means keeping all feature data synchronized on the timeline. The time-aligned feature matrix is ​​a two-dimensional data table formed by arranging all aligned features in chronological order.

[0103] Step 402: Multi-scale feature extraction is performed on the time-aligned feature matrix through the temporal convolution module in the feature transformation layer to extract the processing mode features at different time scales.

[0104] In step 402, the temporal convolution module is a convolutional neural network component specifically designed for processing time-series data. Multi-scale feature extraction refers to the technique of simultaneously analyzing data patterns from different time spans. Disposal pattern features are feature representations extracted from time-series data that reflect the patterns of unit disposal behavior.

[0105] In this embodiment, the temporal convolution module receives a time-aligned feature matrix as input and simultaneously scans the time series data using multiple convolution kernels of different sizes. It captures the change patterns in the data from different time spans, such as short-term, medium-term, and long-term, thereby extracting feature representations that reflect the characteristics of the unit's handling behavior at different time scales.

[0106] Step 403: The spatial attention module in the feature transformation layer performs weighted fusion of the treatment mode features at all time scales to generate a weighted feature representation.

[0107] In step 403, the spatial attention module is a neural network component capable of automatically learning the importance of different features. Weighted fusion is a feature integration method that assigns different weights based on feature importance. The weighted feature representation is a comprehensive feature representation formed after importance weighting.

[0108] In this embodiment, the spatial attention module receives processing pattern features extracted from different time scales, calculates the importance score of each feature, assigns higher weights to important features and lower weights to minor features, and then sums all features according to their respective weights to generate a comprehensive feature representation that retains important information while removing redundancy.

[0109] Step 404: Input the weighted feature representation into the fully connected network in the feature transformation layer for dimensionality reduction, and output a handling efficiency feature vector containing event handling timeliness features, accuracy features, and compliance features. The event handling timeliness feature is used to reflect the time efficiency from the occurrence of the event to the completion of the handling, the accuracy feature is used to reflect the degree of conformity between the handling operation and the standard procedure, and the compliance feature is used to reflect the degree to which the handling process complies with aviation safety regulations.

[0110] In step 404, a fully connected network is a neural network structure composed of multiple fully connected layers of neurons. Dimensionality reduction is a data processing technique that transforms high-dimensional features into low-dimensional representations.

[0111] In this embodiment, the weighted feature representation is fed into a fully connected network. After nonlinear transformation and feature compression of multiple neuron layers, the high-dimensional weighted feature representation is converted into a low-dimensional, dense feature vector. This vector contains three core dimensions that respectively characterize the timeliness, accuracy, and compliance of the handling, thereby comprehensively reflecting the unit's handling efficiency level.

[0112] In the embodiments of this application, the above complete steps achieve a refined characterization of the unit's handling efficiency through time alignment, multi-scale feature extraction, attention weighting, and dimensionality reduction, providing a reliable feature basis for accurately determining the management status and improving the accuracy and interpretability of status determination.

[0113] To systematically uncover the capability factors behind unsafe incidents, in some embodiments, step 102: the mapping relationship between management status, competence, and behavioral indicators provided by the aviation safety knowledge graph, identifying positive competence and first behavioral indicators corresponding to the managed status, and negative competence and second behavioral indicators corresponding to the unmanaged status, through graph computation, includes:

[0114] Step 501: Construct an aviation safety knowledge graph, where threats, errors, and undesirable aircraft states are designated as corresponding event entity nodes. Competencies and their corresponding behavioral indicators are designated as capability entity nodes, interconnected through attribute relationships. Managed and unmanaged states are used as relationship types to connect event entity nodes and capability entity nodes.

[0115] In step 501, event entity nodes are graphical nodes representing specific elements of an unsafe event. Capability entity nodes are graphical nodes representing pilot capabilities and specific behavioral performance. Attribute relationships are connections that link different categories of nodes and define the logical connections between them. Relationship types are classifications that describe the specific nature of the relationships between nodes.

[0116] Step 502: Store the aviation safety knowledge graph using the Neo4j graph database, and define the semantic relationship pattern between event entity nodes and capability entity nodes, wherein the node attributes include node ID, node type, node name and weight value.

[0117] In step 502, Neo4j graph database is a database system specifically designed for storing and querying graph structure data. Semantic relation schemas are predefined rules governing the connections between nodes. Node identifiers are codes that uniquely identify each node. Node types describe the category to which a node belongs. Node names are the specific designations of the node. Weight values ​​represent the importance of a node or relation.

[0118] Step 503: For each aviation safety incident whose management status has been determined, a graph traversal query is performed based on the corresponding management status type, selecting the appropriate relational path. Specifically: For managed events, the query traverses the managed relational path to find all positive competency nodes and their corresponding first-behavioral indicator nodes associated with the managed event. For unmanaged events, the query traverses the unmanaged relational path to find all negative competency nodes and their corresponding second-behavioral indicator nodes associated with the unmanaged event. In Step 503, graph traversal query is a query method that searches along the connections between nodes in a graph database. A relational path refers to a sequence of nodes connected by a specific relation type. Positive competency refers to the effective capabilities demonstrated in managed events. Negative competency refers to the capability deficiencies exposed in unmanaged events.

[0119] In this embodiment of the application, the above complete steps achieve automatic and accurate mapping from security incidents to personnel capabilities by constructing and querying a structured knowledge graph. This provides reliable data support for in-depth analysis of capability gaps and precise design of training content, thereby enhancing the insight and pertinence of aviation safety management.

[0120] To dynamically identify and prioritize key capability factors affecting aviation safety, in some embodiments, step 103 involves using a time series analysis model to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period, generating a first analysis result. This first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side, including:

[0121] Step 601: Calculate the frequency of occurrence of various competency and behavioral indicators within a preset period on a monthly basis, and generate a time series frequency matrix.

[0122] In step 601, the time series frequency matrix is ​​a two-dimensional data table, in which rows represent consecutive months and columns represent different competency or behavioral indicator types. The value of each cell in the table represents the number of times a specific competency or behavioral indicator is identified within that month.

[0123] Step 602: Establish a time series forecasting model using the Holt-Winters three-parameter exponential smoothing method. Perform multi-scale feature extraction and predictive analysis on the frequency change trends of various competencies and behavioral indicators in the time series frequency matrix. Predict the trend changes of various competencies and behavioral indicators in the future preset time period. Various competencies and behavioral indicators include positive competencies, the first behavioral indicator corresponding to positive competencies, negative competencies, and the second behavioral indicator corresponding to negative competencies.

[0124] In step 602, the Holt-Winters three-parameter exponential smoothing method is a time series forecasting method that simultaneously considers three factors—level, trend, and seasonality—to predict future values. Multi-scale feature extraction refers to identifying patterns of change across different time spans, such as short-term fluctuations and long-term trends, from the data.

[0125] In this embodiment, the system uses a three-parameter exponential smoothing method to build a predictive model for each competency and behavioral indicator column in the time series frequency matrix. These models analyze the frequency data of each competency item in the past months, capture its changing patterns, and predict whether its frequency of occurrence will increase, decrease, or remain stable in the coming period.

[0126] Step 603: Set a dynamic frequency threshold. Based on the comparison result between the occurrence frequency and the dynamic frequency threshold, and the weights corresponding to the frequency dimension and the trend change dimension, generate priority scores for each competency and behavioral indicator.

[0127] In step 603, the dynamic frequency threshold is a baseline that automatically adjusts based on the overall data distribution, used to determine whether the frequency of a certain capability item is high. The frequency dimension refers to the current and historical frequency of occurrence of the capability item. The trend change dimension refers to the predicted future direction and magnitude of the frequency change of the capability item. The priority score is a comprehensive numerical value reflecting the degree to which the capability item needs to be given priority. Capabilities with high frequency and an upward trend will receive a higher score.

[0128] Step 604: Sort the priority scores of each competency from high to low, select the top N1 competency items, and generate a priority list of competencies for aviation safety data. Sort the priority scores of behavioral indicators from high to low, select the top N2 competency items, and generate a priority list of behavioral indicators. N1 is greater than or equal to 2, and N2 is greater than or equal to 2.

[0129] In step 604, the aviation safety data-side competency priority list is a list arranged from highest to lowest priority score, listing several competencies that most need attention. The behavioral indicator priority list is also a list arranged from priority score, listing several behavioral indicators that most need attention. These two lists together indicate the capability gaps that should be prioritized from a safety data perspective, exemplified by N1=5 and N2=5.

[0130] Step 605: Combine the aviation safety data side competency priority list and the behavioral indicator priority list into a first analysis result.

[0131] In this embodiment of the application, the system packages the generated competency priority list and behavioral indicator priority list together as the first analysis result output. This result clearly shows the competency areas that need priority intervention based on historical security data.

[0132] In the embodiments of this application, the above complete steps, through time-series dynamic analysis of the frequency of occurrence of capability items, can keenly capture those capability shortcomings that occur frequently and show a worsening trend, thereby providing accurate and forward-looking data basis for prioritizing the allocation of safety management and training resources, and improving the timeliness of safety risk warning and intervention.

[0133] To accurately identify the most important training priorities, in some embodiments, step 105 involves fusing the first analysis result with the second analysis result, and calculating a targeted training list based on a preset weighted algorithm. The targeted training list includes competency items and behavioral indicator items that require priority training, including:

[0134] Step 701: Perform Min-Max normalization on the priority score in the first analysis result to map the priority score to a preset numerical range and obtain a security data weight vector.

[0135] In step 701, Min-Max normalization is a data processing method that linearly transforms the original data to a specific numerical range. The preset numerical range is a predefined, uniform numerical range, such as [0,1]. The security data weight vector is a list of values, where each value represents the importance of a certain competency or behavioral indicator from the perspective of security data analysis, and all values ​​are on the same comparable scale.

[0136] Step 702: Perform Z-score standardization on the rating ranking and the number of negative reviews in the second analysis result, and convert them into training data weight vectors within a preset numerical range.

[0137] In step 702, Z-score normalization is a standardization method that calculates the difference between the data and its mean, using the standard deviation as the unit of measurement. Z-score normalization eliminates the influence of differences in the dimensions and distributions of these data, transforming them into a preset numerical range identical to the weight vector of the safety data, thus obtaining a standardized weight vector that reflects the relative importance of the training data.

[0138] Step 703: Using a weighted summation algorithm, combined with the weight coefficients set by domain experts, calculate the comprehensive priority score for each competency item and behavioral indicator item.

[0139] In step 703, the weighting coefficients set by the domain experts are determined by experienced professionals based on the actual situation, serving as proportional parameters to balance the importance of both security and training data. The overall priority score is a final calculated value that comprehensively reflects the overall priority of a competency or behavioral indicator across both the security and training dimensions.

[0140] Step 704: Sort the comprehensive priority scores from high to low, and generate a targeted training list based on the sorting results.

[0141] In this embodiment, the above-described complete steps effectively integrate information from both operational safety and training performance dimensions through data standardization and weighted fusion. This enables the objective and balanced identification of the most pressing common capability shortcomings, thereby providing a scientific basis for the precise allocation of training resources and improving the targeting and overall effectiveness of training planning.

[0142] Figure 3 A schematic diagram of the structure of an integrated analysis system for aviation safety data and training data provided in this application embodiment is shown in the detailed implementation section.

[0143] Extraction module 31 is used to extract the threats, errors and undesirable aircraft states corresponding to each aviation safety incident from multiple heterogeneous data sources using a semantic analysis model based on deep learning, and to 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.

[0144] The identification module 32 is used to identify the positive competence and first behavioral indicator corresponding to the managed state, and the negative competence and second behavioral indicator corresponding to the unmanaged state, based on the mapping relationship between management status, competence and behavioral indicators provided by the aviation safety knowledge graph, through graph calculation.

[0145] The first analysis module 33 is used to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period using a time series analysis model, and generate a first analysis result. The first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side.

[0146] The second analysis module 34 is used to acquire training data from the same period, perform statistical analysis on the performance data of trainees in various training scenarios, and generate a second analysis result. The second analysis result includes a ranking list of competency scores and a ranking list of negative evaluations of behavioral indicators from the training data side.

[0147] The fusion analysis module 35 is used to fuse the first analysis result and the second analysis result, and calculate the target training list based on the preset weighting algorithm. The target training list includes competency items and behavioral indicator items that need to be trained first.

[0148] The matching optimization module 36 is used to combine the existing threat training priority list, match and optimize at least one threat with the target training list, and generate a training scenario configuration scheme. The training scenario configuration scheme includes the threat elements to be embedded in the training scenario and the corresponding competency items and behavioral indicators.

[0149] The generation module 37 is used to generate training course content, evaluation criteria and training resource optimization configuration scheme based on the training scenario configuration scheme and combined with resource constraints. The training resource optimization configuration scheme includes instructor resource configuration, simulator usage plan and training time arrangement.

[0150] The aviation safety data and training data integration and analysis system of this application embodiment is used to implement the aforementioned aviation safety data and training data integration and analysis method. Therefore, the specific implementation of the aviation safety data and training data integration and analysis system can be found in the embodiment section of the aviation safety data and training data integration and analysis method above. The specific implementation can be referred to the description of the corresponding embodiment, and will not be repeated here.

[0151] 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 above-described method for integrating and analyzing aviation safety data and training data.

[0152] 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 integrating and analyzing aviation safety data and training data.

[0153] 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.

[0154] The 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 integrated analysis method for aviation safety data and training data.

[0155] Those skilled in the art will further recognize that 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 the various examples 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 implementation should not be considered beyond the scope of this application.

[0156] The foregoing has provided a detailed description of the method, system, electronic device, and storage medium for integrating and analyzing aviation safety data and training 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 method for integrating and analyzing aviation safety data and training data, characterized in that, include: Using a deep learning-based semantic analysis model, threats, errors, and undesirable aircraft states corresponding to various aviation safety incidents are extracted from multiple heterogeneous data sources, and the management status of the crew regarding the threats, errors, and undesirable aircraft states is determined, wherein the management status is either managed or unmanaged. Based on the mapping relationship between management status, competence and behavioral indicators provided by the aviation safety knowledge graph, the positive competence and first behavioral indicator corresponding to the managed status, and the negative competence and second behavioral indicator corresponding to the unmanaged status are identified by graph computing. The frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period is statistically analyzed using a time series analysis model to generate a first analysis result. The first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side. Acquire training data from the same period, perform statistical analysis on the performance data of trainees in various training scenarios, and generate a second analysis result. The second analysis result includes a ranking list of competency scores and a ranking list of negative evaluations of behavioral indicators from the training data side. The first analysis result and the second analysis result are fused and analyzed, and a targeted training list is calculated based on a preset weighting algorithm. The targeted training list includes competency items and behavioral indicator items that need to be trained first. By combining the existing threat training priority list, at least one threat with the highest priority is matched and optimized with the target training list to generate a training scenario configuration scheme. The training scenario configuration scheme includes the threat elements to be embedded in the training scenario and the corresponding competency items and behavioral indicators. Based on the training scenario configuration scheme, and combined with resource constraints, training course content, evaluation criteria, and training resource optimization configuration scheme are generated. The training resource optimization configuration scheme includes instructor resource configuration, simulator usage plan, and training time arrangement. The method utilizes a time series analysis model to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period, generating a first analysis result. This first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side, including: The frequency of occurrence of various competency and behavioral indicators within a preset period is statistically analyzed on a monthly basis, and a time series frequency matrix is ​​generated. A time series forecasting model was established using the Holt-Winters three-parameter exponential smoothing method. Multi-scale feature extraction and predictive analysis were performed on the frequency change trends of various competencies and behavioral indicators in the time series frequency matrix to predict the trend changes of various competencies and behavioral indicators in the future within a preset time period. The various competencies and behavioral indicators include positive competencies, the first behavioral indicator corresponding to positive competencies, negative competencies, and the second behavioral indicator corresponding to negative competencies. Set a dynamic frequency threshold, and based on the comparison result between the occurrence frequency and the dynamic frequency threshold, and the weights corresponding to the frequency dimension and the trend change dimension, generate priority scores for each competency and behavioral indicator. The priority scores of each competency are sorted from high to low, and the top N1 competency items are selected to generate a priority list of aviation safety data side competencies. The priority scores of behavioral indicators are sorted from high to low, and the top N2 competency items are selected to generate a priority list of behavioral indicators. N1 is greater than or equal to 2, and N2 is greater than or equal to 2. The aviation safety data-side competency priority list and the behavioral indicator priority list are combined into the first analysis result.

2. The method for integrating and analyzing aviation safety data and training data according to claim 1, characterized in that, The process utilizes a deep learning-based semantic analysis model to extract threats, errors, and undesirable aircraft states corresponding to various aviation safety incidents from multiple heterogeneous data sources. It then determines the crew's management status of these threats, errors, and undesirable aircraft states, which is categorized as either managed or unmanaged. This management status includes: Data cleaning and format standardization are performed on multi-source heterogeneous data to unify the timestamp format and flight number format. The multi-source heterogeneous data comes from: QAR data list, LOSA report, company unsafe incident report and aircraft maintenance data record. The multi-source heterogeneous data after unification of format is input into a semantic analysis model based on deep learning. The semantic analysis model includes an aviation-enhanced BERT encoder, an entity feature extractor, and a management state decision network. The entity feature extractor consists of a BiLSTM layer and an attention mechanism. By using an aviation-enhanced BERT encoder, domain-adaptive encoding is performed on text-based event descriptions to convert aviation terminology into fixed-dimensional vector representations containing contextual semantics. Simultaneously, numerical flight parameters are embedded to generate feature vectors that fuse multimodal information. The entity feature extractor performs sequence modeling and key information enhancement on the feature vector to identify different types of threats, including wind shear, turbulence and thunderstorms, errors including program execution errors and standard announcement errors, and undesirable aircraft states including unnecessary weather crossings, vertical deviation and horizontal deviation. Based on a pre-defined TEM classification rule base, the threats, errors, and undesirable aircraft states are automatically classified and coded to obtain a structured event coding sequence. Based on the structured event coding sequence and event handling result data, the management status of the crew for each aviation safety incident is determined through a management status determination network.

3. The method for integrating and analyzing aviation safety data and training data according to claim 2, characterized in that, The step of determining the crew's management status for each aviation safety incident based on the structured event coding sequence and event handling result data through a management status determination network includes: The structured event coding sequence and the event handling result data are subjected to feature association processing to generate an event handling association feature set containing event type, severity, handling duration and number of operation corrections; The event handling related feature set is dimensionally mapped and fused with spatiotemporal features by the feature transformation layer of the management status determination network to obtain the handling efficiency feature vector. The handling efficiency feature vector includes event handling timeliness features, accuracy features, and compliance features. The multilayer perceptron of the management status determination network performs a nonlinear transformation on the handling effectiveness feature vector, and outputs the probability distribution value of aviation unsafe events belonging to the managed state and the unmanaged state. The probability distribution value reflects the likelihood of aviation unsafe events belonging to the managed state and the unmanaged state. Based on a preset judgment threshold, the probability distribution value is judged to obtain the management status corresponding to each aviation safety event.

4. The method for integrating and analyzing aviation safety data and training data according to claim 3, characterized in that, The feature transformation layer of the management status determination network performs dimensional mapping and spatiotemporal feature fusion on the event handling related feature set to obtain a handling effectiveness feature vector. This handling effectiveness feature vector includes event handling timeliness features, accuracy features, and compliance features, including: The event handling associated feature set is aligned in time dimension, and the event type, severity, handling duration, and number of operation corrections are aligned in time sequence according to the event occurrence timestamp to generate a time alignment feature matrix. The time-aligned feature matrix is ​​subjected to multi-scale feature extraction by the temporal convolution module in the feature transformation layer to extract the processing mode features at different time scales. The spatial attention module in the feature transformation layer performs weighted fusion of treatment mode features across all time scales to generate a weighted feature representation. The weighted feature representation is input into a fully connected network in the feature transformation layer for dimensionality reduction, and the output is a handling efficiency feature vector containing event handling timeliness, accuracy, and compliance features. The event handling timeliness feature reflects the time efficiency from event occurrence to handling completion, the accuracy feature reflects the degree of conformity between the handling operation and standard procedures, and the compliance feature reflects the degree to which the handling process complies with aviation safety regulations.

5. The method for integrating and analyzing aviation safety data and training data according to claim 1, characterized in that, The mapping relationship between management status, competence, and behavioral indicators provided by the aviation safety knowledge graph, and the identification of positive competence and first behavioral indicator corresponding to the managed status, and negative competence and second behavioral indicator corresponding to the unmanaged status through graph computation, include: Construct an aviation safety knowledge graph, in which: threats, errors, and undesirable aircraft states are respectively regarded as corresponding event entity nodes; competencies and their corresponding behavioral indicators are regarded as capability entity nodes, and are interconnected through attribute relationships; managed states and unmanaged states are regarded as relationship types, used to connect event entity nodes and capability entity nodes. The aviation safety knowledge graph is stored using the Neo4j graph database, and a semantic relationship pattern between event entity nodes and capability entity nodes is defined, wherein node attributes include node ID, node type, node name and weight value; For each aviation safety incident that has been determined to be under management status, a graph traversal query is performed based on the corresponding relationship path selected according to the management status type. Specifically: for events under management status, all positive competency nodes and corresponding first behavioral indicator nodes associated with the events under management status are traversed and queried along the managed relationship path; for events under unmanaged status, all negative competency nodes and corresponding second behavioral indicator nodes associated with the events under unmanaged status are traversed and queried along the unmanaged relationship path.

6. The method for integrating and analyzing aviation safety data and training data according to claim 1, characterized in that, The first analysis result and the second analysis result are fused and analyzed to calculate a targeted training list based on a preset weighted algorithm. The targeted training list includes competency items and behavioral indicator items that need to be trained first, including: The priority scores in the first analysis results are subjected to Min-Max normalization to map the priority scores to a preset numerical range, thereby obtaining a security data weight vector. The rating ranking and the number of negative reviews in the second analysis result are Z-score standardized and converted into training data weight vectors within a preset numerical range; A weighted summation algorithm is used, combined with weighting coefficients set by domain experts, to calculate the comprehensive priority score for each competency item and behavioral indicator item; The comprehensive priority scores are sorted from high to low, and a targeted training list is generated based on the sorting results.

7. An integrated analysis system for aviation safety data and training data, characterized in that, include: The extraction module is used to extract the threats, errors and undesirable aircraft states corresponding to each aviation safety incident from multiple heterogeneous data sources using a semantic analysis model based on deep learning, and to 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. The identification module is used to identify the positive competence and first behavioral indicator corresponding to the managed state, and the negative competence and second behavioral indicator corresponding to the unmanaged state, based on the mapping relationship between management status, competence and behavioral indicators provided by the aviation safety knowledge graph and graph computing. The first analysis module is used to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period using a time series analysis model, and generate a first analysis result. The first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side. The second analysis module is used to acquire training data from the same period, perform statistical analysis on the performance data of trainees in various training scenarios, and generate a second analysis result. The second analysis result includes a ranking list of competency scores and a ranking list of negative evaluations of behavioral indicators from the training data side. The fusion analysis module is used to fuse the first analysis result and the second analysis result, and calculate the target training list based on the preset weighting algorithm. The target training list includes competency items and behavioral indicator items that need to be trained first. The matching optimization module is used to combine the existing threat training priority list, match and optimize at least one threat with the target training list, and generate a training scenario configuration scheme. The training scenario configuration scheme includes the threat elements to be embedded in the training scenario and the corresponding competency items and behavioral indicators. The generation module is used to generate training course content, evaluation criteria, and training resource optimization configuration scheme based on the training scenario configuration scheme and resource constraints. The training resource optimization configuration scheme includes instructor resource configuration, simulator usage plan, and training time arrangement. The method utilizes a time series analysis model to statistically analyze the frequency of occurrence of the positive competence, the first behavioral indicator, the negative competence, and the second behavioral indicator within a preset period, generating a first analysis result. This first analysis result includes a competence priority list and a behavioral indicator priority list from the aviation safety data side, including: The frequency of occurrence of various competency and behavioral indicators within a preset period is statistically analyzed on a monthly basis, and a time series frequency matrix is ​​generated. A time series forecasting model was established using the Holt-Winters three-parameter exponential smoothing method. Multi-scale feature extraction and predictive analysis were performed on the frequency change trends of various competencies and behavioral indicators in the time series frequency matrix to predict the trend changes of various competencies and behavioral indicators in the future within a preset time period. The various competencies and behavioral indicators include positive competencies, the first behavioral indicator corresponding to positive competencies, negative competencies, and the second behavioral indicator corresponding to negative competencies. Set a dynamic frequency threshold, and based on the comparison result between the occurrence frequency and the dynamic frequency threshold, and the weights corresponding to the frequency dimension and the trend change dimension, generate priority scores for each competency and behavioral indicator. The priority scores of each competency are sorted from high to low, and the top N1 competency items are selected to generate a priority list of aviation safety data side competencies. The priority scores of behavioral indicators are sorted from high to low, and the top N2 competency items are selected to generate a priority list of behavioral indicators. N1 is greater than or equal to 2, and N2 is greater than or equal to 2. The aviation safety data-side competency priority list and the behavioral indicator priority list are combined into the first analysis result.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the integrated analysis method for aviation safety data and training data as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the integration and analysis method of aviation safety data and training data as described in any one of claims 1 to 6.

Images