Flight technology quality assessment methods, systems and electronic equipment

Abstract

This application discloses a method, system, and electronic device for flight technical quality assessment. The method includes: selecting parameter types from historical QAR data based on the maximum relevance and minimum redundancy criterion as indicators for assessing flight quality; determining the weight coefficients of each indicator based on historical QAR data; determining the standard thresholds of each indicator based on historical target flight data; training and generating a personal flight habit model corresponding to the pilot to be assessed based on the first indicator parameter value, weight coefficient, and standard threshold of the pilot to be assessed in historical QAR data; and determining the flight quality assessment result corresponding to the pilot to be assessed based on the second indicator parameter value of the pilot to be assessed in current QAR data, the personal flight habit model corresponding to the pilot to be assessed, as well as the weight coefficient and standard threshold.

Description

Technical Field

[0001] This application relates to the field of flight data processing technology, specifically to flight technical quality assessment methods, systems, and electronic equipment. Background Technology

[0002] Currently, flight skill quality assessment models generally employ a hierarchical labeling system to characterize pilots' technical capabilities. This typically involves extracting risk-layer labels (such as taxiing speed, takeoff attitude, landing gear retraction height, localizer maintenance, 5ft descent rate, landing load, and landing curve control) from raw flight data (e.g., Quick Access Recorder (QAR) data, Line Operations Safety Audit (LOSA) reports, and unsafe incident records). These labels are then summarized into training-layer labels (taxiing capability, takeoff capability, climb and cruise landing capability, approach capability, and landing capability), and finally, a decision-level profile is formed through weighting. In terms of modeling methods, most models combine human experience with machine learning techniques, with human modeling being more widely adopted due to its ability to incorporate domain expert knowledge.

[0003] However, existing flight technology quality assessment methods based on artificial modeling have many drawbacks, such as strong subjectivity in the assessment system, lack of data-driven objectivity, strong commonality in assessment standards and lack of individual adaptability, coarse analysis granularity and lack of practicality, neglect of individual dynamic evolution and habitual patterns of pilots, and static and rigid models that lack adaptability and continuous evolution capabilities. Summary of the Invention

[0004] This application aims to address one of the technical problems in related technologies to a certain extent. To this end, this application provides a flight technology quality assessment method, system, and electronic equipment to improve the objectivity of the assessment system, balance the commonality and individual adaptability of assessment standards, improve the granularity and practicality of analysis, identify the individual dynamic evolution and habitual patterns of pilots, and enhance the adaptive and continuous evolution capabilities of the assessment.

[0005] To achieve the above objectives, this application adopts the following technical solution: A method for evaluating flight technical quality, comprising: Based on the maximum relevance and minimum redundancy criterion, parameter types are selected from historical QAR data as indicators for evaluating flight quality. Based on historical QAR data, determine the weighting coefficients for each indicator; Based on historical target flight data, standard thresholds for each indicator are determined. Based on the first indicator parameter values, weighting coefficients, and standard thresholds of the pilots to be evaluated in historical QAR data, a personal flight habit model corresponding to the pilots to be evaluated is trained and generated. Based on the second indicator parameter value of the pilot to be evaluated in the current QAR data, the personal flight habit model of the pilot to be evaluated, as well as the weighting coefficient and standard threshold, the flight quality assessment result of the pilot to be evaluated is determined.

[0006] In some feasible implementations, a personal flight habit model for the pilot to be evaluated is trained and generated based on the first indicator parameter value, weighting coefficient, and standard threshold of the pilot to be evaluated in historical QAR data, including: Construct a reward function based on each weight coefficient and each standard threshold; Based on the first indicator parameter value, a state vector is constructed. According to the reward function, the reward value corresponding to each state vector is calculated. The optimization objective is to maximize the cumulative reward value. A personal flight habit model is generated by training based on the reinforcement learning algorithm.

[0007] In some feasible implementations, a state vector is constructed based on the first index parameter value. The reward value corresponding to each state vector is calculated according to the reward function, with maximizing the cumulative reward value as the optimization objective. A personal flight habit model is then trained using a reinforcement learning algorithm, including: Based on each flight flown by the pilot to be evaluated in historical QAR data, the first indicator parameter value for that flight is obtained, and the deviation of the obtained first indicator parameter value from the baseline of the historical flight indicator parameter values ​​of the pilot to be evaluated is determined; wherein, the state vector includes the first indicator parameter value and the corresponding deviation. Based on the actual flight data of the pilot to be evaluated in each flight, the actions taken by the pilot to be evaluated in each flight under the state vector are inferred posteriorly. Based on the reward function, calculate the reward value obtained by the pilot to be evaluated after performing the action in each flight; Using state vectors, actions, and reward values ​​as samples, the Q-Learning algorithm is used to iteratively update and maximize the cumulative reward value to determine the Q-value function of the pilot to be evaluated; A personal flight habit model is generated by training based on the Q-value function.

[0008] In some feasible implementations, a reward function is constructed based on each weight coefficient and each standard threshold, including: Based on the standard threshold values ​​of the indicator parameters, the corresponding values ​​of the individual safety scores for each indicator are determined by mapping. The security compliance reward component is constructed by weighting and aggregating each weight coefficient and each individual security score. Calculate the green economy reward based on the preset green economy indicators and their corresponding weights; The reward function is determined by weighting and combining the safety compliance reward component and the green economy reward component.

[0009] In some feasible implementations, the flight quality assessment result for the pilot to be assessed is determined based on the second indicator parameter value of the pilot to be assessed in the current QAR data, the pilot's corresponding personal flight habit model, and the weighting coefficients and standard thresholds, including: Based on the standard thresholds, the values ​​of the second indicator parameters of the pilots to be evaluated in the current QAR data are mapped to the individual accuracy scores of each indicator. Based on the weighting coefficients, the accuracy scores of each individual item are weighted and aggregated to determine the overall accuracy score; Determine the deviation of the second indicator parameter value from the baseline of the historical flight indicator parameter values ​​of the pilot to be evaluated, and construct the current state vector based on the second indicator parameter value and the corresponding deviation. Input the current state vector into the individual flight habit model to determine the overall stability score; The flight quality assessment results for the pilots to be evaluated are determined based on the overall accuracy score and the overall stability score.

[0010] In some feasible implementations, parameter types are selected from historical QAR data based on the maximum relevance and minimum redundancy criterion as indicators for evaluating flight quality, including: Obtain the parameter values ​​of each candidate parameter and the corresponding flight quality label from the historical QAR data; Based on the Pearson correlation coefficient, the correlation between each candidate parameter and the flight quality label was calculated; Based on the information gain of Pearson correlation coefficient and / or information entropy, calculate the redundancy between any two candidate parameters; Based on the maximum correlation and minimum redundancy criterion, indicators for evaluating flight quality are selected according to correlation and redundancy.

[0011] In some feasible implementations, based on the maximum correlation and minimum redundancy criterion, indicators for evaluating flight quality are screened according to correlation and redundancy. The screening of these indicators includes: Initialize the selected indicator set to empty; Identify the target candidate parameters that have the strongest correlation with the flight quality labels from the candidate parameters; Based on the target candidate parameters, generate a set of selected target indicators; The following steps are executed iteratively until a preset stopping condition is met, at which point the final selected indicators are used as indicators for evaluating flight quality: The comprehensive score is calculated based on the correlation between the remaining candidate parameters of the target and the flight quality label, as well as the redundancy between them and the indicators in the target's selected indicator set. Among them, the remaining candidate parameters of the target correspond to the remaining candidate parameters of the target that were not selected from the candidate parameters; Select the remaining candidate parameters of the target with the highest comprehensive score and add them to the target's selected indicator set.

[0012] In some feasible implementations, the types of indicators include habitual indicators, absolute indicators, and Boolean indicators; Based on historical target flight data, standard thresholds for each indicator are determined, including: Statistical analysis of the numerical distribution of each indicator in historical target flight data; For habitual indicators, based on the preset excellent coverage ratio, the values ​​at the corresponding lower and higher quantiles are determined and used as the first quality threshold and the second quality threshold, respectively; wherein, the first quality threshold is lower than the second quality threshold. For absolute indicators, based on the preset optimal value point and preset warning multiple, the first and second warning boundaries, which are composed of multiple standard deviations of the mean, are determined, and one or both are selected as standard thresholds according to the physical meaning of the indicator. The first warning boundary is larger than the second warning boundary; For Boolean indicators, 0 and 1 are used as the standard thresholds for events that have not occurred and events that have occurred, respectively.

[0013] Furthermore, this application also provides a flight technology quality assessment system, including: The filtering module is used to filter parameter types from historical QAR data based on the maximum relevance and minimum redundancy criterion, as indicators for evaluating flight quality. The first determination module is used to determine the weight coefficients of each indicator based on historical QAR data; The second determination module is used to determine the standard thresholds for each indicator based on historical target flight data. The training module is used to train and generate a personal flight habit model for the pilot to be evaluated based on the first indicator parameter value, weight coefficient and standard threshold of the pilot to be evaluated in historical QAR data. The third determination module is used to determine the flight quality assessment result of the pilot to be evaluated based on the second indicator parameter value of the pilot to be evaluated in the current QAR data, the personal flight habit model of the pilot to be evaluated, as well as the weighting coefficient and standard threshold.

[0014] In addition, this application also provides an electronic device, including: One or more processors; A memory having stored one or more computer programs that, when executed by one or more processors, cause the one or more processors to implement any of the flight technical quality assessment methods described above.

[0015] The flight technical quality assessment method, system, and electronic equipment provided in this application include the following methods: selecting parameter types from historical QAR data based on the maximum relevance and minimum redundancy criterion, as indicators for assessing flight quality; determining the weight coefficients of each indicator based on historical QAR data; determining the standard thresholds of each indicator based on historical target flight data; training and generating a personal flight habit model corresponding to the pilot to be assessed based on the first indicator parameter value, weight coefficient, and standard threshold of the pilot to be assessed in historical QAR data; and determining the flight quality assessment result corresponding to the pilot to be assessed based on the second indicator parameter value of the pilot to be assessed in current QAR data, the personal flight habit model corresponding to the pilot to be assessed, as well as the weight coefficient and standard threshold. Historical QAR data is used to drive indicator selection and weight coefficient calculation to improve the objectivity of the evaluation system. By training individual flight habit models and obtaining group standard thresholds from historical target flight data, the evaluation combines individual characteristics with group commonalities to balance the commonality of evaluation standards with individual adaptability. Training individual flight habit models using historical QAR data before evaluating current QAR data (single flight data) significantly improves analytical granularity. Using the indicator parameter values ​​of the pilot being evaluated in historical QAR data to train the pilot's individual flight habit model allows learning of stable patterns formed during long-term flight, identifying individual dynamic evolution and habitual patterns. By binding the indicator system, weights, standard thresholds, and individual flight habit models to updatable historical data, all elements naturally possess the ability to adapt to environmental changes rather than being fixed. Therefore, when historical data is updated, the above process can be re-executed, allowing all elements (indicators, weight coefficients, standard thresholds, individual flight habit models, etc.) to automatically adapt to the latest "historical data," enhancing adaptability and continuous evolution.

[0016] These features and advantages of this application will be disclosed in detail in the following specific embodiments and accompanying drawings. The best embodiments or means of this application will be shown in detail in conjunction with the accompanying drawings, but are not intended to limit the technical solutions of this application. In addition, each of these features, elements and components appearing in the following text and drawings is multiple and is labeled with different symbols or numbers for convenience, but all represent parts with the same or similar structure or function. Attached Figure Description

[0017] The following description, in conjunction with the accompanying drawings, further illustrates this application: Figure 1This is a flowchart illustrating the flight technology quality assessment method provided in the embodiments of this application; Figure 2 A flowchart illustrating the process of training and generating a personal flight habit model, provided as an embodiment of this application; Figure 3 This is a flowchart illustrating a process for training and generating a personal flight habit model based on a reinforcement learning algorithm, as provided in an embodiment of this application. Figure 4 This is a flowchart illustrating how a reward function is constructed based on various weight coefficients and standard thresholds, as provided in an embodiment of this application. Figure 5 This is a flowchart illustrating a process for determining the flight quality assessment result for a pilot to be evaluated, as provided in an embodiment of this application. Figure 6 This is a flowchart illustrating a screening index provided in an embodiment of this application. Figure 7 A flowchart illustrating yet another screening criterion provided in an embodiment of this application; Figure 8 This is a flowchart illustrating a method for determining standard thresholds for various indicators, provided as an embodiment of this application. Figure 9 A structural schematic diagram of a flight technology quality assessment system provided in this application embodiment; Figure 10 This is a structural schematic diagram of an electronic device provided in an embodiment of this application; Figure 11 This is a structural schematic diagram of a computer-readable medium provided in an embodiment of this application.

[0018] Explanation of reference numerals in the attached figures 101: Processor, 102: Memory, 103: I / O interface, 104: Bus, 201: Filtering module, 202: First determination module, 203: Second determination module, 204: Training module, 205: Third determination module. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments in the implementation are intended to explain this application and should not be construed as limiting this application.

[0020] The terms "an embodiment," "example," or "example" used in this specification refer to a particular feature, structure, or characteristic described in connection with the embodiment itself that may be included in at least one embodiment disclosed in this application. The phrase "in an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0021] Currently, flight skill quality assessment models generally employ a hierarchical labeling system to characterize pilots' technical capabilities. This typically involves extracting risk-layer labels (such as taxiing speed, takeoff attitude, landing gear retraction height, localizer maintenance, 5ft descent rate, landing load, and landing curve control) from raw flight data (e.g., Quick Access Recorder (QAR) data, Line Operations Safety Audit (LOSA) reports, and unsafe incident records). These labels are then summarized into training-layer labels (taxiing capability, takeoff capability, climb and cruise landing capability, approach capability, and landing capability), and finally, a decision-level profile is formed through weighting. In terms of modeling methods, most models combine human experience with machine learning techniques, with human modeling being more widely adopted due to its ability to incorporate domain expert knowledge.

[0022] However, existing flight technology quality assessment methods based on manual modeling have the following drawbacks: 1. The assessment system is highly subjective and lacks data-driven objectivity. The selection of indicators and their weighting rely on expert experience, leading to pre-existing subjective biases. Differences in risk perception and indicator importance among experts directly affect the indicator structure and weighting, making it difficult for the assessment results to fully reflect pilots' actual technical performance and safety contributions. In the field of aviation safety, subjective weighting may weaken the actual impact of key operational parameters (such as energy management and standard call-out compliance rate) while amplifying the role of non-critical qualitative indicators, resulting in distorted assessments.

[0023] 2. The evaluation criteria are too standardized and lack individual adaptability. The current evaluation method, which divides pilots into "high-level" and "low-level" categories based on a unified industry standard, essentially uses a common yardstick to measure all pilots, ignoring the individual performance of pilots in terms of experience, technical characteristics, physiological and psychological state, and specific operating environments. This model is unable to identify subtle technical declines in experienced pilots under complex conditions and may also misjudge the normal fluctuations in the habit curve of new pilots.

[0024] 3. The analysis granularity is coarse and lacks practicality. Current models provide comprehensive profiles based on fixed flight batches (e.g., 50-400 flights), obscuring key details and immediate risks within a single flight, thus failing to meet the needs of refined safety management. Furthermore, the data relied upon is mostly manually processed qualitative descriptions or aggregated scores, rather than the raw high-frequency parameters from the Flight Data Recorder (FDR / QAR). This approach prevents pilots from directly linking assessment conclusions to specific maneuvers, system states, or environmental parameters, resulting in assessment results detached from real-world scenarios and limiting their diagnostic and improvement guidance.

[0025] 4. Existing models suffer from numerous drawbacks, such as neglecting the individual dynamic evolution and habitual patterns of pilots. They treat pilots as static evaluation subjects, failing to fully consider the dynamic changes in individual capabilities with factors such as training, experience, and fatigue cycles. They also fail to identify and quantify long-established habitual operating patterns (such as typical approach attitude preferences and throttle management styles). The lack of establishing and analyzing longitudinal baselines of individual historical performance makes it difficult for models to predict declines in performance relative to their baseline, and also makes it impossible to predict individual behavioral tendencies under specific pressure or complex situations.

[0026] 5. Static models lack adaptability and continuous evolution capabilities. Currently, once a model is deployed, its indicator system, weights, and evaluation logic are often fixed for a long period, making it unable to adapt to changes in the external environment, such as regulatory updates, aircraft upgrades, changes in route characteristics, and generational shifts in pilot skill levels. This static nature leads to "model drift," meaning that the model's evaluation standards gradually deviate from actual safe operating conditions, and its evaluation effectiveness decreases over time.

[0027] In response, the applicant of this application proposes that historical QAR data can be used to drive indicator selection and indicator weight coefficient calculation to improve the objectivity of the evaluation system; individual flight habit models can be trained and group standard thresholds can be obtained from historical target flight data to combine individual characteristics with group commonalities for evaluation, thereby balancing the commonality of evaluation standards and individual adaptability; individual flight habit models can be trained using historical QAR data before evaluating current QAR data, i.e., single flight data, to significantly improve the granularity of analysis; the indicator parameter values ​​of the pilot being evaluated in historical QAR data can be used to train the pilot's individual flight habit model, learning the stable patterns formed by the pilot in long-term flight, to identify the individual dynamic evolution and habitual patterns of the pilot; by binding the indicator system, weights, standard thresholds, and individual flight habit models all to updatable historical data, all elements naturally possess the ability to adapt to environmental changes rather than being fixed, so that by re-executing the above process when historical data is updated, all elements (indicators, weight coefficients, standard thresholds, individual flight habit models, etc.) can automatically adapt to the latest "historical data", improving the ability to adapt and continuously evolve.

[0028] As a first aspect of the embodiments of this application, a method for evaluating flight technology quality is provided. Figure 1 This is a flowchart illustrating the flight technology quality assessment method 100 provided in this application embodiment, as shown below. Figure 1 As shown, method 100 includes: Step S110: Based on the maximum correlation and minimum redundancy criterion, filter parameter types from historical QAR data as indicators for evaluating flight quality.

[0029] For example, the maximum correlation and minimum redundancy criterion is a purely data-driven feature selection method. By calculating the correlation between indicators and flight technical quality (maximum correlation) and the redundancy between indicators (minimum redundancy), it automatically selects the most representative indicators from the original QAR parameters, eliminating the arbitrariness of expert subjective selection. Furthermore, the basis for indicator selection is "historical QAR data." QAR itself is a raw high-frequency parameter, not a qualitative description processed manually. All subsequent evaluations (including weight calculation, standard determination, and model training) are based on the raw data, and the evaluation conclusions can be directly traced back to specific control parameters, which can significantly improve the granularity of analysis.

[0030] Step S120: Determine the weighting coefficients of each indicator based on historical QAR data.

[0031] For example, the calculation of weighting coefficients also depends on data, and the weighting coefficients are determined by the statistical characteristics of the data itself (such as the degree of variation and the conflict between indicators), rather than expert scores.

[0032] The above steps S110 and S120 use historical QAR data to drive indicator selection and indicator weight coefficient calculation, replacing expert experience with data statistics, which can effectively improve the objectivity of the evaluation system.

[0033] Step S130: Determine the standard threshold for each indicator based on historical target flight data.

[0034] For example, historical target flight data refers to historical flight data under compliant, safe, and normal operating conditions, and the standard threshold represents the commonalities of the group.

[0035] Step S140: Based on the first indicator parameter value, weight coefficient and standard threshold of the pilot to be evaluated in the historical QAR data, train and generate the personal flight habit model corresponding to the pilot to be evaluated.

[0036] For example, instead of training a universal model applicable to all pilots, each pilot is trained with their own personalized flight habit model. This personal flight habit model represents individual characteristics. Training the personal flight habit model uses the pilot's own "historical QAR data," which naturally constructs the pilot's longitudinal historical baseline. The model learns the habitual, stable patterns formed by the pilot over long-term flight, laying the foundation for subsequent evaluations "relative to their own baseline." Furthermore, when a pilot's skill improves with training or their condition declines due to fatigue, the difference between their current operation and the historical baseline (the habits represented by the model) is captured, reflected in changes in the evaluation results, and the individual dynamic evolution of the pilot can be identified.

[0037] The standard thresholds and individual flight habit models determined in steps S130 and S140 above together determine the flight technology quality assessment results, naturally realizing a two-dimensional assessment of "common standards + individual habits", balancing the commonality of assessment standards and individual adaptability.

[0038] Step S150: Based on the second indicator parameter value of the pilot to be evaluated in the current QAR data, the personal flight habit model of the pilot to be evaluated, as well as the weighting coefficient and standard threshold, determine the flight quality evaluation result of the pilot to be evaluated.

[0039] For example, the evaluation is based on the current QAR data. Here, "current" naturally refers to a single flight, rather than an aggregated batch of several flights, thus achieving refined analysis at the single-flight level. It can be understood that the first indicator parameter value and the second indicator parameter value involved in steps S140 and S150, respectively, are the indicator parameter values ​​corresponding to the pilot to be evaluated. "First" and "second" are only used to distinguish the sources of the two.

[0040] This method binds the indicator system, weights, standard thresholds, and personal flight habit models to updatable historical data. All elements are inherently capable of adapting to environmental changes rather than being fixed. Therefore, by re-executing the above process when historical data is updated, all elements (indicators, weight coefficients, standard thresholds, personal flight habit models, etc.) can automatically adapt to the latest "historical data," thereby improving the ability to adapt and continuously evolve.

[0041] In some embodiments, a training model for generating a personal flight habit model is also provided (corresponding to step S140 above: training a personal flight habit model for the pilot to be evaluated based on the first indicator parameter value, weight coefficient and standard threshold of the pilot to be evaluated in historical QAR data). Figure 2 This is a flowchart illustrating a process for training and generating a personal flight habit model, as provided in an embodiment of this application. Figure 2 As shown, it includes: Step S210: Construct a reward function based on each weight coefficient and each standard threshold.

[0042] For example, a reward function is constructed using the weighting coefficients of the indicators and standard thresholds, so that the personal flight habit model is constrained by the group's operational characteristics while learning the individual pilot's operational characteristics.

[0043] Step S220: Based on the first index parameter value, construct a state vector, calculate the reward value corresponding to each state vector according to the reward function, take maximizing the cumulative reward value as the optimization objective, and train and generate a personal flight habit model based on reinforcement learning algorithm.

[0044] For example, a state vector is constructed based on the pilot's own historical QAR data. By maximizing the cumulative reward through reinforcement learning, the trained model is a digital representation of the pilot's stable operating habits. The trained model can serve as a dynamic baseline for the pilot's own historical performance, providing a basis for comparison with their personal habits in subsequent evaluations.

[0045] In some embodiments, a personal flight habit model is also provided based on reinforcement learning algorithm training (corresponding to step S220 above: constructing a state vector based on the first index parameter value, calculating the reward value corresponding to each state vector according to the reward function, taking maximizing the cumulative reward value as the optimization objective, and training a personal flight habit model based on reinforcement learning algorithm). Figure 3 This application provides a flowchart illustrating a process for training and generating a personal flight habit model based on a reinforcement learning algorithm, as shown in the embodiments below. Figure 3 As shown, it includes: Step S310: Based on each flight flown by the pilot to be evaluated in the historical QAR data, obtain the first index parameter value for that flight, and determine the deviation of the obtained first index parameter value from the baseline of the historical flight index parameter values ​​of the pilot to be evaluated; wherein, the state vector includes the first index parameter value and the corresponding deviation.

[0046] For example, by calculating the deviation of the first indicator parameter value from the baseline of the individual's historical flight indicator parameter value, a state vector that reflects the deviation of the pilot's current operation from his own habitual deviation is constructed, thereby capturing personalized characteristics in flight.

[0047] Step S320: Based on the actual flight data of the pilot to be evaluated in each flight, the actions taken by the pilot to be evaluated in each flight under the state vector are inferred posteriorly.

[0048] For example, the actual flight data of the pilot to be evaluated in each flight includes not only the index parameter values, but also a large number of other parameter values ​​that were not selected as indicators. The actions taken by the pilot to be evaluated in the state vector (i.e., possible behaviors that may be taken in a given state) can be inferred from its posterior, thereby reconstructing the pilot's decision-making behavior and establishing the correspondence between state and action.

[0049] Step S330: Calculate the reward value obtained by the pilot to be evaluated after performing actions in each flight, based on the reward function.

[0050] For example, the reward function is a criterion for evaluating the pilot's operational performance. Each action is quantitatively scored according to the reward function to measure the quality of the operation.

[0051] Step S340: Using the state vector, action, and reward value as samples, the Q-Learning algorithm is used to iteratively update the function to maximize the cumulative reward value, thereby determining the Q-value function of the pilot to be evaluated.

[0052] For example, by iteratively updating the Q-value function through the Q-Learning algorithm, the long-term cumulative reward that a pilot can obtain by taking various actions under different conditions can be learned, thereby quantifying the merits of their operating habits.

[0053] Step S350: Based on the Q-value function, train and generate a personal flight habit model.

[0054] For example, a personal flight habit model is ultimately generated, which can characterize the pilot's unique operational preferences and decision-making patterns, and can be used for subsequent flight technology quality assessment and auxiliary analysis.

[0055] As a preferred implementation, the Q-Learning algorithm is used for iterative updates to maximize the cumulative reward value, which can be achieved through the following update rules: (1); in, For the state vector Take action below Q value, For learning rate, To perform the action The reward value obtained later As a discount factor, For the next state vector The maximum Q value for all possible actions.

[0056] By iteratively applying the above update rules until the Q-value function converges, the individual flight habit model of the pilot to be evaluated can be obtained.

[0057] As a preferred implementation, a hierarchical temporal fusion mechanism can be further introduced when constructing the state vector to balance the stability and evolution of pilot habits. Specifically, when calculating the deviation of the index parameter values ​​from the historical flight index parameter value baseline, multiple deviations can be calculated based on historical data from three different time windows: recent (e.g., the last 30 flights), medium-term (e.g., 100 flights), and long-term (e.g., 200 flights), and denoted as follows: , , Then, the original parameter values ​​of the same indicator, along with these three deviations, are used as elements of the state vector corresponding to that indicator. In this way, the state vector still contains "indicator parameter values ​​and corresponding deviations," but the deviations are refined into deviations at multiple time scales, enabling the model to simultaneously perceive pilots' short-term fluctuations, medium-term trends, and long-term habits. Deviations from different time windows can be dynamically fused through attention mechanisms or learnable weight networks, thereby adaptively determining the influence weights of habits from different periods on the current state. This mechanism allows the model to retain long-term stable habits while sensitively capturing recent subtle changes, further improving the accuracy of personalized assessments.

[0058] As a preferred implementation, when training a personal flight habit model using the Q-Learning algorithm, a model adaptive initialization mechanism can be further introduced to address the cold start problem caused by insufficient data for new pilots.

[0059] In some feasible implementations, the model adaptive initialization mechanism can be implemented in the following way: For example, parameter initialization can be based on meta-learning. A meta-Q network is pre-trained on a large amount of historical QAR data from pilots, learning the general state-action value mapping relationship for the pilot group. When a new pilot is encountered, the parameters of the meta-Q network are used as the initial parameters of their personal Q network. Then, the Q network is quickly fine-tuned using only a small amount of the pilot's historical flight data to obtain their personal flight habit model. This fine-tuning process can still use the Q-Learning iterative update method described above.

[0060] Alternatively, a personalized layer structure can be parameterized. The Q-network can be designed as a structure containing a global shared layer and a pilot-specific adaptation layer. The global shared layer is pre-trained and fixed on data from all pilots, learning common flight operation characteristics; the adaptation layer parameters are dynamically generated based on each pilot's individual data. During training, only the adaptation layer parameters are iteratively updated using Q-Learning, while the global shared layer remains unchanged. This structure can significantly reduce the amount of training data required for new pilots while ensuring personalized adaptation.

[0061] The adaptive initialization mechanism described above does not change the Q-Learning training framework mentioned above, but only provides optimal solutions for model initialization and network structure, which can effectively improve the practicality and efficiency of personalized evaluation.

[0062] As a preferred implementation method, after obtaining a personal flight habit model, in order to further adapt to the dynamic evolution of pilot skills and changes in the external environment, a continuous learning and optimization mechanism can be introduced to update the model regularly.

[0063] In some feasible implementations, a data stream processing framework can be introduced.

[0064] For example, a fixed update cycle (such as every two weeks) is set to import newly generated flight data into the training set, re-triggering the model update process so that the model can promptly absorb the latest operating habits of pilots.

[0065] In some feasible implementations, catastrophic forgetting can be overcome by adopting the following methods: For example, during incremental learning, continuous learning techniques such as Elastic Weight Consolidation (EWC), Experience Replay Buffer, or regularization constraints are employed to ensure that the model does not forget previously formed stable habit features while learning new habits. For instance, a portion of historical flight samples can be retained through the Experience Replay Buffer and mixed with new samples during each update to maintain the memory of historical habits.

[0066] The aforementioned continuous learning and optimization mechanism does not change the Q-Learning training framework mentioned above, but only adds a periodic update step after the model is deployed, enabling the personal flight habit model to have the ability to evolve dynamically and always maintain a high degree of consistency with the pilot's current state.

[0067] In some embodiments, a reward function is also provided based on each weight coefficient and each standard threshold (corresponding to step S210 above: constructing a reward function based on each weight coefficient and each standard threshold). Figure 4 This application provides a flowchart illustrating the construction of a reward function based on weight coefficients and standard thresholds, as shown in the embodiments. Figure 4 As shown, it includes: Step S410: Based on the standard threshold values ​​of the indicator parameters, map and determine the corresponding values ​​of the individual safety scores for each indicator.

[0068] For example, each indicator (such as "ground load") has a standard threshold for evaluating its flight quality (e.g., <1.3G is excellent, 1.3-1.6G is good, >1.6G is poor). The specific values ​​of the pilot's flight (e.g., 1.4G) are mapped to a single safety score (e.g., 75 points). The units and meanings of the original data are different (some are better the smaller, some are better the more stable). They are unified into a comparable "score" to facilitate subsequent calculations.

[0069] Step S420: Construct a security compliance reward component based on the weighted aggregation of each weight coefficient and each individual security score.

[0070] For example, different indicators have different levels of importance (e.g., "ground load" may be more important than "taxi speed"). The individual safety scores for each indicator are calculated and then summed according to their respective weighting coefficients to obtain a total score. This total score represents the overall performance of the flight in the "safety compliance" dimension. Indicators with higher weighting coefficients have a greater impact on the total score, while those with lower weighting coefficients have a smaller impact.

[0071] Step S430: Calculate the green economy reward amount based on the preset green economy indicators and their corresponding weights.

[0072] For example, in addition to safety, "green economy" (such as fuel saving and emission reduction) should also be considered. A separate green economy score should be calculated for indicators related to fuel efficiency (such as "tipping off the accelerator" and "cruise altitude maintenance"), and environmental targets related to the green QAR should also be included in the evaluation system.

[0073] Step S440: Weight the safety compliance reward component and the green economy reward component to determine the reward function.

[0074] For example, the safety compliance reward component and the green economy reward component are added together in a certain proportion (e.g., safety accounts for 80% and green accounts for 20%) to obtain the final comprehensive reward function. The final reward function can simultaneously reflect the performance of both "safety compliance" and "green economy" dimensions. It is understandable that the proportions of the safety compliance reward component and the green economy reward component are adjustable (for example, airlines can adjust the emphasis on safety and green aspects at different times), further enhancing the flexibility of the calculation.

[0075] As a preferred implementation method, the security compliance reward can be calculated using the following formula: (2); Rsafe represents the security compliance reward component. Let j be the weight coefficient of the j-th indicator. The total number of indicators. For the j-th indicator, the individual safety score is... This represents the parameter value of the j-th indicator.

[0076] In some embodiments, a method is also provided for determining the flight quality assessment result corresponding to the pilot to be assessed (corresponding to step S150 above: determining the flight quality assessment result corresponding to the pilot to be assessed based on the second indicator parameter value of the pilot to be assessed in the current QAR data, the personal flight habit model corresponding to the pilot to be assessed, and the weighting coefficient and standard threshold). Figure 5 This application provides a flowchart illustrating how to determine the flight quality assessment result for a pilot to be evaluated, as illustrated in the embodiments of this application. Figure 5 As shown, it includes: Step S510: Based on each standard threshold, map the values ​​of each second indicator parameter of the pilot to be evaluated in the current QAR data to the individual accuracy scores of each indicator.

[0077] For example, the quality of the current flight operation is evaluated by calculating the individual accuracy score of the indicator using standard thresholds. Indicators can be categorized into habitual indicators, absolute indicators, and Boolean indicators. Habitual indicators are scored according to convention within the standard range, such as throttle reduction altitude and taxi speed. Absolute indicators are scored as lower or higher as possible, such as heading deviation and touchdown bank angle. Boolean indicators are scored as follows: 0 is good, and everything else is poor, such as low-altitude go-around. For absolute indicators, 0 and 6 delta are the standard thresholds. Scoring can be forced at 0 and outside of 6 delta, with 100 points at 0 and 0 points outside of 6 delta. Scoring between 0 and 6 delta can be achieved through non-linear mapping using modeling. For Boolean indicators, 0 and 1 are the standard thresholds themselves, and a score of 0 or 1 can be directly forced. For habitual indicators, the accuracy score of a single item can be determined to fall within a certain range. For example, if historical target flight data shows that "330-380 ft" is the optimal range for throttle reduction altitude, then: if the throttle reduction altitude measurement value <= 330, then low zone (0,60]; if 330 < throttle reduction altitude measurement value <= 380, then medium zone (60,90]; if the throttle reduction altitude measurement value > 380, then high zone (90,100).

[0078] Step S520: Based on each weight coefficient, the accuracy scores of each individual item are weighted and aggregated to determine the overall accuracy score.

[0079] For example, the overall accuracy score is obtained by weighting and aggregating the individual accuracy scores of each indicator based on the weight coefficient of each indicator.

[0080] Step S530: Determine the deviation of the second indicator parameter value from the baseline of the historical flight indicator parameter values ​​of the pilot to be evaluated, and construct the current state vector based on the second indicator parameter value and the corresponding deviation.

[0081] For example, calculate the degree to which the current flight operation deviates from the pilot's personal habits: d =(current value) The deviation is calculated as (personal historical mean) / (personal historical standard deviation). For example, if a pilot's habitual throttle reduction altitude is 350 ft with a standard deviation of 20 ft, and their current throttle reduction altitude is 400 ft, then the deviation = (400-350) / 20 = 2.5 (indicating a deviation of 2.5 standard deviations, which is considered significant). Subsequently, a current state vector for the pilot being evaluated is constructed (containing the second indicator parameter value and its corresponding deviation).

[0082] Step S540: Input the current state vector into the personal flight habit model to determine the overall stability score.

[0083] For example, a current state vector of the pilot to be evaluated (including the second indicator parameter value and the corresponding deviation) is constructed, and their personal flight habit model is input. The Q value corresponding to the current state vector is queried to obtain the overall stability score. Based on the deviation and the personal habit model, the stability score is obtained to determine whether the current flight operation conforms to the pilot's stable operation mode.

[0084] Step S550: Determine the flight quality assessment result for the pilot to be evaluated based on the overall accuracy score and the overall stability score.

[0085] For example, the final evaluation result is obtained by combining the "performance under group standards" (i.e., the overall accuracy score) and the "stability under individual habits" (i.e., the overall stability score).

[0086] As a preferred implementation, in practical applications, the current QAR data may contain missing values ​​for some indicator parameters. To ensure the smooth progress of the evaluation process, a data null value self-derivation method can be used to fill in the missing indicator parameter values ​​before constructing the current state vector. Specific completion methods include: For example, prediction based on relevant indicators: using the parameter values ​​of other complete indicators for the same pilot on the current flight, the parameter values ​​of the missing indicator are predicted by a pre-trained sub-model.

[0087] Alternatively, reverse derivation based on a personal flight habit model: input the currently known index parameter values ​​into a trained personal flight habit model, and reverse derive the missing index parameter values ​​that best match the pilot's personal habits.

[0088] Alternatively, fill-in can be based on similar pilots: using pilot group clustering information, select corresponding indicator parameter values ​​from similar pilots with similar operating styles to fill in the gaps.

[0089] The above-mentioned completion method improves the applicability and robustness of the evaluation method in practical applications by completing missing values ​​during the data preprocessing stage.

[0090] Similarly, when constructing the state vectors of historical flights to train an individual flight habit model, if some indicator parameter values ​​are missing in the historical QAR data, the above-mentioned data null value self-derivation method can be used to fill in the missing values, so as to ensure the integrity of the training data and improve the robustness of model training.

[0091] As a preferred implementation method, after obtaining the flight quality assessment results of the pilot to be evaluated, the individual flight habit model can be further applied to pilot management and training, specifically including: For example, personalized assessment and real-time alerts: Utilizing a personal flight habit model, the deviation between the pilot's current operation and their historical habits is calculated in real time. When the deviation exceeds a preset threshold, an alert is automatically triggered, reminding the pilot or management personnel to pay attention to potential fluctuations in technical condition.

[0092] Alternatively, personalized training programs can be pushed out: based on the individual flight habit model, the pilot's operating characteristics or potential risk patterns (such as persistent deviations in a certain type of approach attitude) can be identified, and combined with their technical weaknesses, targeted training reminders or recommended training subjects can be automatically generated and pushed to the pilot or training department to achieve precise and personalized skills improvement.

[0093] The above application does not change the flight technology quality assessment method mentioned above, but only makes subsequent use of the assessment results and models, giving full play to the industrial value of this method in flight safety management and personnel training.

[0094] In some embodiments, a screening index is also provided (corresponding to step S110 above: screening parameter types from historical QAR data based on the maximum correlation and minimum redundancy criterion, as an index for evaluating flight quality). Figure 6 This is a flowchart illustrating a screening index provided in an embodiment of this application, as shown below. Figure 6 As shown, it includes: Step S610: Obtain the parameter values ​​of each candidate parameter and the corresponding flight quality label from the historical QAR data.

[0095] For example, the embodiments of this application do not impose special limitations on the source of flight quality labels. Flight quality labels may be: expert ratings of historical flights (such as excellent, good, qualified, unqualified), binary labels based on post-event safety event marking (normal / abnormal), comprehensive scores based on continuous safety indicators, etc.

[0096] It should be noted that, in this embodiment, the candidate parameters can correspond to the original flight parameters obtained from the flight data recording system, which are a set of parameters used to characterize the aircraft's operating state or flight control state during flight, such as altitude, speed, attitude angle, engine parameters, or control input parameters. For example, candidate parameters can be filtered using a feature selection algorithm (e.g., the minimum redundancy maximum correlation mRMR algorithm) to obtain indicators for flight technical quality assessment.

[0097] In this embodiment of the application, the index parameter value can correspond to the actual value or statistical value of the corresponding index during the flight of a specific flight, which is used to characterize the flight performance of the flight under the index dimension.

[0098] In this embodiment, flight quality labels can be used to characterize the flight technical quality level of historical flights. Flight quality labels can be determined based on historical flight quality assessment results, such as safety event levels recorded by the Flight Operations Quality Assurance (FOQA) system, flight technical quality assessment results, or manually labeled by flight technical experts based on historical flight data. Flight quality labels can be expressed in a graded or categorized form, such as Excellent, Good, Average, or Needs Improvement.

[0099] In this embodiment, the actions in the reinforcement learning model can represent the flight control strategies or control adjustments taken by the pilot in the current flight state. For example, given a flight state vector, an action can be represented as an adjustment method to flight control parameters, including throttle adjustment, attitude adjustment, speed control strategy, or other flight control decisions. The reinforcement learning model can learn the relationship between actions and flight quality results under different states, thereby obtaining a flight technical quality assessment model.

[0100] Step S620: Calculate the correlation between each candidate parameter and the flight quality label based on the Pearson correlation coefficient.

[0101] For example, when calculating the correlation between each candidate parameter and the flight quality label, the Pearson correlation coefficient can be used to measure the linear correlation. The Pearson correlation coefficient measures the degree of linear correlation, with a value range of [-1, 1]. The larger the absolute value, the stronger the linear correlation, and it can only capture linear relationships.

[0102] The Pearson correlation coefficient can be calculated using the following formula: (3); Where r represents the Pearson correlation coefficient, and They are variables and variables The i-th observation sample, and They are variables and variables The mean, This represents the number of observations in the sample. If used to calculate correlation, the variable... and variables These can represent candidate parameters and flight quality labels, respectively.

[0103] Step S630: Calculate the redundancy between any two candidate parameters based on the information gain of the Pearson correlation coefficient and / or information entropy.

[0104] For example, when calculating redundancy between candidate parameters, the Pearson correlation coefficient can be used to measure linear correlation, while information gain can be used to measure more generalized statistical dependencies (including nonlinear relationships). Combining both can improve the accuracy and robustness of indicator selection. Information gain measures the reduction in uncertainty about one feature after knowing another, with a value ranging from [0, ∞). A larger value indicates a stronger dependency and can capture nonlinear relationships and dependencies of any type. In this embodiment, information gain is used to measure the predictive ability of one candidate parameter for another; a larger information gain indicates higher redundancy.

[0105] In step S630, the variables involved in the Pearson correlation coefficient and variables These can represent two different candidate parameters. For candidate parameters... and First, calculate the information entropy based on its value distribution. and Then calculate the conditional entropy. Information gain Indicates knowledge back The degree to which uncertainty is reduced. The larger, the more it means right The stronger the predictive ability, the higher the redundancy of both.

[0106] Information entropy can be calculated using the following formula: (4); The logarithmic base is usually 2, and the unit is bits. Indicates statistical expectation. for Take the first The probability of each value.

[0107] Conditional entropy can be calculated using the following formula: (5); (6); in, For a given hour The conditional probability, for Take the first The probability of each value.

[0108] When combining Pearson correlation coefficient and information gain to calculate redundancy, a weighted fusion strategy can be adopted: Redundancy ( , )=α ∣r( , )∣+β Normalization (IG) ; )),in These are weighting coefficients, which can be adjusted according to the application scenario.

[0109] Step S640: Based on the maximum correlation and minimum redundancy criterion, select indicators for evaluating flight quality according to correlation and redundancy.

[0110] In some embodiments, another screening criterion is also provided (corresponding to step S640 above: based on the maximum correlation and minimum redundancy criterion, screening criters for evaluating flight quality according to correlation and redundancy). Figure 7 This is a flowchart illustrating yet another screening metric provided in an embodiment of this application, as shown below. Figure 7 As shown, it includes: Step S710: Initialize the selected indicator set to empty.

[0111] Step S720: Determine the target candidate parameter that has the greatest correlation with the flight quality label from the candidate parameters.

[0112] Step S730: Generate a set of selected target indicators based on the target candidate parameters.

[0113] Step S740: Iteratively execute the following steps until the preset stopping condition is met, then use the final selected indicators in the target indicator set as indicators for evaluating flight quality: Step S741: Calculate the comprehensive score based on the correlation between the remaining candidate parameters of the target and the flight quality label, as well as the redundancy between them and the indicators in the selected indicator set of the target; wherein, the remaining candidate parameters of the target correspond to the remaining candidate parameters of the target that were not selected in the candidate parameters.

[0114] Step S742: Select the remaining candidate parameters of the target with the highest comprehensive score and add them to the target selected index set.

[0115] In some embodiments, the types of indicators include habitual indicators, absolute indicators, and Boolean indicators; This application embodiment also provides a standard threshold for determining each indicator (corresponding to step S130 above: determining the standard threshold for each indicator based on historical target flight data). Figure 8 This is a flowchart illustrating a process for determining standard thresholds for various indicators, as provided in an embodiment of this application. Figure 8 As shown, it includes: Step S810: Statistically analyze the numerical distribution of each indicator in the historical target flight data.

[0116] Step S820: For habitual indicators, based on the preset excellent coverage ratio, determine the values ​​at the corresponding lower quantile and higher quantile, which are respectively used as the first quality threshold and the second quality threshold; wherein, the first quality threshold is lower than the second quality threshold.

[0117] For example, for habitual indicators (such as throttle drop height), the values ​​of this indicator are collected from historical target flight data, and their distribution is statistically analyzed. Assuming the preset excellent coverage ratio is 80% (i.e., it is desired that 80% of the target flights are rated at least "good" or "excellent"), the specific values ​​corresponding to the 15th percentile and 85th percentile (such as 330ft and 380ft) are calculated respectively, serving as the first quality threshold and the second quality threshold. Flights below 330ft or above 380ft are rated "poor," while those in between are rated "good."

[0118] In some feasible implementations, the first quality threshold corresponds to the low quality threshold; the second quality threshold corresponds to the high quality threshold.

[0119] Step S830: For absolute indicators, determine a first warning boundary and a second warning boundary composed of multiple standard deviations of the mean, based on a preset optimal value point and a preset warning multiple; wherein the first warning boundary is greater than the second warning boundary.

[0120] For example, for absolute indicators, the numerical distribution of the indicator in historical target flights is statistically analyzed, and its mean and standard deviation are calculated. Based on the nature of the indicator, an optimal value point (usually 0 or an extreme value, such as 0G for ground load and 0 for heading deviation) and a warning multiple k (such as 6 times the standard deviation) are preset. For indicators where "the smaller the better" (such as ground load), the warning boundary is the mean + k times the standard deviation, serving as the first warning boundary; for indicators where "the larger the better" (such as brake pressure), the warning boundary is the mean + k times the standard deviation. k times the standard deviation is used as the second warning boundary; for indicators that require stability within a certain range (such as heading maintenance), it may be necessary to use ±k times the standard deviation of the mean as the first / second warning boundaries. These boundary values, together with the optimal point, constitute the standard threshold of the indicator.

[0121] Step S840: For Boolean indicators, assign 0 and 1 to events that did not occur and events that occurred, respectively, as standard thresholds.

[0122] For example, for Boolean indicators (such as low-altitude go-arounds), the value itself defines whether the event has occurred: 0 indicates that it has not occurred, and 1 indicates that it has occurred. Therefore, the standard threshold is {0,1}, which can be determined without statistical analysis.

[0123] In some embodiments, determining the weight coefficients of each indicator based on historical QAR data includes: determining the weight coefficients of each indicator based on historical QAR data and subjective / objective weighting methods.

[0124] Objective weighting relies entirely on the statistical characteristics of historical QAR data itself, and the objective weight of indicators is comprehensively measured from the following two dimensions: Indicator variability: Measured by standard deviation. The greater the fluctuation of an indicator's value in historical flights (i.e., the larger the standard deviation), the greater the amount of information it carries, the stronger its ability to distinguish between different flight qualities, and the higher its weighting coefficient should be assigned.

[0125] Indicator conflict: Measured by correlation coefficient. If two indicators are highly correlated, it indicates that they reflect overlapping information. To avoid redundancy, their weights should be appropriately reduced. Specifically, the stronger the overall correlation between an indicator and other indicators, the lower its weight coefficient.

[0126] The above-mentioned subjective and objective weighting methods, by comprehensively considering the variability (information content) and conflict (redundancy) of the indicators, can use objective weighting methods such as the CRITIC method to calculate the objective weight coefficient of each indicator.

[0127] To incorporate the knowledge and experience of domain experts into subjective weighting, the Analytic Hierarchy Process (AHP) can be used. Experts compare each indicator pairwise to quantify its importance, construct a judgment matrix, and then calculate the subjective weight coefficient for each indicator. Subjective weighting can reflect prior information such as safety management strategies and regulatory requirements that cannot be fully obtained from data.

[0128] To balance the objective laws of the data with the experiential judgment of experts, the aforementioned objective weights and subjective weights can be combined to obtain the final comprehensive weight coefficient. The combination method can employ multiplicative synthesis or linear weighting, followed by normalization to ensure that the sum of the weight coefficients of all indicators equals 1. The combined weight coefficient reflects both the informational characteristics of the data itself and the experiential knowledge of domain experts, thus possessing higher scientific validity and rationality.

[0129] As a second aspect of this application, a flight technology quality assessment system is provided. Figure 9 This is a structural schematic diagram of a flight technology quality assessment system 200 provided in an embodiment of this application, as shown below. Figure 9 As shown, system 200 includes: The filtering module 201 is used to filter parameter types from historical QAR data based on the maximum relevance and minimum redundancy criterion, as indicators for evaluating flight quality. The first determining module 202 is used to determine the weight coefficient of each indicator based on the historical QAR data; The second determining module 203 is used to determine the standard threshold of each of the indicators based on historical target flight data; Training module 204 is used to train and generate a personal flight habit model corresponding to the pilot to be evaluated based on the first indicator parameter value of the pilot to be evaluated in the historical QAR data, the weight coefficient and the standard threshold. The third determining module 205 is used to determine the flight quality assessment result of the pilot to be assessed based on the second indicator parameter value of the pilot to be assessed in the current QAR data, the personal flight habit model corresponding to the pilot to be assessed, the weighting coefficient and the standard threshold.

[0130] The flight quality assessment system provided in this application includes a screening module that selects parameter types from historical QAR data based on the maximum relevance and minimum redundancy criterion, which are used as indicators for assessing flight quality; a first determining module that determines the weight coefficients of each indicator based on historical QAR data; a second determining module that determines the standard thresholds of each indicator based on historical target flight data; a training module that trains and generates a personal flight habit model corresponding to the pilot to be assessed based on the first indicator parameter value, weight coefficient, and standard threshold of the pilot to be assessed in historical QAR data; and a third determining module that determines the flight quality assessment result corresponding to the pilot to be assessed based on the second indicator parameter value of the pilot to be assessed in the current QAR data, the personal flight habit model corresponding to the pilot to be assessed, as well as the weight coefficient and standard threshold. Historical QAR data is used to drive indicator selection and weight coefficient calculation to improve the objectivity of the evaluation system. By training individual flight habit models and obtaining group standard thresholds from historical target flight data, the evaluation combines individual characteristics with group commonalities to balance the commonality of evaluation standards with individual adaptability. Training individual flight habit models using historical QAR data before evaluating current QAR data (single flight data) significantly improves analytical granularity. Using the indicator parameter values ​​of the pilot being evaluated in historical QAR data to train the pilot's individual flight habit model allows learning of stable patterns formed during long-term flight, identifying individual dynamic evolution and habitual patterns. By binding the indicator system, weights, standard thresholds, and individual flight habit models to updatable historical data, all elements naturally possess the ability to adapt to environmental changes rather than being fixed. Therefore, when historical data is updated, the above process can be re-executed, allowing all elements (indicators, weight coefficients, standard thresholds, individual flight habit models, etc.) to automatically adapt to the latest "historical data," enhancing adaptability and continuous evolution.

[0131] As a third aspect of the embodiments of this application, an electronic device is also provided. Figure 10 This is a structural schematic diagram of an electronic device provided in an embodiment of this application, such as... Figure 10 As shown, it includes: Electronic devices include: One or more processors 101; The memory 102 stores one or more computer programs that, when executed by the one or more processors 101, cause the one or more processors 101 to implement the file-based data storage method provided in the first aspect of the embodiments of this application.

[0132] The electronic device may also include one or more I / O interfaces 103 connected between the processor 101 and the memory 102, configured to enable information interaction between the processor 101 and the memory 102.

[0133] The processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read-write interface) is connected between the processor and the memory, enabling information exchange between the processor and the memory, including but not limited to a data bus (Bus).

[0134] In some embodiments, the processor 101, memory 102, and I / O interface 103 are interconnected via bus 104, and thus connected to other components of the computing device.

[0135] Meanwhile, embodiments of this application also provide a computer-readable medium, Figure 11 This is a structural schematic diagram of a computer-readable medium provided in an embodiment of this application, on which a computer program is stored. When the computer program is executed by a processor, it implements the flight technical quality assessment method provided in the first aspect of this application.

[0136] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. Accordingly, the computer program can be stored in a non-volatile computer-readable storage medium, and when executed, the computer program can implement the methods of any of the above embodiments. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0137] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Those skilled in the art should understand that this application includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments above. Any modifications that do not depart from the functional and structural principles of this application will be included within the scope of the claims.

Claims

1. A method for evaluating the quality of flight technology, characterized in that, The flight technology quality assessment methods include: Based on the maximum relevance and minimum redundancy criterion, parameter types are selected from historical QAR data as indicators for evaluating flight quality. Based on the historical QAR data, determine the weighting coefficients for each of the aforementioned indicators; Based on historical target flight data, standard thresholds for each of the aforementioned indicators are determined. Based on the first indicator parameter value of the pilot to be evaluated in the historical QAR data, the weighting coefficient and the standard threshold, a personal flight habit model corresponding to the pilot to be evaluated is trained and generated. Based on the second indicator parameter value of the pilot to be evaluated in the current QAR data, the personal flight habit model of the pilot to be evaluated, the weighting coefficient, and the standard threshold, the flight quality assessment result of the pilot to be evaluated is determined.

2. The flight technology quality assessment method according to claim 1, characterized in that, The step of training and generating a personal flight habit model for the pilot to be evaluated based on the first indicator parameter value of the pilot to be evaluated in the historical QAR data, the weighting coefficient, and the standard threshold includes: Construct a reward function based on the weight coefficients and standard thresholds described above; Based on the first indicator parameter value, a state vector is constructed, and the reward value corresponding to each state vector is calculated according to the reward function. The cumulative reward value is maximized as the optimization objective, and the personal flight habit model is generated by training based on the reinforcement learning algorithm.

3. The flight technology quality assessment method according to claim 2, characterized in that, The process of constructing a state vector based on the first indicator parameter value, calculating the reward value corresponding to each state vector according to the reward function, and maximizing the cumulative reward value as the optimization objective, and training the personal flight habit model based on a reinforcement learning algorithm includes: Based on each flight flown by the pilot to be evaluated in the historical QAR data, the first indicator parameter value for that flight is obtained, and the deviation of the obtained first indicator parameter value from the baseline of the historical flight indicator parameter value of the pilot to be evaluated is determined; wherein, the state vector includes the first indicator parameter value and the corresponding deviation. Based on the actual flight data of the pilot to be evaluated in each flight, the actions taken by the pilot to be evaluated in each flight under the state vector are inferred posteriorly. Based on the reward function, calculate the reward value obtained by the pilot to be evaluated in each flight after performing the action; Using the state vector, the action, and the reward value as samples, the Q-Learning algorithm is used to iteratively update the data to maximize the cumulative reward value, thereby determining the Q-value function of the pilot to be evaluated. The personal flight habit model is trained and generated based on the Q-value function.

4. The flight technology quality assessment method according to claim 2, characterized in that, The step of constructing a reward function based on each of the weight coefficients and each of the standard thresholds includes: Based on the standard threshold values ​​of each indicator parameter, the corresponding numerical value of the individual safety score for each indicator is determined by mapping. A security compliance reward component is constructed by weighting and aggregating the weighting coefficients and individual security scores as described above. Calculate the green economy reward based on the preset green economy indicators and their corresponding weights; The reward function is determined by weighting and combining the security compliance reward component and the green economy reward component.

5. The flight technology quality assessment method according to claim 1, characterized in that, The step of determining the flight quality assessment result for the pilot under evaluation based on the second indicator parameter value of the pilot under evaluation in the current QAR data, the pilot under evaluation's corresponding personal flight habit model, the weighting coefficient, and the standard threshold includes: Based on the aforementioned standard thresholds, the values ​​of the second indicator parameters of the pilot to be evaluated in the current QAR data are mapped to the individual accuracy scores of each indicator. Based on the weighting coefficients, the individual accuracy scores are weighted and aggregated to determine the overall accuracy score; Determine the deviation of the second indicator parameter value from the baseline of the historical flight indicator parameter value of the pilot to be evaluated, and construct the current state vector based on the second indicator parameter value and the corresponding deviation. The current state vector is input into the personal flight habit model to determine the overall stability score; Based on the overall accuracy score and the overall stability score, the flight quality assessment result corresponding to the pilot to be evaluated is determined.

6. The flight technology quality assessment method according to claim 1, characterized in that, The selection of parameter types from historical QAR data based on the maximum correlation and minimum redundancy criterion, used as indicators for evaluating flight quality, includes: Obtain the parameter values ​​of each candidate parameter and the corresponding flight quality label from the historical QAR data; Based on the Pearson correlation coefficient, the correlation between each candidate parameter and the flight quality label is calculated; Based on the information gain of Pearson correlation coefficient and / or information entropy, calculate the redundancy between any two candidate parameters; Based on the maximum correlation and minimum redundancy criterion, the indicators used to evaluate flight quality are selected according to the correlation and redundancy.

7. The flight technology quality assessment method according to claim 6, characterized in that, The method of selecting indicators for evaluating flight quality based on the maximum correlation and minimum redundancy criterion, according to the correlation and redundancy, includes: Initialize the selected indicator set to empty; From the candidate parameters, determine the target candidate parameter that has the highest correlation with the flight quality label; Based on the target candidate parameters, generate a set of selected target indicators; The following steps are executed iteratively until a preset stopping condition is met, at which point the final set of selected target indicators is used as the indicators for evaluating flight quality: A comprehensive score is calculated based on the correlation between the remaining candidate parameters of the target and the flight quality label, as well as the redundancy between them and the indicators in the selected indicator set of the target; wherein, the remaining candidate parameters of the target correspond to the remaining candidate parameters of the target that were not selected in the candidate parameters. The remaining candidate parameters of the target with the highest comprehensive score are selected and added to the target's selected index set.

8. The flight technology quality assessment method according to claim 1, characterized in that, The types of indicators include habitual indicators, absolute indicators, and Boolean indicators; The determination of standard thresholds for each indicator based on historical target flight data includes: The numerical distribution of each of the aforementioned indicators in the historical target flight data was statistically analyzed. For habitual indicators, based on the preset excellent coverage ratio, the values ​​at the corresponding lower quantile and higher quantile are determined and used as the first quality threshold and the second quality threshold, respectively; wherein, the first quality threshold is lower than the second quality threshold. For absolute indicators, based on the preset optimal value point and the preset warning multiple, a first warning boundary and a second warning boundary are determined, which are composed of multiple standard deviations of the mean. One or both of them are selected as standard thresholds according to the physical meaning of the indicator. The first warning boundary is greater than the second warning boundary. For Boolean indicators, 0 and 1 are used as the standard thresholds for events that have not occurred and events that have occurred, respectively.

9. A flight technology quality assessment system, characterized in that, include: The filtering module is used to filter parameter types from historical QAR data based on the maximum relevance and minimum redundancy criterion, as indicators for evaluating flight quality. The first determining module is used to determine the weight coefficient of each indicator based on the historical QAR data; The second determining module is used to determine the standard threshold of each of the indicators based on historical target flight data; The training module is used to train and generate a personal flight habit model corresponding to the pilot to be evaluated based on the first indicator parameter value of the pilot to be evaluated in the historical QAR data, the weight coefficient, and the standard threshold. The third determining module is used to determine the flight quality assessment result of the pilot to be assessed based on the second indicator parameter value of the pilot to be assessed in the current QAR data, the personal flight habit model corresponding to the pilot to be assessed, the weighting coefficient and the standard threshold.

10. An electronic device, characterized in that, include: One or more processors; A memory having stored one or more computer programs that, when executed by one or more processors, cause the one or more processors to implement the flight technical quality assessment method according to any one of claims 1 to 8.

Images