CN-121981614-A - Pilot competence assessment and event attribution system based on QAR space-time characteristics

CN121981614ACN 121981614 ACN121981614 ACN 121981614ACN-121981614-A

Abstract

The invention belongs to the technical field of aviation safety and pilot capability assessment, relates to a pilot competence assessment and event attribution system based on QAR space-time characteristics, and solves the problem that the prior art cannot accurately assess and attribution. The invention decodes and normalizes the archived flight operation basic data through a decoding and normalizing preprocessing module, a flight stage and event detection module adopts a combination rule, decision tree verification and multi-parameter coupling physical model to sense a flight state and a complex event, a competence feature extraction and quantization scoring module generates accurate scores through each dimension exclusive physical quantization model and a combination weighting algorithm, a clustering portraits and anomaly attribution module realizes dynamic portraits and interpretable attribution through self-adaptive clustering, POT dynamic threshold detection and SHAP value algorithm, and an intelligent disc copying and training recommendation module fuses multi-source data to generate an evidence chain to output an EBT training scheme. The invention improves the objectivity of evaluation, attribution precision and training pertinence.

Inventors

ZENG RUIQI
LI YUFEI
FAN YANJIE
ZHOU XIN
Lv Keyuan
LI JIAXIN
YAO BINGYU
YANG LEI
YANG SHI
SU JIANFEI
WANG ZHONGXING
WANG ZHIFENG
ZHANG ZIHONG
LIU WEIREN
LI SISI

Assignees

南航科技(广东横琴)有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A pilot competency assessment and event attribution system based on QAR space-time features, the system comprising: The data access and archiving module is configured to acquire flight operation basic data and obtain archiving data through hierarchical archiving; the decoding and standardization preprocessing module is configured to decode and standardize the archived data to obtain a standardization parameter matrix; The flight phase and event detection module is used for carrying out parallel processing on the standardized parameter matrix through double paths, wherein one path is used for generating a flight phase index through combination rule judgment, decision tree verification and transition window optimization, the other path is used for detecting complex events through a multi-parameter coupling physical model, window boundaries are defined through a double-threshold hysteresis algorithm, and the severity is dynamically calculated by combining peak overrun, overrun energy integration and environmental risk coefficient, so that a structured event object set is generated; The competence feature extraction and quantification scoring module is used for extracting interpretable derivative features corresponding to a plurality of core competence dimensions based on the stage index and the structured event object set, inputting the interpretable derivative features into a pre-constructed multi-dimensional competence quantification model for weighted fusion and quantification mapping, and generating competence scores corresponding to the core competence dimensions; The clustering portrait and anomaly attribution module is used for self-adaptively determining the clustering quantity through an elbow rule and a contour coefficient based on the competence score and the interpretable derivative features, further generating an initial pilot operation style portrait with a business semantic label, fitting tail distribution to the interpretable derivative features through a POT algorithm of an extremum theory to obtain anomaly scores, calculating a dynamic threshold value based on the anomaly scores through a pollution rate self-adaptive adjustment mechanism, further identifying anomalies, and calculating marginal contribution values of the interpretable derivative features to the anomaly scores through a SHAP value algorithm to obtain anomaly attribution results.
2. The pilot competency assessment and event attribution system based on QAR spatiotemporal features of claim 1, further comprising an intelligent compound and training recommendation module configured to integrate the standardized parameter matrix, the structured event object set, the competency score, the pilot operating style portraits, and the anomaly attribution results to generate a multi-dimensional auditable evidence chain comprising a multi-parameter time series compound graph, a competency radar contrast graph, a clustered PCA dimension reduction graph, and an anomaly root cause report; Based on the short plate dimension, the event type and the quantization index in the multi-dimension auditable evidence chain, a training recommendation index is calculated by combining a ternary mapping matrix formed by the competence short plate, the event type and the training subjects, and the multi-objective knapsack algorithm is used for optimizing the lesson time distribution, so that a personalized scheme adapting to the EBT training is obtained.
3. The QAR spatiotemporal feature-based pilot competence assessment and event attribution system of claim 1, wherein said normalization process comprises parameter alias alignment, discrete mixed bit splitting, heterogeneous sample rate unification, parameter calibration, and data cleansing; The parameter alias alignment adopts a combination mode of static mapping table and semi-automatic matching confirmation; the discrete mixed bit splitting is achieved by a masking and shifting operation.
4. The QAR spatiotemporal feature-based pilot competence assessment and event attribution system of claim 1, wherein said multiparameter coupled physics model comprises a wind shear detection model and an unsteady approach detection model; The wind shear detection model builds quantization factor calculation based on a longitudinal wind shear component, a vertical wind speed and a vacuum speed, and triggers a wind shear event when the radio altitude is lower than a preset altitude threshold and the quantization factor exceeds a preset range for a preset period of time; The unstable approach detection model is used for judging through combination of total energy errors and flap in-place states, the total energy errors are absolute values of current total energy and standard glidepath energy, and the current total energy is calculated through the sum of conversion values of the height and the square of the speed.
5. The QAR spatiotemporal feature-based pilot competence assessment and event attribution system of claim 1, wherein said combining peak overrun, overrun energy integration and environmental risk factor dynamically calculates severity, said severity The calculation formula is as follows: ; Wherein, the As a peak value of the parameter, A threshold value is preset for the parameter, Alpha and beta are fixed weight coefficients for environmental risk coefficients, As a result of the dynamic weight coefficient, As a value of the parameter in real time, As the termination time of the event, Is the trigger time of the event.
6. The pilot competency assessment and event attribution system based on QAR spatiotemporal features of claim 1, wherein said pre-built multidimensional competency model is constructed by: Respectively constructing corresponding feature-score mapping sub-models for a plurality of core competence dimensions; Determining a combination weight for the interpretable derivative features processed by each sub-model by adopting a combination weighting mode combining an AHP analytic hierarchy process and an entropy weighting method; And configuring each sub-model based on the combination weight so as to carry out weighted fusion on the input interpretable derivative features, and setting a threshold value of the quantization mapping according to quantiles obtained by statistics from the fleet historical data.
7. The QAR spatiotemporal feature-based pilot competence assessment and event attribution system of claim 6, wherein said plurality of core competence dimensions comprises flight path management, operational synergy, situational awareness judgment, emergency handling response, and automated equipment management; And (3) supplementing and evaluating competence dimensions which cannot be directly observed by the QAR by adopting a layered completion strategy based on the similar pilot capacity baselines and event association relations.
8. The pilot competence assessment and event attribution system based on QAR spatiotemporal features of claim 1, wherein said determining a number of clusters is self-adaptive with a contour coefficient by an elbow law, said elbow law being based on SSE curve inflection points, the contour coefficient being based on a maximum average contour value; the business semantic tags are generated based on the standard deviation of cluster centroids and the global mean of the fleet; the business semantic tags comprise high-load rough types, automatic dependent types and accurate stable types.
9. The QAR spatiotemporal feature-based pilot competence assessment and event attribution system of claim 1, wherein said specific step of fitting a tail distribution to said interpretable derivative feature by a POT algorithm employing extremum theory is: fitting tail distribution to the interpretable derivative features by using a POT algorithm of an extremum theory, and calculating to obtain an abnormal score value; the abnormal scoring values are subjected to descending order sequencing, and then the difference value of adjacent scoring is calculated, so that a differential sequence of abnormal scoring is obtained; Respectively identifying a maximum value point of the differential sequence and a second derivative peak value point of a curve corresponding to an abnormal score value, and taking the maximum value point and the second derivative peak value point as elbow points to further obtain the number of the elbow points; Calculating the ratio of the number of elbow points to the total number of the interpretable derivative features to obtain a dynamic pollution rate adapted to the current interpretable derivative features; Based on the dynamic contamination rate, a dynamic threshold is determined from the anomaly score values in descending order, and interpretable derivative features having scores exceeding the dynamic threshold are determined to be anomaly.
10. The QAR spatiotemporal feature-based pilot competency assessment and event attribution system of claim 1, further comprising verifying and updating an initial pilot operating style representation based on said anomaly attribution results by: when the abnormal core contribution characteristic calculated by the SHAP value algorithm is highly matched with the business semantic tag of the initial pilot operation style portrait, judging that the abnormal attribution result is a high confidence result, and confirming the initial pilot operation style portrait as a final portrait; When the core contribution features are not matched with the portrait tag, secondary screening is carried out on the interpretable derivative features, the SHAP value marginal contribution is recalculated by combining the time sequence association relation of the scene features and the operation behaviors in the flight stage, the abnormal attribution result is corrected, and the corrected attribution conclusion is reversely updated into the operation style portrait tag of the pilot to obtain a final pilot operation style portrait, so that bidirectional promotion of portrait iteration and attribution accuracy is realized.

Description

Pilot competence assessment and event attribution system based on QAR space-time characteristics Technical Field The invention belongs to the technical field of aviation safety and pilot capability assessment, and particularly relates to a pilot competence assessment and event attribution system based on QAR space-time characteristics. Background In the aviation operation safety management, the competence level of pilots directly relates to the aviation safety, and the accurate attribution of aviation unsafe events is the core of optimizing a training scheme and improving the operation safety. The prior art has various limitations, and is difficult to meet the requirements of fine granularity capability assessment and efficient attribution: The traditional QAR/FOQA overrun event monitoring scheme takes a preset threshold value and a logic rule base as a core, and the overrun event is identified and a list is generated by scanning QAR data, so that manual copying is relied on. The method has the defects of sensitivity to parameter noise, model and environmental change, need to adjust parameters manually in a large amount, can only capture the overt risk of 'overrun', and is missing the problem of the process type, and has high false alarm rate, high manual rechecking cost, low closed-loop efficiency and incapability of supporting fine granularity capability diagnosis; The manual competence scoring scheme is characterized in that scoring is carried out according to a subjective scale of an instructor in an EBT system, so that the problems of strong subjectivity and insufficient reproducibility exist; An unsupervised clustering/anomaly detection scheme, wherein part of the scheme extracts QAR data statistical characteristics, combines KMeans and other algorithms to cluster or detect anomalies, but only outputs clustering labels or anomaly scores, and lacks association mapping with nine competence; The QAR data has the problems of different parameter aliases, complex analysis of discrete mixing amount, heterogeneous sampling rate and the like in the data processing and quantization model, and the lack of a unified standardization mechanism in the prior art leads to poor consistency of cross-model/batch data processing and insufficient reliability of an evaluation result. Meanwhile, the existing scheme adopts a general weighting model to quantify the winner Ren Li, does not design exclusive logic aiming at physical properties of each dimension, has larger deviation between scoring and actual capability, and cannot accurately describe the multidimensional real level of the pilot. Therefore, a technical scheme is needed that can deeply mine the space-time characteristics of the QAR, combine each competence dimension exclusive quantization model to realize accurate evaluation, and can automatically and interpretably ascribe unsafe events so as to fill the blank of the prior art. Disclosure of Invention In order to solve the above problems in the prior art, that is, the lack of standardization of QAR data processing, lack of proprietary quantization model for competence assessment, poor automation and interpretability of unsafe event attribution, and poor connection between assessment result and training decision, which further causes technical problems of inaccurate assessment and attribution, the invention provides a pilot competence assessment and event attribution system based on QAR space-time characteristics, which comprises: The data access and archiving module is configured to acquire flight operation basic data and obtain archiving data through hierarchical archiving; the decoding and standardization preprocessing module is configured to decode and standardize the archived data to obtain a standardization parameter matrix; The flight phase and event detection module is used for carrying out parallel processing on the standardized parameter matrix through double paths, wherein one path is used for generating a flight phase index through combination rule judgment, decision tree verification and transition window optimization, the other path is used for detecting complex events through a multi-parameter coupling physical model, window boundaries are defined through a double-threshold hysteresis algorithm, and the severity is dynamically calculated by combining peak overrun, overrun energy integration and environmental risk coefficient, so that a structured event object set is generated; The competence feature extraction and quantification scoring module is used for extracting interpretable derivative features corresponding to a plurality of core competence dimensions based on the stage index and the structured event object set, inputting the interpretable derivative features into a pre-constructed multi-dimensional competence quantification model for weighted fusion and quantification mapping, and generating competence scores corresponding to the core competence dimensions; The clustering portrait and anomaly