Multi-dimensional based flight skill assessment and personalized flight skill path optimization method

Abstract

The present application relates to aerospace management technical field, especially for based on multi-dimensional flight skill evaluation and individualized flight skill path optimization method, its method includes the following steps: step S1, data acquisition and pretreatment;Step S2, multi-dimensional flight skill evaluation model training;Step S3, individualized flight skill path planning;Step S4, path optimization and iteration;Step S5, real-time monitoring and feedback adjustment;Step S6, knowledge base update and model optimization;The present application is through system identification method and parameter identification algorithm, constructs the key skill index model covering ground preparation, approach, cruise and other multi-flight stages, quantifies the error of pilot operation and standard model, realizes dynamic multi-dimensional evaluation, overcomes the problem of traditional system relying on single dimension or static index, comprehensively reflects the skill level of pilot in spatial orientation, procedure execution and other dimensions, integrates improved ant colony algorithm and multi-objective genetic algorithm, dynamically adjusts the evaporation coefficient and path cost function.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of aerospace management, in particular to a flight skill evaluation method based on multiple dimensions and a personalized flight skill path optimization method. BACKGROUND

[0002] The flight skill evaluation and personalized path optimization technology has formed a multi-dimensional evaluation system and an intelligent optimization framework. At the evaluation level, the industry uses a flight simulator to implement full-dynamic scene testing, reproduces extreme weather, mechanical failure and other sudden conditions through a high-precision motion platform and a visual system, and captures physiological indicators such as the stress response and decision speed of pilots through physiological monitoring equipment to form a quantitative evaluation report.

[0003] However, the existing system lacks dynamic modeling of the multi-dimensional skill indicators of pilots, and it is difficult to comprehensively reflect the skill level of pilots. The existing path planning method mainly relies on the experience of instructors and lacks a data-driven dynamic optimization mechanism, making it difficult to adapt to individual differences of pilots and dynamic environments. The application of digital twin technology in aviation training is mainly concentrated in equipment maintenance and has not been deeply integrated with pilot skill evaluation and path optimization. At the same time, NLP technology has been used for text classification, information extraction, etc., but in the update of aviation knowledge base, the existing method lacks calculation of term similarity and automatic classification of knowledge items, and relies on manual update with low efficiency. SUMMARY

[0004] The purpose of the present application is to provide a flight skill evaluation method based on multiple dimensions and a personalized flight skill path optimization method, which solves the problems of single evaluation dimension, static path planning, insufficient algorithm performance and lagging knowledge base of traditional technology.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a flight skill evaluation method based on multiple dimensions and a personalized flight skill path optimization method, which comprises the following steps:

[0006] Step S1, record flight data through a flight simulator or a GARMIN device in actual flight, preprocess the collected flight data, and establish a standard driving behavior model, the algorithm expression of which is:

[0007] ;

[0008] In the formula, G ( s ) is a transfer function, s is a Laplace operator, a i and b i is a model parameter, m denotes the m th bi model parameters, n denotes the nth a i model parameters, the subscripts of two models i is changed following the order of the model;

[0009] Step S2, using system identification method and parameter identification algorithm to model the cleaned flight data, to establish the key skill index model of pilots in each flight stage, and to obtain the quantitative index error between the standard driving behavior model through comparison for level evaluation;

[0010] Step S3, according to the results of the multi-dimensional flight skill evaluation model, combining multi-source data, using improved ant colony algorithm and path cost function, to build pilot-aircraft-environment digital twin;

[0011] Step S4, real-time monitoring of pilot's skill performance, according to the actual flight data to optimize and iterate the planned path, and constantly adjust the path parameters and weight coefficients;

[0012] Step S5, using the ground control station to monitor the flight path in real time, according to the actual situation changes in the flight process to feed back the key data in the flight process to the multi-dimensional flight skill evaluation model, and to continuously optimize and update the model;

[0013] Step S6, regularly updating the general aviation flight expert knowledge base, and building a continuous improvement ecology for pilot skill improvement.

[0014] Preferably, in step S1, the recorded flight data includes aircraft pitch attitude, roll attitude, pitch angle rate, roll angle rate, target heading and target speed.

[0015] Preferably, in step S2, the parameter identification algorithm is least square method, and the objective function of the least square method is:

[0016] ;

[0017] In the formula: is the actual observation value, is the model prediction value, N is the sample size.

[0018] Preferably, in step S2, each flight stage is ground preparation stage, ground activity stage, approach stage, cruise stage, zone adjustment stage and landing stage.

[0019] Preferably, in step S2, the flight skill level of the pilot is evaluated, and the algorithm formula is as follows:

[0020] ;

[0021] wherein: C is the flight cost, L is the flight length, H is the flight altitude, A is the turn angle, w 1 , w 2 , w 3 is the weight coefficient.

[0022] Preferably, in the step S3, the pilot-aircraft-environment digital twin is constructed, a three-dimensional visualization engine is integrated with a dynamic optimization algorithm, and three types of data sources to be fused for path planning include individual skill atlas, task demand analysis, and knowledge atlas recommendation.

[0023] Preferably, in the step S3, an improved ant colony algorithm is used for path optimization, and a dynamic adjustment mechanism of evaporation coefficient is introduced, and the specific formula is as follows:

[0024] ;

[0025] wherein, p min = 0.1, p max = 0.9, the decay coefficient k is related to the pilot skill improvement rate, t is time, e is a natural constant, about 2.71828, and the path cost function comprehensively considers training time, fuel consumption, and risk index, and the path cost function is as follows:

[0026] ;

[0027] wherein, F ( t ) is the instantaneous fuel flow, R i is the risk event penalty term, w 1 , w 2 , w 3 is the above weight coefficient, T train is the training duration, dt is the differential of the time variable, is the area of the curved trapezoid surrounded by the function and the coordinate axis in the interval [0, T ], in the formula, i = 1 represents the starting term of summation, that is, from i ​=1 start taking values, M The termination term represents the sum, that is, take i= M End, the overall meaning is to take the variable i From 1 to M And the corresponding expression is accumulated.

[0028] Preferably, in the step S5, the actual situation in the flight process includes weather changes and route adjustments.

[0029] Preferably, in the step S6, the continuous improvement ecology of pilot skill improvement is constructed, including the following steps:

[0030] Step S6.1, analyze the latest regulations, accident investigation reports and model manual updates through natural language processing, automatically extract new skill requirements into the evaluation dimension, calculate the term similarity by using the word embedding model, and realize automatic classification of knowledge items;

[0031] Step S6.2, deploy incremental support vector machine algorithm, trigger model fine-tuning when new data accumulation reaches threshold, avoid catastrophic forgetting, and dynamically adjust learning rate through forgetting factor;

[0032] Step S6.3, develop instructor control panel, support manual adjustment of evaluation weight and insertion of customized training module, system records instructor intervention behavior, and finds implicit knowledge through association rule mining.

[0033] Preferably, in the step S6.2, the forgetting factor dynamic adjustment formula is as follows:

[0034] ;

[0035] In the formula: k The control forgetting speed, t The current time, T total The total duration, η 0 is the coefficient.

[0036] Compared with the prior art, the beneficial effects of the present application are as follows:

[0037] 1、The present application constructs a key skill index model covering ground preparation, approach, cruise and other flight stages through system identification method and parameter identification algorithm, quantifies the error between pilot operation and standard model, realizes dynamic multi-dimensional evaluation, overcomes the problem that traditional system relies on single dimension or static index, comprehensively reflects the skill level of pilot in spatial orientation, program execution and other dimensions, integrates improved ant colony algorithm and multi-objective genetic algorithm, combines individual skill atlas, task demand analysis and knowledge graph recommendation, generates a Pareto optimal solution set, and dynamically adjusts the volatility coefficient and path cost function.

[0038] 2、The present application adapts to individual differences of pilots and dynamic environmental changes through a data-driven mechanism, improves training efficiency, constructs a pilot-aircraft-environment digital twin, combines a three-dimensional visualization engine, a dynamic optimization algorithm and a VR device, supports immersive pre-performance and real-time feedback, breaks through the status quo that existing digital twin technology is limited to equipment maintenance, deeply combines skill evaluation and path optimization, enhances the authenticity and interactivity of the training scene, and uses NLP technology to automatically analyze regulatory documents and accident reports, calculates the similarity of terms and classifies knowledge items, combines an incremental support vector machine to realize model fine-tuning, solves the problem of low efficiency of traditional methods which rely on manual updating, and improves the timeliness and accuracy of the knowledge base through an automated process. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The present application is a flowchart of the method. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0041] The method for multi-dimensional flight skill evaluation and individualized flight skill path optimization includes the following steps:

[0042] Step S1, data acquisition and preprocessing: flight data is recorded by a flight simulator or a GARMIN device in actual flight, including aircraft pitch attitude, roll attitude, pitch angle rate, roll angle rate, target heading, target speed, etc., the collected flight data is cleaned, standardized and processed, abnormal values and noise data are removed, and the accuracy and reliability of the data are ensured;

[0043] Flight simulator: used for standard subject training and data collection by experienced pilots or flight instructors, GARMIN navigation map device: used for recording navigation data such as heading and speed during flight, flight data recording: flight data is recorded by a flight simulator or a GARMIN device in actual flight, including aircraft pitch attitude, roll attitude, pitch angle rate, roll angle rate, target heading, target speed, etc., a large amount of flight data is cleaned, analyzed and mined, key skill indicators of pilots in each flight stage are extracted, system identification method: used for analyzing the driving behavior of pilots, a standard driving behavior model is established, and the algorithm expression is:

[0044] ​

[0045] wherein, G ( s ) is a transfer function, s is a Laplace operator, a i and b i is a model parameter, m denotes the nth m b i model parameter, n denotes the nth a i model parameter, the subscripts of two models i vary with the order of the following models;

[0046] Parameter identification algorithm (such as least square method): parameter identification for standard driving behavior model, the objective function of least square method is:

[0047] ;

[0048] wherein: is an actual observation value, is a model prediction value, N is a sample number;

[0049] Step S2, multi-dimensional flight skill evaluation model training: using system identification method and parameter identification algorithm (such as least square method) to model the cleaned flight data, to establish the key skill index model of pilots in each flight stage (such as ground preparation stage, ground activity stage, approach stage, cruising stage, zone adjustment stage, landing stage), by comparing the key skill index model of the measured pilot with the standard driving behavior model, the quantitative index error between the standard driving behavior model is obtained, and the flight skill level of the pilot is evaluated;

[0050] ;

[0051] wherein: C is a flight cost, L is a flight length, H is a flight height, A is a turning angle, w 1 , w 2 , w 3 is a weight coefficient;

[0052] ​Step S3, personalized flight skill path planning: according to the results of the multi-dimensional flight skill evaluation model, combined with flight mission, weather information, aeronautical chart and other multi-source data, using improved ant colony algorithm and path cost function, the pilot-aircraft-environment digital twin is constructed, and the personalized flight skill improvement path is planned for the pilot. The path should fully consider the skill shortcomings and improvement needs of the pilot to ensure the pertinence and effectiveness of the path;

[0053] The pilot-aircraft-environment digital twin is constructed, the three-dimensional visualization engine and dynamic optimization algorithm are integrated, and the path planning needs to integrate three types of data sources:

[0054] Individual skill map: through the error heat map generated in step two, identify the weak links of the pilot in spatial orientation, energy management, procedure execution and other dimensions;

[0055] Task demand analysis: automatically decode the flight mission, extract the key points of the route, the minimum safe height, and the information of the alternate field, and generate the constraint conditions combined with the weather information (TAF / METAR);

[0056] Knowledge graph recommendation: match similar cases from the expert knowledge base and recommend targeted training subjects (such as VOR / DME approach, single engine failure procedure);

[0057] An improved ant colony algorithm (ACO) is used for path optimization, and a volatility coefficient dynamic adjustment mechanism is introduced, the specific formula is as follows:

[0058] ;

[0059] In the formula, p min =0.1, p max =0.9, decay coefficient k related to the pilot's skill improvement rate, t is time, e is a natural constant, about 2.71828, the path cost function considers training time, fuel consumption, and risk index, the path cost function is as follows:

[0060] ;

[0061] In the formula, F ( t ) is the instantaneous fuel flow, R i is the risk event penalty term, w 1 , w 2 , w 3 is the above weight coefficient, Ttrain for the training duration, dt for the differential of the time variable, for the area of the curved trapezoid formed by the function and the coordinate axes within the interval [0, T ] in the formula , i =1 represents the starting term of the summation, i.e. the value starts from i =1, M represents the termination term of the summation, i.e. it goes up to i= M , the overall meaning is to take the variable i from 1 to M in turn, and to perform cumulative summation on the corresponding expression, generate a Pareto optimal solution set through a multi-objective genetic algorithm (NSGA-II), and select the final path according to the training target by the instructor, the planning result is presented in the form of three-dimensional visualization, including dynamic flight path, key control node prompt, virtual instructor voice guidance, and supports Meta-Quest and other VR devices for immersive preview;

[0062] Step S4, path optimization and iteration: during the flight, the skill performance of the pilot is monitored in real time, the planned path is optimized and iterated according to the actual flight data, and the adaptability and optimality of the path are ensured by continuously adjusting the path parameters and weight coefficients;

[0063] Step S5, real-time monitoring and feedback adjustment: the flight path is monitored in real time by the ground control station, the path is adjusted in time according to the actual situation (such as weather changes, route adjustments, etc.) during the flight, and the key data during the flight is fed back to the multi-dimensional flight skill evaluation model, and the model is continuously optimized and updated;

[0064] Step S6, knowledge base update and model optimization: regularly update the air navigation flight expert knowledge base, and incorporate the latest flight skill evaluation standards and path optimization methods into the knowledge base to build a continuous improvement ecology for pilot skill improvement, and at the same time, according to the actual data and feedback information during the flight, continuously optimize and improve the multi-dimensional flight skill evaluation model and the individualized flight skill path optimization method, improve the accuracy and reliability of the model;

[0065] To build a continuous improvement ecology for pilot skill improvement, including the following steps:

[0066] Step S6.1, automatically extract new skill requirements into the evaluation dimensions by natural language processing (NLP) analysis of the latest regulations (such as FAA / EASA airworthiness directives), accident investigation reports (such as NTSB database), and model manual updates, use word embedding model (Word2Vec) to calculate the similarity of terms, and realize automatic classification of knowledge items;

[0067] Step S6.2, deploy incremental support vector machine (ISVM) algorithm, trigger model fine-tuning when new data accumulation reaches threshold, avoid catastrophic forgetting, learning rate is dynamically adjusted by forgetting factor, forgetting factor dynamic adjustment formula is as follows:

[0068] ;

[0069] In the formula: k is the control forgetting speed, t is the current time, T total is the total duration, η 0 is the coefficient;

[0070] Step S6.3, develop instructor control panel, support manual adjustment of evaluation weight, insert customized training module, system records instructor intervention behavior, find implicit knowledge through association rule mining (Apriori algorithm), for example, a specific instructor prefers to strengthen stable approach training in landing phase, such patterns are automatically converted into system recommended strategies after verification;

[0071] Establish an iterative verification process for digital twins. New models are first simulated in a simulated environment using Monte Carlo simulation. They can only be deployed to real training after passing the 95% confidence test. The entire system forms a flywheel effect of "data-driven - model optimization - practical verification", ensuring continuous improvement of evaluation accuracy and training effectiveness.

[0072] The application overcomes the problem of traditional system relying on single dimension (such as operation time) or static index by constructing key skill index model covering ground preparation, approach, cruise and other flight stages through system identification method (such as transfer function modeling) and parameter identification algorithm (least square method), quantifying the error (such as pitch attitude, roll angular rate, etc.) of pilot operation and standard model, realizing dynamic multi-dimensional evaluation, comprehensively reflecting the skill level of pilot in spatial orientation, program execution and other dimensions, integrating improved ant colony algorithm (ACO) and multi-objective genetic algorithm (NSGA-II), combining individual skill atlas, task demand analysis (air route, weather data) and knowledge graph recommendation, generating a set of Pareto optimal solutions, dynamically adjusting the evaporation coefficient and path cost function (such as fuel consumption, risk index), abandoning the traditional planning mode relying on subjective experience of instructors, adapting to individual differences of pilots and dynamic environmental changes through data-driven mechanism, improving training efficiency, constructing pilot-aircraft-environment digital twin, combining three-dimensional visualization engine, dynamic optimization algorithm and VR equipment (such as Meta-Quest), supporting immersive pre-rehearsal and real-time feedback, breaking through the limitation of existing digital twin technology to equipment maintenance, deeply combining it with skill evaluation and path optimization, enhancing the authenticity and interactivity of training scene, and using NLP technology (word embedding model Word2Vec) to automatically analyze regulations (FAA / EASA), accident reports, calculate term similarity and classify knowledge items, realize model fine-tuning combined with incremental support vector machine (ISVM), solve the problem of low efficiency of traditional method relying on manual update, improve the timeliness and accuracy of knowledge base through automatic process.

[0073] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-dimensional based flight skill assessment and personalized flight skill path optimization method, characterized in that: The method comprises the following steps: Step S1, data acquisition and preprocessing: record flight data through a flight simulator or a GARMIN device in actual flight, preprocess the collected flight data, and establish a standard driving behavior model, the algorithm expression of which is: ； wherein G s is a transfer function, s is a Laplace operator, a i and b i is a model parameter, m denotes the first m b i model parameter, n denotes the nth a i model parameter, the indices of the two models i vary following the order of the models;​​ Step S2, multi-dimensional flight skill evaluation model training: model the cleaned flight data by using a system identification method and a parameter identification algorithm, establish a key skill index model of the pilot in each flight stage, and obtain a quantitative index error between the standard driving behavior model by comparison to evaluate the level; Step S3, individual flight skill path planning: according to the result of the multi-dimensional flight skill evaluation model, combining multi-source data, using an improved ant colony algorithm and a path cost function, constructing a pilot-aircraft-environment digital twin, using the improved ant colony algorithm for path optimization, and introducing a dynamic adjustment mechanism of a volatilization coefficient, the specific formula is as follows: ； wherein p min = 0.1, p max = 0.9, attenuation coefficient k related to the pilot skill improvement rate, t is time, e is a natural constant, approximately 2.71828, the path cost function comprehensively considers training time, fuel consumption and risk index, and the path cost function is as follows: ； In the formula, F ( t R represents the instantaneous fuel flow rate. i As a penalty item for risk events, w 1 , w 2 , w 3 The above weighting coefficients, T train For training duration, dt The derivative of the time variable, To find the function in [0, T The area of ​​the curvilinear trapezoid enclosed by the coordinate axes within the interval is given by the formula. middle, i =1 indicates the starting term for the summation, i.e., from... i Values ​​start from 1. M This indicates the terminator of the summation, i.e., continuing until i = M The end, the overall meaning is to put the variables i Take values ​​sequentially from 1 to M And sum the corresponding expressions; Step S4, path optimization and iteration: real-time monitoring of the skill performance of the pilot, optimization and iteration of the planned path according to the actual flight data, and continuous adjustment of path parameters and weight coefficients; Step S5, real-time monitoring and feedback adjustment: real-time monitoring of the flight path by using a ground control station, and feeding back key data in the flight process to the multi-dimensional flight skill evaluation model according to actual situation changes in the flight process to continuously optimize and update the model; Step S6, knowledge base updating and model optimization: periodically updating the general aviation flight expert knowledge base, and constructing a continuous improvement ecology for pilot skill improvement, comprising the following steps: Step S6.1, updating the latest regulations, accident investigation reports and aircraft manuals by natural language processing, automatically extracting new skill requirements into the evaluation dimension, calculating the term similarity by using a word embedding model, and realizing automatic classification of knowledge items; Step S6.2, deploying an incremental support vector machine algorithm, triggering model fine-tuning when new data accumulation reaches a threshold, avoiding catastrophic forgetting, and dynamically adjusting the learning rate through a forgetting factor; Step S6.3, developing a teacher control panel to support manual adjustment of evaluation weights and insertion of customized training modules, recording the teacher intervention behavior, and discovering implicit knowledge by association rule mining.

2. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method according to claim 1, wherein: In the step S1, the recorded flight data includes aircraft pitch attitude, roll attitude, pitch angle rate, roll angle rate, target heading and target speed.

3. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method of claim 1, wherein: In the step S2, the parameter identification algorithm is a least square method, and a target function of the least square method is: ； In the formula: is the actual observation value, is the model prediction value, N is the sample number.

4. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method of claim 1, wherein: In the step S2, the flight stages are a ground preparation stage, a ground activity stage, an approach stage, a cruising stage, a zone adjustment stage and a landing stage.

5. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method of claim 1, wherein: In the step S2, the flight skill level of the pilot is evaluated, and the algorithm formula is as follows: ； wherein: C is the flight cost, L is the flight length, H is the flight height, A is the turn angle, w 1 , w 2 , w 3 is a weight coefficient.

6. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method according to claim 1, wherein: In the step S3, the pilot-aircraft-environment digital twin is constructed, a three-dimensional visualization engine and a dynamic optimization algorithm are integrated, and three types of data sources to be fused include an individual skill graph, task demand analysis and knowledge graph recommendation.

7. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method of claim 1, wherein: In the step S5, the actual situation in the flight process includes weather changes and route adjustments.

8. The multi-dimensional based flight skill assessment and individualized flight skill path optimization method of claim 1, wherein: In the step S6.2, the forgetting factor dynamic adjustment formula is as follows: ; where: k is a control forgetting rate, t is a current time, T total is a total duration, η 0 is a coefficient.

Images