CN115293225B

CN115293225B - Method and device for analyzing causes of pilot flat-floating ejector rod

Info

Publication number: CN115293225B
Application number: CN202210690209.XA
Authority: CN
Inventors: 郑林江; 尚家兴; 王启星; 陈逢文; 陈红年; 张锐祥; 陈浩东
Original assignee: Chongqing University
Current assignee: Chongqing University
Priority date: 2022-06-17
Filing date: 2022-06-17
Publication date: 2023-04-28
Anticipated expiration: 2042-06-17
Also published as: CN115293225A

Abstract

The application provides a pilot flat-floating ejector rod cause analysis method, which comprises the following steps: acquiring QAR parameters of a rapid access recorder of a plurality of air segments; data preprocessing and parameter selection are carried out on QAR parameters; feature extraction is carried out on QAR parameters subjected to data preprocessing and parameter selection, so that feature data of a plurality of air segments are obtained; performing state coding on the characteristic data of the plurality of navigation segments to obtain characteristic vectors of the plurality of navigation segments; and performing cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a clustering result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the clustering result. According to the method and the device, the pilot flat-drift ejector rod event is subjected to interpretable clustering through the unsupervised clustering model, so that the pilot flat-drift ejector rod event is subjected to deep interpretability research and analysis, and flight specialists can better find the reason of the flat-drift ejector rod.

Description

Method and device for analyzing causes of pilot flat-floating ejector rod

Technical Field

The application relates to the technical field of aviation safety, in particular to a pilot flat-floating ejector rod cause analysis method and device.

Background

With the development of civil aviation, aircraft are becoming one of the preferred long-distance transportation modes. Since the beginning of the 21 st century, the number of flights in the civil aviation industry has steadily increased. However, with the concern and importance of aviation safety, aviation accidents are more likely to be fear of people than other vehicles such as ships and trains, and bring about great attention worldwide. Therefore, the aviation safety is researched, and the early warning of avoiding in advance or the cause analysis of accidents or safety events possibly occurring in aviation are necessary.

In the field of civil aviation safety, there is always a "black 11 minutes" statement that most flight accidents occur within 3 minutes of the take-off phase of the aircraft and 8 minutes of the approach to the landing phase. When the aircraft takes off and approaches to land, the pilot is busiest and highly stressed, particularly in the landing stage, the aircraft flying height is lower, instruments on the aircraft need to receive signals of various navigation devices such as an omnidirectional beacon, a range finder, a heading beacon and the like on the ground, the specific flying height and the flying attitude are maintained according to the condition of the environment state such as weather, wind and the like of the aircraft at the moment, a great deal of manual operation is involved, and once any interference occurs, the altitude, the heading and the like can deviate, so that the flying accident is caused. The final approach and landing phases are the flight phases most prone to major safety incidents. In general, although the overall probability of occurrence of a serious safety accident for an aircraft is low, the occurrence probability of an overrun event that may affect the safety of the aircraft cannot be ignored.

The fast access recorder (Quick Access Recorder, QAR) is an on-board flight recorder with a protection device for monitoring and recording a large amount of time-series flight data, covering most of the parameters of the flight operation quality analysis. QAR monitoring is a scientific and effective technical means for guaranteeing flight safety and improving operation efficiency, and the monitoring result is an important basis for flight safety assessment, technical inspection, safety event investigation and aircraft maintenance. Problems in the aspects of unit operation, aviation performance and the like in flight can be found in time through QAR data, the reasons are analyzed and found, safety dynamics are mastered, and targeted measures are taken, so that potential safety hazards are eliminated, and flight safety is ensured. The QAR device is capable of acquiring a number of flight parameters in real time through a plurality of sensors during the flight of an aircraft, including aircraft state parameters such as radio altitude, airspeed, ground speed, pitch angle, etc., aircraft pilot operating parameters such as aircraft pitch stick control, throttle stick position, roll angle, etc., and aircraft environmental parameters such as current wind speed, wind direction, temperature, etc., as well. Airlines generally provide flight data of overrun events to aviation specialists for case analysis or accident investigation, but with the great increase of flight shifts and the great increase of QAR data, the manual analysis of the aviation specialists is only worry, the manual cost is high, the efficiency is low, the analysis result is influenced by personal experience of the specialists, and objective and effective data support is lacking. With the prevalence of machine learning and the deepening application of QAR data, the method provides possibility for developing advanced early warning prediction and risk event interpretability analysis of overrun events for mining QAR data. For example, by means of a deep learning prediction method, a pilot who is flying and landing can be pre-warned in advance of risk events, so that the pilot is given an opportunity to perform corresponding treatment to avoid a risk state, and landing safety is improved. In addition, in order to more comprehensively ensure aviation safety, airlines not only wish to early warn of overrun events in advance, but also wish to explore and analyze reasons for occurrence of overrun events, especially to analyze reasons for unsafe or improper operation behaviors of pilots, so that flight training can be better and more specifically carried out, pilot skill level and safety consciousness are improved, and aviation flight safety is more comprehensively ensured.

The final approach and landing phases include two very critical phases of flight throughout the flight phase of the aircraft: a leveling stage and a leveling stage. Specifically, when the aircraft descends to 50 feet from the ground, the pilot will typically begin to pull the fly pitch stick at this altitude, an operation stage called the flare-up stage, which is primarily to cause the aircraft's vertical descent rate and horizontal flight speed to drop rapidly. At this point, the flight speed of the aircraft is still high and cannot be immediately grounded, and at this stage, the aircraft needs to continue to drop in speed and land again when it drops to a certain speed, this stage is called a flat-fly stage for the pilot, and it is mainly ensured that the aircraft can be grounded smoothly and softly, avoiding heavy landing. Since the aircraft is grounded immediately at this stage, the pilot can have a large number of operations in a short time, particularly on the aircraft pitch lever, and any minor erroneous operation can have serious consequences, particularly in the class of operations in which the pilot pushes the aircraft pitch lever during the fly-flat stage. In the plane-floating stage, the airspeed of the aircraft is gradually reduced, so that the lift force is also continuously reduced, in order to keep the lift force and the gravity of the aircraft balanced and enable the aircraft to slowly approach the ground to descend, a pilot should further pull a pitching control lever within the controllable range of the attitude of the aircraft to increase the pitch angle of the aircraft so as to lift the lift force of the aircraft to reduce the vertical landing speed (namely the descent rate) of the aircraft, and finally land smoothly and softly. In practice, however, the pilot may push the pitch lever of the aircraft forward during the fly-flat phase due to the aircraft pitch angle being too great or the head wind being too great, an operation known as a fly-flat ram. This action can result in a drop in lift of the aircraft, thereby indirectly increasing the vertical landing speed of the aircraft, resulting in excessive vertical landing loads, increased risk of heavy landing events, and ultimately possibly causing damage to the landing gear of the aircraft, critical passenger personal safety.

At present, the traditional flight overrun event research is mainly developed by flight specialists in the aviation field based on professional knowledge, and the flight specialists have profound knowledge and knowledge in the aviation field, and analyze overrun events by studying and analyzing the flight overrun event, so that the flight specialists analyze and think from the field professional point of view. For example Zhang Zhuo et al, consider the important impact on aviation flight safety from the standpoint of adverse weather, including: various low-altitude wind shears enable the state of an aircraft to be poor, aeronautical accidents are easy to cause, hail weather causes the aircraft to be impacted by hail easily to generate aeronautical accidents, airflow caused by thunderstorm weather is extremely unstable to easily cause aeronautical accidents, and low visibility can cause misoperation of pilots to cause aeronautical accidents. The suggestions given for adverse weather conditions are as follows: a perfect weather observation system is established, the management of weather observation data is enhanced, and special weather observation is performed.

With the rapid development of related technologies in big data age, various data analysis methods are layered endlessly, for the aviation field, the effect brought by analyzing and researching aviation safety events only by relying on the professional knowledge in the aviation field is very limited, the analysis result is relatively rough and the analysis efficiency is low, so that the analysis and mining of the aviation safety events by using QAR data are very important.

Current research into QAR data also has the following problems:

(1) QAR data presents challenges. The QAR data is the core data of airlines, and most airlines do not share the outer disclosures, which makes it difficult for professionals with computer background to obtain the QAR data for analysis, and thus the current academic research methodology for aviation security is not deep enough. On the other hand, the QAR data itself contains ultra-high dimensional parameters (QAR parameters of full-range decoding are up to 2000) and are dynamic time sequence data, and related parameters are closely connected with the expertise of the flight field, and the whole flight flow is involved, and deep communication with pilots and aviation specialists is needed for analysis. Thus, even if QAR data is taken, analysis without the relevant background knowledge is very difficult for professionals with only a computer background, resulting in a high research barrier in this area. In addition, the accuracy and integrity of the flight data cannot be guaranteed, and the original data of the QAR generated by the flight of the aircraft are binary data, and decoding software is required to perform decoding operation so as to convert the binary data into a usable data format. However, since there is no unified industry decoding specification at present, the decoded data may be incomplete or have deviation, and since the QAR data is data collected by the aircraft sensor, the sensor itself collection error (such as the GPS track of the aircraft) or the collection error may also cause inaccurate data.

(2) The interpretability analysis is weak. Most of the current research based on QAR data focuses on statistical analysis of overrun events or investigation of security incidents at specific voyages, and there is little interpretative research effort on revealing common causes and laws of typical security incidents using machine learning methods. In addition, some research remains at the level of the QAR parameters themselves, lacking in-depth profiling of the underlying physical meaning and flight traffic.

Disclosure of Invention

The present application aims to solve, at least to some extent, one of the technical problems in the related art.

Therefore, the first aim of the application is to provide a pilot flat-drift ejector rod cause analysis method, which solves the technical problem that the interpretability of the existing method on the flat-drift ejector rod is weak, and carries out interpretable clustering on pilot flat-drift ejector rod events by providing an unsupervised clustering model, so that deep interpretability research analysis is carried out on the pilot flat-drift ejector rod, and flight specialists can better find the cause of the flat-drift ejector rod, thereby guaranteeing the safety of civil aviation flight.

A second object of the present application is to propose a pilot fly-flat ejector rod cause analysis device.

A third object of the present application is to propose a computer device.

A fourth object of the present application is to propose a non-transitory computer readable storage medium.

In order to achieve the above object, an embodiment of a first aspect of the present application provides a method for analyzing a cause of a pilot flat-floating ejector rod, including: acquiring QAR parameters of a rapid access recorder of a plurality of air segments; data preprocessing and parameter selection are carried out on QAR parameters; feature extraction is carried out on QAR parameters subjected to data preprocessing and parameter selection, so that feature data of a plurality of air segments are obtained; performing state coding on the characteristic data of the plurality of navigation segments to obtain characteristic vectors of the plurality of navigation segments; and performing cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a clustering result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the clustering result.

Optionally, in one embodiment of the present application, the data preprocessing and parameter selection of the QAR parameters includes:

data cleaning is carried out on QAR parameters;

the pitch angle of the aircraft, the descent rate of the aircraft, the upwind and the radio altitude are selected as parameters for feature extraction.

Optionally, in an embodiment of the present application, feature extraction is performed on the QAR parameters subjected to data preprocessing and parameter selection, to obtain feature data of a plurality of air segments, including:

And performing feature extraction on QAR parameters subjected to data preprocessing and parameter selection on each navigation segment to obtain feature data corresponding to each navigation segment so as to obtain the feature data of a plurality of navigation segments.

Optionally, in an embodiment of the present application, feature extraction is performed on QAR parameters of each leg subjected to data preprocessing and parameter selection, so as to obtain feature data corresponding to each leg, including:

acquiring a time interval of feature extraction;

and respectively extracting parameter values of the pitch angle, the descent rate, the upwind and the radio altitude of the airplane at each moment in the time interval as characteristic data.

Optionally, in an embodiment of the present application, performing state encoding on feature data of a plurality of segments to obtain feature vectors of the plurality of segments, including:

and respectively carrying out state coding on the parameter values of the pitch angle, the descent rate, the upwind and the radio altitude of the airplane at each moment in each air section time interval to obtain the characteristic vector corresponding to each air section so as to obtain the characteristic vectors of a plurality of air sections.

Optionally, in one embodiment of the present application, performing state encoding on parameter values of an aircraft pitch angle, an aircraft descent rate, an upwind and a radio altitude at each moment in each leg time interval to obtain a feature vector corresponding to each leg, including:

Respectively comparing the aircraft pitch angle parameter value of each moment in the current navigation section time interval with the aircraft pitch angle parameter values of all navigation sections at corresponding moments, and if the aircraft pitch angle parameter value of a certain moment in the current navigation section is larger than the aircraft pitch angle parameter value of the navigation section at the same moment in a preset proportion, encoding the aircraft pitch angle of the moment in the current navigation section into an abnormal state, otherwise encoding the abnormal state into a normal state, and obtaining the state variable characteristics of the aircraft pitch angle of each moment in the time interval;

respectively comparing the aircraft descent rate parameter value of each moment in the current leg time interval with the aircraft descent rate parameter values of all the legs at corresponding moments, if the aircraft descent rate parameter value of a certain moment in the current leg is larger than the aircraft descent rate parameter value of the preset proportion of legs at the same moment, encoding the aircraft descent rate of the moment in the current leg into an abnormal state, otherwise encoding the aircraft descent rate into a normal state, and obtaining the state variable characteristics of the aircraft descent rate of each moment in the time interval;

comparing the upwind parameter value at each moment in the current air section time interval with the upwind parameter values at the corresponding moments of all the air sections, if the upwind parameter value at one moment in the current air section is larger than the upwind parameter value of the air section at the same moment in the preset proportion, encoding the upwind at the moment in the current air section into an abnormal state, otherwise encoding the upwind into a normal state, and obtaining the state variable characteristics of the upwind at each moment in the time interval;

Comparing the radio height parameter value at each moment in the current air section time interval with the radio height parameter values at the corresponding moments of all the air sections respectively, if the radio height parameter value at one moment in the current air section is smaller than the radio height parameter value of the air section at the same moment in the preset proportion, encoding the radio height at the moment in the current air section into an abnormal state, otherwise encoding the radio height into a normal state, and obtaining the state variable characteristics of the radio height at each moment in the time interval;

and splicing the state variable characteristics of the pitch angle of the airplane, the state variable characteristics of the descent rate of the airplane, the state variable characteristics of the upwind and the state variable characteristics of the radio altitude at each moment in the time interval of the current navigation section to obtain the characteristic vector corresponding to the current navigation section.

Optionally, in an embodiment of the present application, performing cluster analysis on feature vectors of a plurality of segments by using a K-media algorithm to obtain a cluster result, and obtaining a cause of a behavior of the flat-floating ejector rod according to the cluster result, where the method includes:

and (3) inputting the feature vectors of the plurality of navigation segments as sample data into a K-media algorithm, selecting a proper clustering parameter K for clustering analysis to obtain a clustering result, and carrying out statistical analysis on the clustering result to obtain the cause of the behavior of the flat-floating ejector rod.

To achieve the above object, an embodiment of a second aspect of the present application provides a pilot flutter ejector rod cause analysis device, including:

the acquisition module is used for acquiring QAR parameters of the quick access recorder of a plurality of air segments;

the preprocessing module is used for carrying out data preprocessing and parameter selection on the QAR parameters;

the feature extraction module is used for carrying out feature extraction on the QAR parameters subjected to data preprocessing and parameter selection to obtain feature data of a plurality of navigation segments;

the state coding module is used for carrying out state coding on the characteristic data of the plurality of air segments to obtain characteristic vectors of the plurality of air segments;

and the cluster analysis module is used for carrying out cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a cluster result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the cluster result.

To achieve the above objective, an embodiment of a third aspect of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor executes the computer program to implement the method for analyzing the cause of a pilot fly ejector pin according to the above embodiment.

To achieve the above object, a fourth aspect of the present application proposes a non-transitory computer-readable storage medium, which when executed by a processor, is capable of performing a pilot fly ram cause analysis method.

According to the pilot flat-drift ejector rod cause analysis method, device, computer equipment and non-transitory computer readable storage medium, the technical problem that the existing method lacks of interpretive study on flat-drift ejector rod behaviors is solved, and an unsupervised clustering model is provided to conduct interpretive clustering on pilot flat-drift ejector rod events, so that deep interpretive study analysis is conducted on pilot flat-drift ejector rods, flight specialists can find flat-drift ejector rod causes better, and civil aviation flight safety is guaranteed.

Additional aspects and advantages of the application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application.

Drawings

The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

FIG. 1 is a flow chart of a pilot flutter ejector rod cause analysis method provided in an embodiment of the present application;

FIG. 2 is a graph of wind speed and direction change of a pilot's fly-over ejector pin cause analysis method in an embodiment of the present application.

FIG. 3 is a graph showing an example of the variation of the contour coefficients of the K-means algorithm of the pilot fly-over ejector rod cause analysis method according to the embodiment of the present application.

FIG. 4 is an exemplary graph of the upwind impact of the pilot fly ejector pin cause analysis method in accordance with an embodiment of the present application.

FIG. 5 is an exemplary diagram of pitch impact of a pilot fly ram cause analysis method according to an embodiment of the present application.

FIG. 6 is a diagram showing an example of the high impact of the pilot fly ejector pin cause analysis method according to the embodiment of the present application.

FIG. 7 is a diagram showing an example of the self-influence of the pilot flutter ejector rod cause analysis method according to the embodiment of the present application.

Fig. 8 is a schematic structural diagram of a pilot flat-floating ejector rod cause analysis device according to a second embodiment of the present disclosure.

Detailed Description

Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and intended for the purpose of explaining the present application and are not to be construed as limiting the present application.

The following describes a pilot flat-nose ejector rod cause analysis method and device according to the embodiment of the application with reference to the accompanying drawings.

Fig. 1 is a flow chart of a method for analyzing the cause of a pilot's flat-nose lift pin according to an embodiment of the present application.

As shown in FIG. 1, the pilot flat-floating ejector rod cause analysis method comprises the following steps:

step 101, acquiring QAR parameters of a quick access recorder of a plurality of navigation segments;

102, carrying out data preprocessing and parameter selection on QAR parameters;

step 103, extracting features of the QAR parameters subjected to data preprocessing and parameter selection to obtain feature data of a plurality of air segments;

104, performing state coding on the characteristic data of the plurality of navigation segments to obtain characteristic vectors of the plurality of navigation segments;

and 105, performing cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a clustering result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the clustering result.

According to the pilot flat-floating ejector rod cause analysis method, quick access recorder QAR parameters of a plurality of air segments are obtained; data preprocessing and parameter selection are carried out on QAR parameters; feature extraction is carried out on QAR parameters subjected to data preprocessing and parameter selection, so that feature data of a plurality of air segments are obtained; performing state coding on the characteristic data of the plurality of navigation segments to obtain characteristic vectors of the plurality of navigation segments; and performing cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a clustering result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the clustering result. Therefore, the technical problem that the prior method lacks of interpretive study on the behavior of the flat-floating ejector rod can be solved, and the non-supervision clustering model is provided to carry out interpretive clustering on the events of the flat-floating ejector rod of the pilot, so that the pilot flat-floating ejector rod is subjected to deep interpretive study and analysis, and flight specialists can better find the reasons of the flat-floating ejector rod, thereby guaranteeing the safety of civil aviation flight.

In the embodiment of the application, the problem definition of the flat floating ejector rod comprises that

Given a flight leg and corresponding QAR parameters thereof, first find the first grounding moment of the landing stage of the aircraft according to the left landing gear state (LDGL) and the right landing gear state (LDGR) in the QAR parameters, namely the first grounding moment of any landing gear on the left and right, and record as t _landing . The time when the aircraft radio altitude drops to 10 feet (10 ft) is then found from the left radio altitude (ralt_lh) parameter of the QAR parameters, denoted as t _10ft Satisfy t _10ft <t _landing . If in section [ t ] _10ft ,t _landing ]Occurrence of pilotEjector rod behavior, i.e. there is a moment t e t _10ft ,t _landing ]Such that the PITCH CONTROL parameter PITCH CONTROL among the QAR parameters _t >And 0, defining that the flat-floating ejector rod event occurs in the navigation segment.

The method finds the cause of the occurrence of the flat-floating ejector rod event by analyzing the characteristics of multidimensional QAR parameters in the flight section.

Further, in the embodiment of the present application, performing data preprocessing and parameter selection on the QAR parameter includes:

data cleaning is carried out on QAR parameters;

Data cleaning is performed on the QAR parameters. Because of possible sensor acquisition errors, decoding errors and other problems, field errors or information missing occur in part of the flight data in the QAR data, and therefore, the error fields need to be filtered out or the missing information needs to be complemented. On the other hand, partial leg data content is not complete, namely, only partial flight process data is contained, and some leg data are flight training data, namely, the departure place and the destination are the same, and the data need to be removed.

Because the QAR data characteristic parameter is various data collected by the aircraft sensor, the data shows two types of characteristics, namely, a continuous numerical value type data value represents the specific size of a measured value of the parameter at a certain moment, and a discrete state type data represents the discrete state of the parameter at a certain moment. Most parameter types are continuous numerical types such as radio altitude, airspeed, pitch, etc.; a small portion of the parameters are discrete state parameters such as landing gear state, flap state, etc. For the processing of discrete state data types, taking the landing gear state as an example, the landing gear state parameters are three, namely a left landing gear state (LDGL), a right landing gear state (LDGR) and a front landing gear state (LDGNOS), which correspond to two states, namely an AIR state and a GROUND state, wherein AIR represents that the aircraft is in the AIR (ungrounded state) at the moment, and GROUND represents that the aircraft is in the GROUND (grounded state) at the moment. To convert such state variables to intuitively available numerical parameters, the AIR state is converted to 1 and the group state is converted to 0. The landing gear state parameters are utilized to help locate the landing time point of the aircraft, which is important for extracting the occurrence interval of the flat-floating ejector rod event and the characteristic extraction interval.

For the processing of the characteristic parameter sampling frequency, in the QAR data set, the sampling frequencies of different parameters are different, the variation range is different from 1Hz to 8Hz, and the application aligns all characteristic parameter frequencies to 1Hz. By consulting flight specialists, the method selects wind speed, wind direction, aircraft magnetic heading, descent rate, altitude and pitch angle parameters as basic characteristics, and combines priori knowledge and expert experience to process different parameters correspondingly. The sampling frequency of wind speed, wind direction, airplane magnetic heading and descent rate is 1Hz, and the sampling frequency of altitude and pitch angle is 4Hz.

According to the opinion of flight specialists, the pitch angle of the aircraft is greatly influenced by the upwind, but the upwind is not included in the original QAR flight parameters, and the method and the device calculate the upwind air quantity of the aircraft by combining the magnetic heading, the wind speed and the wind direction of the aircraft. As shown in fig. 2, firstly, an included angle between the magnetic heading and the wind direction of the aircraft needs to be obtained, and then, components of the wind speed in two directions of the aircraft, namely the vertical direction and the parallel direction, are calculated by utilizing a trigonometric function relation, wherein an upwind calculation formula is expressed as follows:

wherein beta represents wind direction, alpha represents airplane magnetic heading, theta is an included angle between the two, and WIN _spd Indicating wind speed and WIN _crs Indicating the magnitude of crosswind, i.e. wind perpendicular to the direction of flight, WIN _head Representing the upwind magnitude, i.e. the wind parallel to the direction of flight.

Since all parameters are frequency aligned to 1Hz, the pitch angle and radio altitude parameters, which were originally at 4Hz, need to be down-converted. The pitch control of the aircraft is to directly operate the pitch angle, and the ejector rod behavior is closely related to the excessive pitch angle, so that the pitch angle control method selects to take the maximum value of 4 pitch angle values in one second. For the height parameter, the method selects a radio height average value within one second as the height value of the current moment so as to analyze the relation between the height and the flat-floating ejector rod.

Finally, the aircraft descent rate parameter is closely related to the aircraft pitch angle and landing, and therefore this parameter is also selected as the characteristic parameter.

Further, in the embodiment of the present application, feature extraction is performed on QAR parameters subjected to data preprocessing and parameter selection, so as to obtain feature data of a plurality of air segments, including:

Further, in the embodiment of the present application, feature extraction is performed on QAR parameters of each leg subjected to data preprocessing and parameter selection, so as to obtain feature data corresponding to each leg, including:

Acquiring a time interval of feature extraction;

After data preprocessing and parameter selection are performed on QAR parameters, feature extraction is performed first. The present application uses the time t of the flat-floating ejector rod (i.e. interval t _10ft ,t _landing ]The first occurrence of PITCH_CONTROL>0) as a reference time point, aligning all the voyages according to the reference time point, and then taking the interval from 4 seconds to 1 second before the reference time point, namely [ t-4, t-1 ]]As a time interval for feature extraction. Since this section includes 4 seconds in total, the pitch angle, the descent rate, the upwind magnitude, and the radio altitude parameter value at the corresponding time are respectively characterized for each second in the section.

Further, in the embodiment of the present application, performing state encoding on feature data of a plurality of segments to obtain feature vectors of the plurality of segments, including:

After feature extraction is performed on the QAR parameters, each feature is respectively subjected to state coding. The continuous numerical feature may be encoded into discrete boolean state features by a state encoding method. The basic idea of state coding is: and comparing the individual characteristic value of the current navigation section with the group values of all navigation sections at the same time, if the individual characteristic value exceeds a certain threshold value, encoding the individual characteristic value into an abnormal state (with the value of 1), and otherwise, encoding the individual characteristic value into a normal state (with the value of 0).

Further, in the embodiment of the present application, status encoding is performed on parameter values of an aircraft pitch angle, an aircraft descent rate, an upwind and a radio altitude at each moment in each leg time interval, so as to obtain a feature vector corresponding to each leg, including:

Illustratively, the extracted pitch angle, descent rate, headwind size and radio altitude characteristics at each instant in the above-obtained [ t-4, t-1] time interval are state-coded below.

With respect to the pitch angle feature, in combination with expert experience, it is known that when the state of the aircraft is too great (i.e., the pitch angle is too great), the pilot tends to take ejector pin operations to avoid the risk of tail rub. Therefore, the pitch angles of all the voyages at the moment are ranked from large to small, if the pitch angle value of the current voyage at the moment is ranked by 5%, the pitch angle is coded into an abnormal state 1, and otherwise, the pitch angle is coded into a normal state 0. Since there are 4 times (t-1, t-2, t-3, t-4), the resulting state variables are PITCH_1, PITCH_2, PITCH_3, PITCH_4.

The rate of descent is typically negative (indicative of descent) during the aircraft landing phase, and a greater absolute value indicates a faster descent. For the descent rate characteristics, according to expert experience, the descent speed of the aircraft before landing is faster and unsafe events are more likely to occur, so that the descent rates of all the voyages are ranked from large to small according to absolute values, if the absolute value of the descent rate of the current voyage at the moment is ranked 5% before, the voyage is coded into an abnormal state 1, and otherwise, the voyage is coded into a normal state 0. The encoded state variables are IVV _1, IVV_2, IVV_3, IVV_4, respectively.

According to the experience of flight specialists, when the upwind of the aircraft is larger, the aircraft is more likely to be excessively large in attitude, so that ejector rod operation is caused. Therefore, the upwind of all the voyages is ranked according to the slave size, if the upwind size of the current voyage at the moment is ranked by 5%, the upwind is encoded into an abnormal state 1, otherwise, the upwind is encoded into a normal state 0. The encoded state variables are WIN_1, WIN_2, WIN_3, WIN_4, respectively.

According to flight expert experience, when an aircraft is subjected to low-altitude flat-drift (i.e., the aircraft is very low but not grounded at all times) for a long period of time, an overrun event is easily triggered, which is far from flat-drift, and in order to avoid this event, pilots often also take ejector pin operations. Thus, the left radio altitude parameter of the aircraft is ordered from small to large, and if the left radio altitude at that time of the current leg is ordered by the first 5%, it is encoded as an abnormal state 1, otherwise it is encoded as a normal state 0. The encoded state variables are HEIGHT_1, HEIGHT_2, HEIGHT_3, and HEIGHT_4, respectively.

And splicing the 16 state variable characteristics to obtain the final state variable characteristic, wherein the dimension is 16. And then, carrying out cluster analysis on the final state variable characteristics by using a K-media algorithm to obtain the cause of the behavior of the flat-floating ejector rod.

The K-means clustering algorithm is a partition-based unsupervised clustering algorithm that can divide a given data set into K clusters. Each cluster has at least one data object, and as such, each data object can only exist in one cluster. In the K-means clustering algorithm, the center point of each cluster must be a sample point in the cluster, and the similarity between sample points is measured by its manhattan distance, with smaller distances having higher similarity. For the state variable feature vector defined in this application, the Manhattan distance between two sample points is equivalent to its number of state variables that are not equal, so the K-means clustering algorithm naturally applies to the state feature vector. Furthermore, the K-means algorithm generally has better cluster robustness than the K-means algorithm.

K-media algorithm basic idea: first, k data objects, each representing the center of the initial cluster, are randomly selected, and then the following allocation and update steps are iteratively performed until the clusters are stable.

And (3) distribution: for each data object x in the sample set _i The minimum Manhattan distance from each cluster center is calculated and assigned to the cluster corresponding to the cluster center most similar to it (smallest distance), the calculation formula is as follows, where u _j Distance is Manhattan Distance calculation formula for the j-th cluster center.

C _i ＝arg _j minDistance(x _i ,u _j )

Wherein Distance is Manhattan Distance calculation formula, x _i Data object, u _j Is the j-th cluster center.

Updating: recalculating each cluster center u _j J=1, 2,..k, calculated as: for each data object in the jth cluster, calculating the average distance between the data object and other objects in the cluster, and then selecting a sample with the smallest average distance from the clusters as a new cluster center.

The algorithm repeats the above allocation and update process until convergence, which ultimately enables minimization of the following objective function:

wherein J is an objective function of a K-means algorithm, K is the number of clustered categories (clusters), N is the number of data samples, i is the number of samples, distance is a Manhattan Distance calculation formula, and x _i Data object, u _j Is the j-th cluster center.

For the super-parameter k selection problem, the application uses the contour coefficient (Silhouette Coefficient) to select the most appropriate k value and evaluate the algorithm performance. Profile coefficientThe combination of the degree of aggregation and the degree of separation is evaluated, the value is between-1 and 1, and the larger the value is, the better the clustering effect is. Given data object x _i Let a _i Representing data object x _i Average distance to other objects in the same cluster, d _ij Data object x _i Average distance to all objects in other cluster j, and

is x _i Minimum average distance to other clusters, data object x _i The profile coefficients of (a) are as follows:

wherein s is _i Data object x _i Profile coefficient, a _i Representing data object x _i Average distance to other objects in the same cluster, d _ij Data object x _i Average distance to all objects in the other cluster j,

b _i is x _i Minimum average distance to other clusters.

The contour coefficient of the whole clustering result is defined as the contour coefficient mean value of all data objects in the sample set, namely:

/>

wherein N is the number of data samples, s _i Data object x _i Is a contour coefficient of (c).

The application relates to a super parameter setting experiment of a K-media cluster model. For the K-means model, the total number of iterations is 500. In addition, the super parameter k is mainly selected according to the contour coefficient and expert experience, and the result is shown in fig. 3, the contour coefficient change condition that k takes 2 to 10 is given in the graph, and as can be seen from the graph, when k takes 4, the clustering effect of the model is good.

The behavior types of pilot flat-floating ejector rod events are classified into 4 types through a K-media algorithm, and because the input characteristic data of the model are all Boolean state coding data, abnormal coding characteristics in clustering results are directly counted in visual display, and the distribution condition of the abnormal coding characteristics of each type, namely the number of times of occurrence of abnormal characteristic states, is given. Each type of results is analyzed separately below to provide an explanation of pilot fly ejector pin behavior.

First category: as shown in fig. 4, it can be seen that this type is manifested in a very high proportion of 4 seconds of sustained upwind anomalies, which are very prominent relative to the other several time series parameters, meaning that the pilot pushing the pitch horn is primarily due to the effects of wind, so that such a flat-nose ram behavior may be referred to as an upwind impact class.

The second category: as shown in fig. 5, it can be seen that this type of behavior is represented by a continuously increasing proportion of the pitch angle value of the aircraft for 4 seconds, with a small portion of the following rise in the rate of descent (IVV) of the aircraft, indicating that this type of pilot performs a flat-nose stick behavior because the pitch angle of the aircraft has been continuously increasing until then, while the rate of descent of the aircraft is relatively high at the beginning, and later, as the pitch angle is continuously increasing, the abnormal rate of descent of the aircraft is gradually relieved, however, at this point the pitch angle of the aircraft is already great, and in order to avoid the risk of tail-scrubbing, the pilot performs a stick behavior, which may be referred to as a pitch-influencing class.

Third category: as shown in fig. 6, it can be seen that this type of aircraft is represented by a continuously low 4 second aircraft altitude value, which is very prominent relative to the other several time series parameters, and that this type of fly-over ejector behavior may be referred to as altitude-influencing class, since the aircraft is continuously flying at a relatively low altitude, the pilot's pushing operation may bring the aircraft into contact with the ground as soon as possible, thereby avoiding runway-out risks.

Fourth category: as shown in fig. 7, it can be seen that this type of behavior is that the various parameter anomalies occupy relatively small amounts at various time points, and most of the characteristic parameter states are normal, so that the behavior of the pilot's fly-away ejector pin of this type can be regarded as an unnecessary dangerous operation behavior, called self-influencing class, and training should be focused on this type of pilot.

In general, the unsupervised clustering model provided by the application can perform interpretable clustering on pilot flat-drift ejector rod events, can reflect time characteristics, has more outstanding obtained various types of characteristics, and shows very visual interpretability. Through deep interpretation research and analysis on the pilot flat-drift ejector rod, flight experts can better find the reason of the flat-drift ejector rod, so that the flight skill level of the pilot and the safety management level of an airline company are further improved, and the safety of civil aviation flight is guaranteed.

As shown in fig. 8, the pilot's flying ejector pin cause analysis device includes:

an acquisition module 10 for acquiring quick access recorder QAR parameters for a plurality of voyages;

A preprocessing module 20 for performing data preprocessing and parameter selection on the QAR parameters;

the feature extraction module 30 is configured to perform feature extraction on the QAR parameters subjected to data preprocessing and parameter selection, so as to obtain feature data of a plurality of air segments;

the state coding module 40 is configured to perform state coding on the feature data of the plurality of segments to obtain feature vectors of the plurality of segments;

the cluster analysis module 50 is configured to perform cluster analysis on the feature vectors of the plurality of navigation segments by using a K-mediums algorithm, obtain a cluster result, and obtain a cause of behavior of the flat-floating ejector rod according to the cluster result.

The pilot flat-floating ejector rod cause analysis device of this application embodiment includes: the acquisition module is used for acquiring QAR parameters of the quick access recorder of a plurality of air segments; the preprocessing module is used for carrying out data preprocessing and parameter selection on the QAR parameters; the feature extraction module is used for carrying out feature extraction on the QAR parameters subjected to data preprocessing and parameter selection to obtain feature data of a plurality of navigation segments; the state coding module is used for carrying out state coding on the characteristic data of the plurality of air segments to obtain characteristic vectors of the plurality of air segments; and the cluster analysis module is used for carrying out cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a cluster result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the cluster result. Therefore, the technical problem that the prior method lacks of interpretive study on the behavior of the flat-floating ejector rod can be solved, and the non-supervision clustering model is provided to carry out interpretive clustering on the events of the flat-floating ejector rod of the pilot, so that the pilot flat-floating ejector rod is subjected to deep interpretive study and analysis, and flight specialists can better find the reasons of the flat-floating ejector rod, thereby guaranteeing the safety of civil aviation flight.

In order to implement the above embodiment, the application further provides a computer device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor implements the pilot flutter ejector rod cause analysis method according to the above embodiment when executing the computer program.

In order to implement the above embodiment, the present application further proposes a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the pilot fly ram cause analysis method of the above embodiment.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "plurality" is at least two, such as two, three, etc., unless explicitly defined otherwise.

Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and additional implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present application.

Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

It is to be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As with the other embodiments, if implemented in hardware, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.

Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.

In addition, each functional unit in each embodiment of the present application may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.

The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like. Although embodiments of the present application have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the application, and that variations, modifications, alternatives, and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the application.

Claims

1. The pilot flat-floating ejector rod cause analysis method is characterized by comprising the following steps of:

acquiring QAR parameters of a rapid access recorder of a plurality of air segments;

performing data preprocessing and parameter selection on the QAR parameters;

feature extraction is carried out on QAR parameters subjected to data preprocessing and parameter selection, so that feature data of a plurality of air segments are obtained;

performing state coding on the characteristic data of the plurality of navigation segments to obtain characteristic vectors of the plurality of navigation segments;

performing cluster analysis on the feature vectors of a plurality of navigation segments by using a K-mediums algorithm to obtain a clustering result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the clustering result;

wherein, the data preprocessing and parameter selection of the QAR parameter comprises:

data cleaning is carried out on the QAR parameters;

Selecting an aircraft pitch angle, an aircraft descent rate, an upwind and a radio altitude as parameters for feature extraction;

the feature extraction is performed on the QAR parameters subjected to data preprocessing and parameter selection to obtain feature data of a plurality of air segments, and the feature extraction comprises the following steps:

respectively carrying out feature extraction on QAR parameters of each navigation segment subjected to data preprocessing and parameter selection to obtain feature data corresponding to each navigation segment so as to obtain feature data of a plurality of navigation segments;

the feature extraction is performed on QAR parameters of each navigation segment subjected to data preprocessing and parameter selection to obtain feature data corresponding to each navigation segment, and the method comprises the following steps:

acquiring a time interval of feature extraction;

respectively extracting parameter values of pitch angle, descent rate, upwind and radio altitude of the airplane at each moment in the time interval as characteristic data;

the step of performing state coding on the characteristic data of the plurality of air segments to obtain characteristic vectors of the plurality of air segments comprises the following steps:

respectively carrying out state coding on parameter values of the pitch angle, the descent rate of the airplane, the upwind and the radio height of the airplane at each moment in the time interval of each navigation segment to obtain a characteristic vector corresponding to each navigation segment so as to obtain characteristic vectors of a plurality of navigation segments;

The step of carrying out state coding on the parameter values of the pitch angle, the descent rate, the upwind and the radio altitude of the airplane at each moment in the time interval of each navigation segment to obtain the characteristic vector corresponding to each navigation segment comprises the following steps:

respectively comparing the aircraft pitch angle parameter value of each moment in the time interval of the current navigation section with the aircraft pitch angle parameter values of all the navigation sections at corresponding moments, and if the aircraft pitch angle parameter value of a certain moment in the current navigation section is larger than the aircraft pitch angle parameter value of the navigation section at the same moment in a preset proportion, encoding the aircraft pitch angle of the moment in the current navigation section into an abnormal state, otherwise encoding the abnormal state into a normal state, and obtaining the state variable characteristics of the aircraft pitch angle of each moment in the time interval;

respectively comparing the aircraft descent rate parameter value of each moment in the time interval of the current leg with the aircraft descent rate parameter values of all corresponding moments of the legs, if the aircraft descent rate parameter value of a certain moment of the current leg is larger than the aircraft descent rate parameter value of the leg with the preset proportion at the same moment, encoding the aircraft descent rate of the moment of the current leg into an abnormal state, otherwise encoding the aircraft descent rate into a normal state, and obtaining the state variable characteristics of the aircraft descent rate of each moment in the time interval;

Comparing the upwind parameter value of each moment in the time interval of the current navigation section with the upwind parameter values of all navigation sections at corresponding moments, if the upwind parameter value of a certain moment in the current navigation section is larger than the upwind parameter value of the navigation section with preset proportion at the same moment, encoding the upwind of the moment in the current navigation section into an abnormal state, otherwise encoding the upwind into a normal state, and obtaining the state variable characteristics of the upwind of each moment in the time interval;

respectively comparing the radio height parameter value of each moment in the time interval of the current leg with the radio height parameter values of all the legs at corresponding moments, if the radio height parameter value of a certain moment in the current leg is smaller than the radio height parameter value of the legs at the same moment in the preset proportion, encoding the radio height of the moment in the current leg into an abnormal state, otherwise encoding the radio height into a normal state, and obtaining the state variable characteristics of the radio height of each moment in the time interval;

2. The method of claim 1, wherein the performing cluster analysis on the feature vectors of the plurality of legs using a K-means algorithm to obtain a cluster result, and obtaining a cause of the behavior of the fly-back ejector according to the cluster result comprises:

and inputting the feature vectors of the plurality of navigation segments as sample data into a K-media algorithm, selecting a proper clustering parameter K for clustering analysis to obtain a clustering result, and carrying out statistical analysis on the clustering result to obtain the cause of the behavior of the flat-floating ejector rod.

3. A pilot flutter ejector pin cause analysis device, characterized by comprising:

the cluster analysis module is used for carrying out cluster analysis on the feature vectors of the plurality of navigation segments by using a K-media algorithm to obtain a cluster result, and obtaining the cause of the behavior of the flat-floating ejector rod according to the cluster result;

data cleaning is carried out on the QAR parameters;

acquiring a time interval of feature extraction;

4. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method according to any of claims 1-2 when executing the computer program.

5. A non-transitory computer readable storage medium, having stored thereon a computer program, which when executed by a processor, implements the method according to any of claims 1-2.