CN107045638A - A kind of flight safety affair analytical method based on context-aware model - Google Patents
A kind of flight safety affair analytical method based on context-aware model Download PDFInfo
- Publication number
- CN107045638A CN107045638A CN201611260332.9A CN201611260332A CN107045638A CN 107045638 A CN107045638 A CN 107045638A CN 201611260332 A CN201611260332 A CN 201611260332A CN 107045638 A CN107045638 A CN 107045638A
- Authority
- CN
- China
- Prior art keywords
- feature
- context
- classification
- aware model
- aware
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of flight safety affair analytical method based on context-aware model, comprise the following steps：S1, is analyzed history flying quality, sets up the context-aware model based on Bow tie；S2, refines the characteristic point for the context-aware model set up, and primary data dictionary is set up according to context-aware model and the characteristic point refined；S3, flying quality is obtained by QAR, and the selected characteristic classification in primary data dictionary is classified according to the feature classification of selection to the flying quality of acquisition；S4, is ranked up in each category content of the feature classification of selection by clustering to feature, obtains key feature according to sequence, and optimize primary data dictionary.Testing efficiency is the method increase, testing time and testing cost is reduced.
Description
Technical field
The present invention relates to a kind of flight safety affair analytical method, more particularly to a kind of flight based on context-aware model
Safety case investigation method, belongs to flight information processing technology field.
Background technology
With the fast development of flight cause, flight safety event is more and more more valued by people.In order to solve to fly
Flight safety event in machine flight course, the analysis of flight safety event is essential.The analysis of flight safety event is in QAR
It is an important job in (quick access recorder) data analysis.Context-aware (Situation awareness, referred to as
SA critically important height) has been mentioned, as important tools of analysis.It helps us from cockpit and outside, goes to perceive
With understand its environment, and help pilot to expect potential difficulty in advance, appointing based on target (such as taking off or land)
Business instructs how pilot overcomes these difficult.
People can form continuing based on relevant information and comprehensive brain is drawn when being related to more complicated running environment
Face, is used as the foundation of decision-making.This is referred to as context-aware.Why context-aware SA is so important in flying security evaluation
As flight new hand, each step that context-aware has permeated training is set up, is trained from the most basic flight that meets accident to simulation
Machine training, to advanced flying training, or even finally enters airline and starts daily aerial mission.As long as theory, technical ability, knowledge
Storage level is sufficient, and the foundation of context-aware gradually will be strengthened and repeat.Therefore, context-aware can be understood as entering and fly
The blood of office staff, when we assess safe flight performance, check the security of context-aware transfinites for our research safeties
The analysis of event just turns into most suitable point of penetration.
QAR monitors whole context-aware process, and possibility is provided for ex-post analysis.Excavated by QAR big datas, can
To reappear context-aware processing procedure, and assess the performance of pilot.But how according to the processing procedure pair of context-aware
Flight safety event is analyzed, and feature is ranked up according on the importance that flight safety event influences, so as to obtain
Key feature, improves the importance of principal character and payes attention to dynamics, reduces the importance of secondary feature and payes attention to dynamics, existing
Research in do not relate to.
The content of the invention
In view of the shortcomings of the prior art, the technical problems to be solved by the invention are to provide a kind of based on context-aware mould
The flight safety affair analytical method of type.
For achieving the above object, the present invention uses following technical schemes：
A kind of flight safety affair analytical method based on context-aware model, comprises the following steps：
S1, is analyzed history flying quality, sets up the context-aware model based on Bow-tie；
S2, refines the characteristic point for the context-aware model set up, and is set up according to context-aware model and the characteristic point refined
Primary data dictionary；
S3, flying quality, the selected characteristic classification in primary data dictionary, according to the feature class of selection are obtained by QAR
The other flying quality to acquisition is classified；
S4, is ranked up in each category content of the feature classification of selection by clustering to feature, according to sequence
Key feature is obtained, and optimizes primary data dictionary.
Wherein more preferably, in step sl, the context-aware model based on Bow-tie, is replaced with context-aware
The inducement event changed in Bow-tie, the operation of abstract pilot is risky situation；
Context-aware model based on Bow-tie includes 7 parts：Risky situation, reason summary, correction/control measure, wind
Dangerous event, further correction/control, further event and potential result；
Each part includes the situation classification occurred in aircraft flight.
Wherein more preferably, S31, chooses the feature classification as class categories；
S32, by carrying out learning to obtain Weak Classifier to training dataset, Gentle is obtained by Weak Classifier weighted sum
Adaboost graders；
S33, by the Gentle Adaboost graders of acquisition to the feature classification of the flying quality of acquisition according to selection
Classified, obtain other feature classifications and the relation for the feature classification chosen.
Wherein more preferably, in step s 4, by clustering to spy in each category content of the feature classification of selection
Levy and be ranked up, key feature is obtained according to sequence, comprised the following steps：
S41, calculates cluster centre, and obtain each feature to the distance of cluster centre using GK clustering algorithms；
S42, is ranked up according to the distance of each feature to cluster centre to feature, and key feature is obtained according to sequence；
S43, repeat step S41~S42 are until the key feature in each category content for the feature classification chosen is equal
Untill acquisition.
Wherein more preferably, in step s 4, it is described optimization primary data dictionary be selection feature classification each
Feature is ranked up by clustering in category content, the feature after sequence is updated in primary data dictionary in sequence.
Wherein more preferably, the flight safety affair analytical method based on context-aware model, it is characterised in that also
Comprise the following steps：
S5, existing flying quality is obtained by QAR in real time, and existing flying quality is carried out according to the data dictionary after optimization
Analysis judges, when there is occurrence risk event risk, sends early warning.
Flight safety affair analytical method provided by the present invention based on context-aware model, sets up and is based on Bow-tie
Context-aware model；And the characteristic point of the context-aware model of foundation is refined, set up primary data dictionary；By being obtained to QAR
The flying quality taken is classified and clustering processing, obtains the incidence relation between feature classification in primary data dictionary, according to
Incidence relation optimizes primary data dictionary.Data dictionary after optimization can be carried out with the context-aware model based on Bow-tie
Displaying, clearly shows that the relation between various pieces, and context-aware that can be in aircraft flight is carried out to result
Aircraft safety business reasons can also be analyzed, and assess the performance of pilot by prediction and warning according to flight result.
Brief description of the drawings
Fig. 1 is the flow chart of the flight safety affair analytical method provided by the present invention based on context-aware model；
Fig. 2 is the structural representation of existing basic Bow-tie models；
Fig. 3 is the structural representation of existing strengthening version Bow-tie models；
Fig. 4 is in one embodiment provided by the present invention, the part-structure of the context-aware model based on Bow-tie shows
It is intended to；
Fig. 5 is in one embodiment provided by the present invention, the primary data dictionary after optimization is used based on Bow-tie's
The part-structure schematic diagram of context-aware model display.
Embodiment
Detailed specific description is carried out to the technology contents of the present invention with specific embodiment below in conjunction with the accompanying drawings.
As shown in figure 1, the flight safety affair analytical method based on context-aware model that the present invention is provided, is specifically included
Following steps：First, history flying quality is analyzed, sets up the context-aware model based on Bow-tie；Secondly, refine
The characteristic point of the context-aware model of foundation；And set up according to the context-aware model based on Bow-tie and the characteristic point refined
Primary data dictionary；Then, flying quality, the selected characteristic classification in primary data dictionary, according to selection are obtained by QAR
Feature classification is classified to the flying quality of acquisition；Finally, pass through in each category content of the feature classification of selection
Cluster is ranked up to feature, and key feature is obtained according to sequence.Detailed specific description is done to this process below.
S1, is analyzed history flying quality, sets up the context-aware model based on Bow-tie.
Before analysis context-aware is started, the analysis of the frequency occurs according to the accident to mission phase, it can be seen that
61% event occurs in landing period, and the time shared by it is only the 24% of whole flight route.The limit of safety,
In this stage, fade away, it is meant that the ability of pilot can not meet mission requirements sometimes.Therefore, landing period is all
The focus that airline focuses more on, therefore, just starts with when analyzing context-aware from the event of landing period.
Exist in the context-aware model provided by the present invention based on Bow-tie and traditional context-aware method very big
Difference.Traditional context-aware method and the particular model for not setting up the event that transfinited based on flight safety, and it is more powerful
Method be namely based on the network analysis of accident/event tree and correcting method, i.e. Bow-tie models.Traditional Bow-tie models
Top event is listed, for example, airplane fault.Adopting said method, input series is potential to cause the inducement event of failure, such as Fig. 2
It is shown, for example：Fuel oil, which lacks, causes engine cut-off, and the result of the output of this Bow-tie model is similar to land in an emergency.By portion
The possibility and the seriousness of accident divided are defined.Fig. 3 gives the Bow-tie models of a strengthening version, covers control
System/corrective measure and degradation factors.
In embodiment provided by the present invention, Bow-tie models are combined with context-aware model, anticipated using scene
Know to replace the inducement event in Bow-tie, the operation of abstract pilot is affair character, and applies artificial intelligence/machine
Learning algorithm automates judgement accident/event tree, gives more preferable logic.In addition, machine learning can easily by
Pilot's behavior is classified or clustered, with significant statistical significance.
As shown in figure 4, analyzing history flying quality, the context-aware model based on Bow-tie is set up, is based on
Bow-tie context-aware model includes 7 parts：Risky situation, reason summary, correction/control measure, risk case, enter one
Walk correction/control, further event and potential result.Each part includes the situation classification occurred in aircraft flight.
It the following specifically describes the situation categorised content of each part.In embodiment provided by the present invention, portion is only described
Dividing influences larger and more typical situation categorised content on flight safety event.On risky situation, the classification of its situation about including
It is more, in embodiment provided by the present invention, only describe wherein to influence larger preceding 12 risky situations.First risk feelings
Scape refers to relative trajectory (glide path), related throttle and target airspeed etc..If these values have exception, then aircraft can enter naturally
Enter risk situation.We should be noted the influence factor of throttle：Full weight, height, temperature, slip down to gradient, the influence of wind, target
Speed etc..Second risky situation is wing flap configuration.When flap angle is small, flying speed quickly increases, and slows down slower.The front of a garment which button on the right
The wing is than small flap angle more suitable for downwind landing, but small wing flap configuration is more suitable for flying in strong wind than big wing flap.3rd
Individual risky situation is to even up evening on opportunity.4th risky situation is that the process of evening up falls head.5th risky situation is that processing is jumped
Landing concept is incorrect.6th risky situation is that excessive throttle of receiving produces stronger nose-down pitching moment.7th risky situation be
Mistake is accustomed to, including pilot thinks that aircraft is suitable, and leaves control-rod and gas pedal, and two pilots remind mutually
After start adjustment, it is possible to become big from very little is started to the later stage.We are also recalled that near and descent is entered, even secondary
Drive, hand and pin should not also leave control-rod and pedal, as it is possible that can not quick sensing risk.30ft is changed in landing
The bottom line gone out, prompting is that one kind is lost time and chance each other.Left-hand seat adjustment amount is small, and later stage change has greatly same risk.
Above-mentioned 7 kinds of risky situations are possible to cause big rate of descent or control stick slightly (to belong to the feelings of reason conclusion suddenly
Condition is classified), the two are the major reasons for causing landing again (the situation classification for belonging to risk case).Risky situation is returned with reason
Receive, the situation of the relation between risk case and reason conclusion, risk case etc. is sorted in and is subsequently described in detail.The
Eight risky situations are to the flat amendment floatd.9th risky situation is the amendment too slow to low latitude sinking track.Tenth wind
Dangerous situation scape is downwind landing, is generally all that aerodynamic performance also can gradually improve from big to small with the wind.If wind changes
Quickly, flight path can be influenceed, if change is very slow, throttle is influenceed.If operation is that steady bar (belongs to correction/control measure
Situation classification) reduce aircraft pitch angle, under this control, downwind landing is inappropriate, it is possible to cause to land again
Or earth point is remote.
11st risky situation is low clearance high crosswind, and best solution is " drift method " (adjustment machine in theory
Brilliance degree) and " sideslip method " (change roll angle).If downwind landing, during low clearance wide-angle, there is very big head angle
Change, is noted also that with the wind, and new hand can become in a rush (the situation classification for belonging to reason conclusion), can pass through simplification
The countermeasure situation of correction/control measure (belong to classification) is controlled and revised, it may occur however that risk case situation classification
It is that earth point is remote, jumps and land and wipe tail.Last risky situation be land before posture it is small, i.e., it is so-called fall head (belong to
The situation classification that reason is concluded).It is in a rush and fall head not only easily trigger land again, landing point it is remote, can equally cause jump
Land or wipe tail.
On risky situation, the classification of its situation about mainly including includes：Rate of descent is big, pumping rod (bar has retardance), steady bar
It is improper, in a rush and fall head etc..
On correction/control measure (precautionary measures i.e. before accident), the classification of its situation about mainly including includes：Steady bar has
Amount, steady, deceleration of being floatd with the wind with steady bar, simplified countermeasure, drawing, are grounded again after posture increase.
On risk case, the classification of its situation about mainly including includes：Land again, earth point is remote, jumps and lands, wipes tail
Deng.
On further correction/control, the classification of its situation about mainly including includes：Note the landing distance of high-altitude aerodrome more
Situation further classification etc. in long, jump landing mission.
The main cause for jumping landing is three undercarriage engineerings while landing, or nose-gear first suffers ground.Because aircraft
First half apart from center of gravity farther out, nose-gear can rebound it is aerial, posture becomes big, and lift also becomes big, therefore aircraft can takeoff
Come.The main cause for wiping tail is that before aircraft jumps up, spoiler stretches out and opened, and lift is destroyed, the increase of posture
Trend is more serious, or even has wiped tail.When landing, remote landing point, jump landing and wiping tail generation again, we can enter one
Amendment/control is walked to avoid continuing deteriorating.For example, if landing point is remote, airport is located at plateau, it would be desirable to which reserved ratio is in Plain
The remote landing distance in airport.
When jumping landing generation, it would be desirable to which concern caused by the excessive reduction angle of pitch due to falling head.In addition, if
Aircraft rebound is aerial, and spoiler, which stretches out, to rise, lift destruction, it is equally possible to occurs to land again, jump and land.Occurring to land again
When, if runway is remaining not enough, estimating jump landing can be very serious, and it is exactly to go around to have a control recovery measure.Again land and again,
Jump the structural damage that the latent consequences landed is likely to result in tire or airframe again.If pick-up point is remote, latent consequences can
Can gun off the runway.If high crosswind, gun off the runway it can also happen that.
On further event (control measure i.e. before accident), the classification of its situation about mainly including includes：It is expected that overweight,
There is no information to runway Distance Remaining and it is expected that overweight etc..
On potential result, the classification of its situation about mainly including includes：Aero tyre even fuselage produce structural damage,
Gun off the runway.
On the relation between seven parts in the context-aware model based on Bow-tie, pass through the flight number obtained to QAR
According to being classified and clustering processing, it can obtain, subsequently be described in detail.
S2, refines the characteristic point for the context-aware model set up, and is set up according to context-aware model and the characteristic point refined
Primary data dictionary.
According to the context-aware model based on Bow-tie of foundation, gem-pure business reasons analysis knot can be obtained
Structure.It is preceding to have addressed, the analysis of the frequency occurs according to the accident to mission phase, it can be seen that 61% event occurs in landing rank
Section, and the time shared by it is only the 24% of whole flight route.Therefore, landing period is that all airlines focus more on
Focus, analyze context-aware when just start with from the event of landing period.Analyzed according to aircraft history flying quality, in landing
Stage, 30ft is the minimum terrain clearance gone around.Overshoot maneuver, which is formed, needs event delay, therefore for changing from touch down attitude
Go out, it is necessary in liftoff at least more than 30ft.And generally aircraft is entering above runway mouthful untill 50ft, luffing angle 1 to 2 is spent, when
During ground connection, the angle of pitch is typically 4 to 5 degree.In embodiment provided by the present invention, the spy for the context-aware model set up is refined
Levying can a little use in SIFT algorithms, spin image algorithms, HoG algorithms, RIFT algorithms, Textons algorithms, GLOH algorithms
One kind, because above-mentioned algorithm is the algorithm of existing traditional extraction characteristic point, is just repeated no more herein.Provided by the present invention
Embodiment in, use SIFT algorithms extract characteristic point may be referred to number of patent application for 201510111168.4 China apply
A kind of improved scale invariant feature matching algorithm based on wavelet transformation carries out the extraction of characteristic point disclosed in patent.
In consideration of it, the logic of the concern information and above-mentioned business reasons analytical structure in aircraft flight is determined
The justice characteristic point of 75 dimensions.For example：In elevation dimension, " field is high " parameter in QAR is used.First, for a machine
The same runway of field, calculates the monthly average height that below 100ft is measured, to eliminate the influence in season and environment.Then make
Real flight path is described with 5 characteristic points in 100ft, 50-100ft, 50ft, 50ft to ground connection, earth point, these stages
Whether average height is higher than.Because this is an experimental research, in the present invention, QAR data sampling frequencies are 1 second 1 time,
It from 50ft to ground connection, generally only have recorded 3 times, therefore do not refine this stage.As long as but need, sampling can be improved
Rate, and the dimension of defined feature again, than expanding 10-20 times now, as that will be compared with average value as feature.On
Glide path deviates latitude, also records its state using 5 characteristic points, in 100ft, 50ft-100ft, 50ft, 50ft to ground connection
Point, earth point.On air speed latitude, in the present invention, 3 characteristic points are have recorded, mcp target velocities are in 100ft, 50ft-
100ft、50ft.Other 2 characteristic points have recorded the Vref related to air speed in 50ft to earth point, earth point.On pitching
Angle latitude, have recorded 6 characteristic points of the angle of pitch, i.e., in 100ft, 50ft-100ft, 50ft, 50ft to earth point, earth point, connect
The later monthly average value in ground.Last part is to assess to land and jump again to land.Similar, 6 angles of pitch are controlled in addition
Characteristic point, and other 5 characteristic points for being directed to rate of descent.For crosswind and with the wind, 5 characteristic points each have recorded theirs
Value and wind direction.One characteristic point is used to record wing flap configuration, the big approach speeds of 500ft-50ft, big touchdown speed, landing flap
Configuration is late in place, and the ground connection angle of pitch is big, and the ground connection angle of pitch is small, and pick-up point is near, lands again, jumps and lands again, touchdown speed is small.
After the characteristic point for refining the context-aware model set up, according to the context-aware model based on Bow-tie and carrying
The characteristic point of refining sets up primary data dictionary.
S3, flying quality, the selected characteristic classification in primary data dictionary, according to the feature class of selection are obtained by QAR
The other flying quality to acquisition is classified.
Flying quality is obtained by QAR, the feature classification included according to primary data dictionary is entered to the flying quality of acquisition
Row classification.In embodiment provided by the present invention, using Gentle Adaboost classifier methods to the flight number to acquisition
According to this risk case and feature point group into feature classification classified for correlation.By correlation, for example whether " 50ft is to connecing
The situation of ground hypertelorism " or " ground connection float time length " is classified, and finds the relation of these features, and these features are such as
What influences flat situation of floaing until ground connection.
First introduce lower used method.AdaBoost is the abbreviation of " adaptive boosting " algorithm, is machine learning member calculation
Method, is proposed, and obtained in 2003 by Yoav Freund and Robert SchapirePrize.Other study are calculated
Method (" weak study ") is also added to the part as weighted sum, is used as a part for last lifting grader.If weak point
Class device has distorted classification before, and AdaBoost is adaptive.
Flying quality is obtained by QAR, the feature classification included according to primary data dictionary is entered to the flying quality of acquisition
Row classification, specifically includes following steps：
S31, chooses the feature classification as class categories；In embodiment provided by the present invention, by risk case with
Feature point group into feature classification be used as class categories.For example：" 50ft is remote to ground distance ", or " ground connection float time length ".
S32, by carrying out learning to obtain Weak Classifier to training dataset, Gentle is obtained by Weak Classifier weighted sum
Adaboost graders, specifically include following steps：
S321, initializes the probability distribution of training data, has just started to be uniformly distributed, D1=(w11, w12..., w1N), its
In,Dm represent m wheel iteration start before, the probability distribution (or weights distribution) of training data,
wmiThe weights in i-th of sample are represented,Assume that training dataset has when initial to be uniformly distributed, i.e., it is each
Training sample acts on identical in the study of Weak Classifier.
S322, to m=1,2 ..., M, carries out learning to obtain one weak point using the training dataset for being distributed Dm with weights
Class device：Gm(x)=X → { -1 ,+1 }.
Wherein, any naive Bayesian can be selected by carrying out study using the training dataset that Dm is distributed with weights, certainly
Plan tree, a kind of model such as SVM, and each round iteration can use different models.Calculate Weak Classifier Gm(x) in training number
According to error in classification the rate em and G on collectionm(x) factor alpham.Wherein,
S323, updates the weights distribution D of training datam+1=(wM+1,1, wM+1,1..., wM+1, N), wherein,
Zm is standardizing factor,So,Make
It is a probability distribution to obtain Dm+1.
S324, linear combination is carried out by M basic classification deviceObtain final
Gentle Adaboost graders
In, each Gm(X) it is a weak typing
Device, x is output as an actual value, it is indicated that the classification belonging to x as input.The output symbol of Weak Classifier has reacted prediction mesh
Target generic, absolute value embodies the confidence level that it belongs to the category.Similar, M-layer graders can output symbol
For just, if sample belongs to positive classification, vice versa.
Boosting algorithms before are all to select G as far as possiblem(X) all test mistakes for reducing and often walking, are minimized.
GentleBoost defines the size of each step.When Weak Classifier is classified, GentleBoost can select Gm(X) it is equal to
yi。
S33, by the Gentle Adaboost graders of acquisition to the feature classification of the flying quality of acquisition according to selection
Classified, obtain other feature classifications and the relation for the feature classification chosen.For example：In risky situation (other feature classifications)
Relative trajectory (glide path), related throttle and target airspeed, wing flap configuration, even up evening on opportunity, the process of evening up and fall head etc. very
In big degree aircraft can be caused to land again (risk case in the feature classification of selection).
S4, is ranked up in each category content of the feature classification of selection by clustering to feature, according to sequence
Key feature is obtained, and optimizes primary data dictionary.
In embodiment provided by the present invention, using risk case and feature point group into feature classification as selection spy
Classification is levied, feature is ranked up by clustering in each category content of the feature classification of selection, for example：To in classification
The feature held in " 50ft to ground distance too far " or " ground connection float time length " is ranked up by cluster, according to sequence acquisition
Key feature, specifically includes following steps：
S41, calculates cluster centre, and obtain each feature to the distance of cluster centre using GK clustering algorithms.
Data-oriented collection X, selects ＜ c ＜ N, weight coefficient the m ＞ 1 of number of clusters 1, terminates tolerable coefficient ε ＞ 0, norm
Induced matrix A.Random initializtion subdivision matrix (Initialize the partitionmatrixrandomly) U(0)So thatFollowing step is repeated to l=1,2 ...：
First, cluster centre is calculated:
Wherein, μikIt is center cluster, represents degree of membership of the data point relative to cluster centre, and xkRepresent data point (mark
Note vector).
Then, the cluster covariance matrix of ith cluster cluster is calculated according to cluster centre：
If adding a scaling matrices on the basis of the above:
γ is proportionality coefficient.Extract eigenvalue λijAnd фij；It was found that
λi,max=maxiλijAnd setForIt can recalculate
Obtain after cluster covariance matrix, calculate each feature to the distance of cluster centre:
Pi is adjustment
Coefficient.
Finally, Matrix dividing is updated:
Until | | U(l)-U(l-1)| | ＜ ∈.
Wherein, use GK algorithms to carry out cluster and obtain cluster centre and each feature to the distance of cluster centre for ability
It is related to further illustrating for formula in domain conventional technical means, algorithm and may refer to Wang Hui et al. issues《GK fuzzy classifications are calculated
Application of the method in GIS partial discharge pattern-recognition》(《Electric power system protection and control》Volume 39, the 17th phase in 2011) and
The GIS partial discharge kind identification method based on GK fuzzy clusterings disclosed in Patent No. 201410394763.9, herein just
Repeat no more.
S42, is ranked up according to the distance of each feature to cluster centre to feature, and key feature is obtained according to sequence.
S43, repeat step S41~S42 are until the key feature in each category content for the feature classification chosen is equal
Untill acquisition.Obtain the key feature of each category content of the feature classification chosen.
Feature is ranked up by clustering in each category content of the feature classification of selection, obtained according to sequence
Key feature, optimizes primary data dictionary.Wherein, optimization primary data dictionary is each classification in the feature classification of selection
Feature is ranked up by clustering in content, the feature after sequence is updated in primary data dictionary in sequence.Optimization
Primary data dictionary afterwards is as shown in Figure 5 using the part-structure schematic diagram of the context-aware model display based on Bow-tie.
The basis for classification/cluster result that good feature set has been.Probably ignore important using too small set
Aspect, therefore the data used are The more the better.In addition, in embodiment provided by the present invention, having beyond expected discovery.Allow
We come feature-rich dimension and solution.For example, some are potential regular in business reasons reasoning, such as：
1) according to sequence, different weights are assigned to risky situation.For example, quick low clearance downside risk ranking:A.
One, 20ft stop increase angle of pitch initiation and fall head；B. second, even up caused by visual problems float it is late；C. the 3rd, enter runway
Before, throttle is excessively dropped, the angle of pitch is reduced, causes to fall the moment of torsion etc. of head and reduction head.
2) details of amendment/control is paid close attention to, for example, the amendment to low clearance rate of descent slowly：Excessively increase flies when a.20ft
The machine angle of pitch is easy to trigger excessive control, causes to fall head, and landing state is unstable；B. too early, earth point is started when evening up
Too far, it is careful not to over-correction/control；C. the increase angle of pitch can cause low clearance is flat to float, such as 20ft, and the reduction angle of pitch must
Must little by little it come, should not be excessive.
Excavated by QAR big datas, we present flight landing process again, and have evaluated the behavior of pilot.Wind
Dangerous situation scape is because some reason firing events.Also consider category in our model for the modification method of risky situation.Manually
Intelligence and machine learning algorithm are used for comprehensive automation and divide pilot's behavior.We use the Gentle based on supervision
Adaboost classification, and based on non-supervisory G-K clustering procedures.The reference of gentle adaboost algorithms transfinites event to instruct
Practice model, and assess test result, G-K clusters then use free pattern.
S5, existing flying quality is obtained by QAR in real time, and existing flying quality is carried out according to the data dictionary after optimization
Analysis judges, when there is occurrence risk event risk, sends early warning, pilot is dealt with time.
According to the primary data dictionary of optimization, QAR data can be analyzed, in time to pilot with early warning,
To make a response in time, the generation of flight safety event is prevented.Analysis can also be carried out according to aircraft flight result to reason to test
Card.For example, the reasoning result provided according to the primary data dictionary of optimization is：Certain day a, flight flies to city from city A
City B, will occur to land again, peak value 1.77G.The same day QAR record weather be：Left side wind 8.38 is saved, and 9.92 is saved with the wind greatly, landing
Shi Shunfeng disappears.QAR data are taught that：During 100ft, enter near substantially steady.Wing flap 30,56.43 tons of full weight, approach speed-
5.25, air speed 134.56 is saved.It is too early to start flat opportunity of floaing to increase luffing angle during landing, and at this afterwards, rate of descent is too small,
Maximum 574ft/ minutes.Due to by influenceing greatly with the wind, aircraft is easy to abnormal flat float.Because empty with the wind from greatly to disappearance
Aerodynamic force becomes big, therefore the increase angle of pitch is slack-off.There is the behavior of a reduction angle of pitch, in 50ft height, air speed is 132.39
Section, is saved, aerodynamic quality can cause the shortage of lift less than mcp target velocities 135.Rate of descent still very little, maximum
444.9ft/min, pilot continues to reduce the angle of pitch.In 50ft to ground connection, rate of descent suddenly becomes too much, and numerical value is changed into
570.38ft/min.This is the angle of pitch because excessively reducing.Angle of pitch increase is too slow, causes head to decline, and minimum is bowed
Elevation angles 2.45.Before landing, aircraft begins with roll, maximum angle 3.12.Course is also changed, 6.11 degree, left side pair
The wing deviates.Pilot resists high crosswind, theoretically the best way using sideslip method and drift method simultaneously.But simultaneously,
Pilot needs to solve greatly with the wind, and situation is extremely complex, easily in a rush.During landing, air speed 132.21 is saved, and is more than
Vref127 is saved, it is meant that remaining air force is too strong.If pilot gets used to accelerating and big power using large throttle, I
Recommend pilot using wing flap 40 in the case where wind speed is small.The landing angle of pitch very well, does not jump significantly.
Another example：2015, a flight flew to plateau city B from plateau city A, occurred to land again during landing,
Peak value 1.82G, and occur to jump the weather data in landing, peak value 1.80G.QAR, 6.6 sections against the wind, are especially suitable for flight.QAR numbers
According to teaching that：Aircraft altitude is higher by the height for the aircraft that other 84% same months land with runway, therefore his energy during 100ft
Amount is very big.Wing flap 40,56.89 tons of full weight, approach speed 1.25, air speed 133.37 is saved.In 50ft, air speed 132.65 is saved, is higher than
Mcp target velocities -131 are saved, and cause lift larger.Pilot reduces the angle of pitch.In 50ft to ground connection, rate of descent became
Greatly, maximum 731.78ft/min.Pilot wants to change the too high situation of aircraft, and the flat increased angle of pitch that floats is excessively slow,
Cause head, minimum 0.63 degree of the angle of pitch.During landing, air speed 129.68 is saved, and is saved greatly than Vref 126, it is meant that air is moved
Power is larger.The landing angle of pitch is too small, only 2.32 degree, close to 3 points of ground connection.Because fuselage forebody point is away from center of gravity, preceding to rise and fall
Frame is very easy to jump.Jump landing and there occurs that first half fuselage takeoffs, the angle of pitch is up to 5.76 degree.Work as aircraft
When still in the air, spoiler and brake are fully open, destroy the lift of aircraft, cause jump and land, land again.
In summary, the flight safety affair analytical method provided by the present invention based on context-aware model, to history
Flying quality is analyzed, and sets up the context-aware model based on Bow-tie；Refine the feature for the context-aware model set up
Point, primary data dictionary is set up according to context-aware model and the characteristic point refined；Flying quality is obtained by QAR, initial
Selected characteristic classification in data dictionary, classifies according to the feature classification of selection to the flying quality of acquisition；In the spy of selection
Feature is ranked up by clustering in each category content for levying classification, key feature is obtained according to sequence, optimization is initial
Data dictionary.Data dictionary after optimization can be shown with the context-aware model based on Bow-tie, be clearly showed that each
Relation between individual part, context-aware that can be in aircraft flight is predicted early warning to result, can also root
Aircraft safety business reasons are analyzed according to flight result, and assess the performance of pilot.
The flight safety affair analytical method provided by the present invention based on context-aware model has been carried out in detail above
Explanation.For those of ordinary skill in the art, to appointing that it is done on the premise of without departing substantially from true spirit
What obvious change, will all constitute to infringement of patent right of the present invention, and will undertake corresponding legal liabilities.
Claims (6)
1. a kind of flight safety affair analytical method based on context-aware model, it is characterised in that comprise the following steps：
S1, is analyzed history flying quality, sets up the context-aware model based on Bow-tie；
S2, refines the characteristic point for the context-aware model set up, and sets up initial according to context-aware model and the characteristic point refined
Data dictionary；
S3, flying quality, the selected characteristic classification in primary data dictionary, according to the feature classification pair of selection are obtained by QAR
The flying quality of acquisition is classified；
S4, is ranked up in each category content of the feature classification of selection by clustering to feature, is obtained according to sequence
Key feature, and optimize primary data dictionary.
2. the flight safety affair analytical method as claimed in claim 1 based on context-aware model, it is characterised in that：
In step sl, the context-aware model based on Bow-tie, is that luring in Bow-tie is replaced with context-aware
Because of event, the operation of abstract pilot is risky situation；
Context-aware model based on Bow-tie includes 7 parts：Risky situation, reason summary, correction/control measure, risk thing
Part, further correction/control, further event and potential result；
Each part includes the situation classification occurred in aircraft flight.
3. the flight safety affair analytical method as claimed in claim 1 based on context-aware model, it is characterised in that：
S31, chooses the feature classification as class categories；
S32, by carrying out learning to obtain Weak Classifier to training dataset, Gentle is obtained by Weak Classifier weighted sum
Adaboost graders；
S33, is carried out by the Gentle Adaboost graders of acquisition to the flying quality of acquisition according to the feature classification of selection
Classification, obtains other feature classifications and the relation for the feature classification chosen.
4. the flight safety affair analytical method as claimed in claim 1 based on context-aware model, it is characterised in that in step
In rapid S4, feature is ranked up by clustering in each category content of the feature classification of selection, obtained according to sequence
Key feature, comprises the following steps：
S41, calculates cluster centre, and obtain each feature to the distance of cluster centre using GK clustering algorithms；
S42, is ranked up according to the distance of each feature to cluster centre to feature, and key feature is obtained according to sequence；
S43, repeat step S41~S42 are until the key feature in each category content for the feature classification chosen is obtained
Untill.
5. the flight safety affair analytical method as claimed in claim 1 based on context-aware model, it is characterised in that：
In step s 4, the optimization primary data dictionary is by poly- in each category content of the feature classification of selection
Class is ranked up to feature, and the feature after sequence is updated in primary data dictionary in sequence.
6. the flight safety affair analytical method as claimed in claim 1 based on context-aware model, it is characterised in that also wrap
Include following steps：
S5, existing flying quality is obtained by QAR in real time, and existing flying quality is analyzed according to the data dictionary after optimization
Judge, when there is occurrence risk event risk, send early warning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611260332.9A CN107045638B (en) | 2016-12-30 | 2016-12-30 | Flight safety event analysis method based on situational awareness model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611260332.9A CN107045638B (en) | 2016-12-30 | 2016-12-30 | Flight safety event analysis method based on situational awareness model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107045638A true CN107045638A (en) | 2017-08-15 |
| CN107045638B CN107045638B (en) | 2021-03-30 |
Family
ID=59543905
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611260332.9A Active CN107045638B (en) | 2016-12-30 | 2016-12-30 | Flight safety event analysis method based on situational awareness model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107045638B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109284925A (en) * | 2018-09-21 | 2019-01-29 | 南华大学 | A kind of measurement method, device, equipment and the storage medium of teams and groups' context-aware |
| CN109918905A (en) * | 2017-12-12 | 2019-06-21 | 财团法人资讯工业策进会 | Behavior inference model generating means and its behavior inference model generating method |
| CN109978168A (en) * | 2019-03-19 | 2019-07-05 | 北京瑞斯克企业管理咨询有限公司 | The origin cause of formation automated reasoning method and system of landing again based on timing QAR parameter curve cluster |
| CN110069070A (en) * | 2019-05-08 | 2019-07-30 | 成都高威节能科技有限公司 | A method of improving the safety of large aircraft take-off process |
| CN110390177A (en) * | 2019-07-31 | 2019-10-29 | 中国民航科学技术研究院 | The flying object that peels off determines method and device |
| CN110533095A (en) * | 2019-08-27 | 2019-12-03 | 中国民航大学 | A kind of schedule flight risk behavior recognition methods based on improvement random forest |
| CN111008669A (en) * | 2019-12-10 | 2020-04-14 | 北京航空航天大学 | Deep learning-based heavy landing prediction method |
| CN111626327A (en) * | 2020-04-15 | 2020-09-04 | 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） | Airplane heavy landing prediction method and device, computer equipment and storage medium |
| CN112862621A (en) * | 2021-01-15 | 2021-05-28 | 广州亚美信息科技有限公司 | Vehicle insurance risk assessment method and device and computer equipment |
| CN113066315A (en) * | 2021-03-03 | 2021-07-02 | 中国民用航空飞行学院 | Scene capture analysis method for civil aircraft approach landing stage flight safety |
| CN114004292A (en) * | 2021-10-29 | 2022-02-01 | 重庆大学 | Pilot flat-floating ejector rod behavior analysis method based on flight parameter data unsupervised clustering |
| CN114647198A (en) * | 2022-03-09 | 2022-06-21 | 深圳市经纬纵横科技有限公司 | Intelligent home control method and system based on Internet of things and electronic equipment |
| CN114937317A (en) * | 2022-04-02 | 2022-08-23 | 珠海翔翼航空技术有限公司 | QAR data management method, system and storage medium for mobile terminal |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113792489B (en) * | 2021-09-15 | 2024-01-05 | 国家石油天然气管网集团有限公司 | Method for establishing pipeline risk prevention and control model and method for establishing safety risk catalogue |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120317058A1 (en) * | 2011-06-13 | 2012-12-13 | Abhulimen Kingsley E | Design of computer based risk and safety management system of complex production and multifunctional process facilities-application to fpso's |
| CN104239970A (en) * | 2014-09-04 | 2014-12-24 | 国家电网公司 | Power transmission line gallop risk early-warning method based on Adaboost |
| CN104331843A (en) * | 2014-10-30 | 2015-02-04 | 华中科技大学 | Transformer fault risk assessment method based on bowknot model |
| CN104537487A (en) * | 2014-12-25 | 2015-04-22 | 云南电网公司电力科学研究院 | Assessment method of operating dynamic risk of electric transmission and transformation equipment |
| CN105654229A (en) * | 2015-11-26 | 2016-06-08 | 国家电网公司 | Power grid automation system and equipment running state risk assessment algorithm |
-
2016
- 2016-12-30 CN CN201611260332.9A patent/CN107045638B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120317058A1 (en) * | 2011-06-13 | 2012-12-13 | Abhulimen Kingsley E | Design of computer based risk and safety management system of complex production and multifunctional process facilities-application to fpso's |
| CN104239970A (en) * | 2014-09-04 | 2014-12-24 | 国家电网公司 | Power transmission line gallop risk early-warning method based on Adaboost |
| CN104331843A (en) * | 2014-10-30 | 2015-02-04 | 华中科技大学 | Transformer fault risk assessment method based on bowknot model |
| CN104537487A (en) * | 2014-12-25 | 2015-04-22 | 云南电网公司电力科学研究院 | Assessment method of operating dynamic risk of electric transmission and transformation equipment |
| CN105654229A (en) * | 2015-11-26 | 2016-06-08 | 国家电网公司 | Power grid automation system and equipment running state risk assessment algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 陆正 等: "基于Bow-tie模型的民机着陆冲出跑道风险分析", 《工业安全与环保》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918905A (en) * | 2017-12-12 | 2019-06-21 | 财团法人资讯工业策进会 | Behavior inference model generating means and its behavior inference model generating method |
| CN109918905B (en) * | 2017-12-12 | 2022-05-10 | 财团法人资讯工业策进会 | Behavior inference model generation device and behavior inference model generation method thereof |
| CN109284925A (en) * | 2018-09-21 | 2019-01-29 | 南华大学 | A kind of measurement method, device, equipment and the storage medium of teams and groups' context-aware |
| CN109978168A (en) * | 2019-03-19 | 2019-07-05 | 北京瑞斯克企业管理咨询有限公司 | The origin cause of formation automated reasoning method and system of landing again based on timing QAR parameter curve cluster |
| CN110069070B (en) * | 2019-05-08 | 2022-01-18 | 成都高威节能科技有限公司 | Method for improving safety of large airplane in takeoff process |
| CN110069070A (en) * | 2019-05-08 | 2019-07-30 | 成都高威节能科技有限公司 | A method of improving the safety of large aircraft take-off process |
| CN110390177A (en) * | 2019-07-31 | 2019-10-29 | 中国民航科学技术研究院 | The flying object that peels off determines method and device |
| CN110533095A (en) * | 2019-08-27 | 2019-12-03 | 中国民航大学 | A kind of schedule flight risk behavior recognition methods based on improvement random forest |
| CN110533095B (en) * | 2019-08-27 | 2023-01-17 | 中国民航大学 | Flight risk behavior identification method based on improved random forest |
| CN111008669A (en) * | 2019-12-10 | 2020-04-14 | 北京航空航天大学 | Deep learning-based heavy landing prediction method |
| CN111626327A (en) * | 2020-04-15 | 2020-09-04 | 中国电子产品可靠性与环境试验研究所（（工业和信息化部电子第五研究所）（中国赛宝实验室）） | Airplane heavy landing prediction method and device, computer equipment and storage medium |
| CN112862621A (en) * | 2021-01-15 | 2021-05-28 | 广州亚美信息科技有限公司 | Vehicle insurance risk assessment method and device and computer equipment |
| CN113066315A (en) * | 2021-03-03 | 2021-07-02 | 中国民用航空飞行学院 | Scene capture analysis method for civil aircraft approach landing stage flight safety |
| CN114004292A (en) * | 2021-10-29 | 2022-02-01 | 重庆大学 | Pilot flat-floating ejector rod behavior analysis method based on flight parameter data unsupervised clustering |
| CN114647198A (en) * | 2022-03-09 | 2022-06-21 | 深圳市经纬纵横科技有限公司 | Intelligent home control method and system based on Internet of things and electronic equipment |
| CN114937317A (en) * | 2022-04-02 | 2022-08-23 | 珠海翔翼航空技术有限公司 | QAR data management method, system and storage medium for mobile terminal |
| CN114937317B (en) * | 2022-04-02 | 2024-05-10 | 珠海翔翼航空技术有限公司 | Mobile-end-oriented QAR data management method, system and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107045638B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107045638A (en) | A kind of flight safety affair analytical method based on context-aware model | |
| Napolitano et al. | Aircraft failure detection and identification using neural networks | |
| CN113486938B (en) | Multi-branch time convolution network-based re-landing analysis method and device | |
| Burnett et al. | Prediction of injuries and fatalities in aviation accidents through machine learning | |
| CN108256285A (en) | Flight path exception detecting method and system based on density peaks fast search | |
| Luo et al. | Using lda2vec topic modeling to identify latent topics in aviation safety reports | |
| CN106127407B (en) | Airplane travel scoring method and system based on multi-sensor information fusion | |
| Silagyi II et al. | Prediction of severity of aviation landing accidents using support vector machine models | |
| CN114692290A (en) | Improved FRAM (FRAM) method-based airplane landing safety quality analysis method | |
| Wu et al. | Flight situation recognition under different weather conditions | |
| CN102521901A (en) | Method for automatically estimating and realizing flight quality of pilots | |
| Fala et al. | Study on machine learning methods for general aviation flight phase identification | |
| Baugh | Predicting general aviation accidents using machine learning algorithms | |
| CN113093568A (en) | Airplane automatic driving operation simulation method based on long-time and short-time memory network | |
| CN114692786B (en) | Civil aircraft fault diagnosis method based on multi-source data fusion | |
| Lang et al. | Deep reinforcement learning-based upset recovery control for generic transport aircraft | |
| CN114003052B (en) | Fixed wing unmanned aerial vehicle longitudinal movement robust self-adaptive control method based on dynamic compensation system | |
| Jiang et al. | Research on the flight anomaly detection during take-off phase based on FOQA data | |
| Heffley et al. | Airworthiness criteria development for powered-lift aircraft: A program summary | |
| Li et al. | Study on flight safety manipulation space under complex conditions | |
| US20150269861A1 (en) | System and Method for Using Pilot Controllable Discretionary Operational Parameters to Reduce Fuel Consumption in Piloted Aircraft | |
| He et al. | Data-driven method for estimating aircraft mass from quick access recorder using aircraft dynamics and multilayer perceptron neural network | |
| Mirot | The future of unmanned aircraft systems pilot qualification | |
| Hinton | Forward-look wind-shear detection for microburst recovery | |
| Tong et al. | Application of Situation Awareness and Bow-tie Method in Flight Safety Event Reasoning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |