A method and system for risk warning and cause identification of aircraft runway excursion events

Abstract

The application discloses a kind of aircraft rush / cause runway event risk early warning and cause identification method and system, method includes: S0: by FOQA standard, SOP and QRH obtains the key parameter set and risk factor set of rush cause runway event;S1: based on the correspondence of key parameters and risk factors is constructed by Delphi method;S2: construct the rush cause runway risk prediction model based on Transformer, realize online risk identification;S3: based on hypothesis testing method obtains the most similar cause, completes risk factor identification.

Description

Technical Field

[0001] This invention belongs to the field of civil aviation safety technology and flight data application, specifically relating to a risk warning and cause identification method and system for aircraft overrun / deviate from runway events. Background Technology

[0002] Runway overrun or runway deviation during the approach and landing phase is one of the leading causes of fatal accidents. Flight quality monitoring technology based on post-flight data analysis is a crucial means of causal analysis and risk assessment for runway overrun / deviation incidents, and is of great significance to flight safety. However, current flight quality monitoring technologies are still insufficient in real-time early warning and causal analysis of runway overrun / deviation incidents. This is mainly reflected in the following aspects:

[0003] First, existing runway overrun / runway deviation detection methods mainly rely on flight data after the flight to determine whether an overrun / runway deviation event has occurred. This makes it difficult to monitor the aircraft status in real time and provide online risk warnings, preventing flight crews from identifying potential risks in advance and taking appropriate measures to prevent threats and errors from escalating into runway overrun / runway deviation events.

[0004] Secondly, there is a lack of automated methods for identifying the causes of runway overrun / runway veer-off incidents. Currently, the identification of the causes of runway overrun / runway veer-off incidents relies on manual analysis by flight safety experts combined with post-flight data to further determine the type of risk induced by the incident. Automated identification of the causes of runway overrun / runway veer-off incidents can assist flight safety experts in better understanding risk patterns and, to some extent, provide flight crews with auxiliary decision-making information to take appropriate preventive measures. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a risk warning and cause identification method and system for aircraft runway overrun / veergence incidents. For runway overrun / veergence incidents, starting from aircraft operational quality monitoring standards, flight standard operating procedures, and flight quick checklists, key parameters related to aircraft takeoff and landing are obtained, and a risk factor fault tree is established to identify the risk factors that cause runway overrun / veergence. Subsequently, the Delphi method is used to obtain the correspondence between key parameters and risk factors to achieve objective analysis of risk factors. Based on this, an anomaly prediction model is established to achieve online risk monitoring and early warning, and hypothesis testing methods are combined to complete automated cause identification.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for risk warning and cause identification of aircraft overrun / veergence incidents, the method comprising:

[0008] S0: Obtain the set of key parameters related to aircraft takeoff and landing through flight quality monitoring standards, flight standard operating procedures and flight quick reference manuals; obtain the set of risk factors that may cause runway overrun / runway deviation events by establishing a risk factor fault tree;

[0009] S1: Construct the correspondence between key parameters and risk factors based on the modified Delphi method;

[0010] S2: Construct a Transformer-based runway overrun / runway deviation risk prediction model, extract runway overrun / runway deviation event features from flight data using the Transformer-based runway overrun / runway deviation risk prediction model, and use historical flight data sequences to predict the distance of the aircraft from the runway centerline and the remaining runway distance at a future time, so as to achieve online risk identification.

[0011] S3: Based on hypothesis testing methods, obtain key parameters that show the difference between warning flights and normal flights exceeding a preset threshold. Combine the obtained key parameters with the correspondence between risk factors to output the most similar cause and complete the risk factor identification.

[0012] Preferably, in S0, the key parameter PV = {v 1 ,v 2 ,...,v d} and risk factors PR={R1,R2,...,R m Specifically, it includes:

[0013] Key parameter PV = {v 1 ,v 2 ,...,v d}, where d is the number of key parameters, x 1 The first key parameter; the key parameter PV consists of two parts, PX and PW. PX is the set of key parameters obtained through flight data, PX = {x 1 ,x 2 ,...,x b}, where b is the number of elements in PX, x 1 PX is the first element in PX; PW is the set of key parameters obtained through meteorological reports and airport pavement data, PW = {w 1 ,w 2 ,...,w c}, where c is the number of elements in PW, and d = b + c, w 1 It is the first element in PW;

[0014] Risk factor PR = {R1, R2, ..., R m}, where m is the number of risk factors, and R1 represents the first risk factor.

[0015] Preferably, in S1, the key parameter PV = {v} is established based on the modified Delphi method. 1 ,v 2 ,...,v d} and risk factors PR={R1,R2,...,R m The correspondence between} specifically includes:

[0016] S1.1: Assemble an interdisciplinary expert team and design and distribute anonymous scoring questionnaires to assess any risk factor R. j j∈[1,m] and any key parameter v i The degree of correlation of i∈[1,d], and the collection of subjective comments from experts;

[0017] S1.2: Based on the rating questionnaires obtained from each expert, for each risk factor R... j For j∈[1,m], construct the parameter correlation matrix. Where n is the number of experts. For the r-th expert, the key parameter v i and risk factors R j The correlation score was then used; subsequently, the mean μ of the evaluation index was introduced. i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j All anonymous expert ratings and mean μ i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j Subjective comments and opinions are provided again to each expert, and each expert should pay close attention to μ. i,j <3.5, ECI i,j <0.7, IQR i,j The score is >2 and the indicators mentioned in the subjective comments are used to improve and update the original score results. This process is repeated 2 to 3 times.

[0018] S1.3: Based on the final score data, if μ i,j If the value is ≥3.5, then the risk factor R is considered to be... j Subject to key parameter v i If the impact exceeds the preset threshold, the corresponding relationship between key parameters and risk factors is finally obtained, denoted as R. j =f(v h ,...,v p ), 1≤h≤p≤m.

[0019] Preferably, in step S2, the method of constructing a Transformer-based runway overrun / runway deviation risk prediction model, extracting runway overrun / runway deviation event features from flight data using the Transformer-based runway overrun / runway deviation risk prediction model, and obtaining the aircraft's deviation distance from the runway centerline and remaining runway distance at a future time from historical parameter sequences to achieve online risk identification includes:

[0020] S2.1: For the flight data sequence of the aircraft within the time interval 0 to t [x1, x2, ..., xt] t ] T The flight parameter sequence is transformed into a high-dimensional feature matrix Z with temporal dependence and multi-parameter correlation by using a 6-layer encoder layer structure. enc =[z1,z2,...,z t ] T The data is then input into a 4-layer Decoder layer for decoding. Based on the global features and historical information provided by the Encoder layer, the future runway state is gradually deduced, and finally the distance d from the runway centerline at the future time t+k is output. t+k and remaining runway distance L t+k ,in This represents the flight data sequence record of the flight at time t. Z represents the recorded value of the first flight data at time t. enc It is a high-dimensional time series feature matrix. Let b be the temporal feature vector of the flight at time t, and b′ be the dimension of the hidden layer of the model.

[0021] S2.2: Based on historical flight data, the 95th percentile of the distance from the runway centerline of all flights at time t+k is set as the maximum value of the centerline deviation threshold range. The 5th percentile is set as the minimum value of the range of deviations from the center line from the threshold. The 5th percentile of the remaining runway distance for all flights at time k is set as the remaining runway distance threshold μ. t+k ;when or When L indicates that the aircraft is at risk of veering off the runway at time t+k; when L t+k ≤μ t+k This indicates that the aircraft is at risk of overrunning the runway at time t+k; when the aircraft is at risk of veering off course or overrunning the runway, record the key parameters at the current warning time t and time t+k. We will then continue to monitor the risk status at the next moment.

[0022] Preferably, in step S3, the method for identifying risk factors by obtaining key parameters that show a difference between the warning flight and the normal flight exceeding a preset threshold based on hypothesis testing, and then outputting the most similar cause by combining the obtained key parameters with the correspondence between risk factors, includes:

[0023] When an early warning is issued at time t, the key parameters at time t will be... Key parameters of other regular flights at time t Perform hypothesis testing, where 1, 2, ..., N represent 1 to N different flights. (i∈[1,N]) represents the key parameters of the i-th flight at time t. The key parameters obtained when the difference between the warning flight and the normal flight exceeds the preset threshold are denoted as the trigger parameter R′={v g ,...,v q}, 1≤g≤q≤m; then calculate the causative parameter R′={v g ,...,v q} and each correspondence R j =f(v h ,...,v p The Jaccard similarity of j∈[1,m] is calculated; the final output is the risk factor R with the highest Jaccard similarity. j Complete the identification of risk factors.

[0024] The present invention also provides a risk warning and cause identification system for aircraft runway overrun / veergence incidents. The system is used to implement the aforementioned method and includes: an acquisition module, a construction module, an online risk identification module, and a risk factor identification module.

[0025] The acquisition module is used to obtain a set of key parameters related to aircraft takeoff and landing through flight quality monitoring standards, flight standard operating procedures and flight quick reference manuals; and to obtain a set of risk factors that could cause runway overrun / runway deviation events by establishing a risk factor fault tree.

[0026] The construction module is used to construct the correspondence between key parameters and risk factors based on the modified Delphi method;

[0027] The online risk identification module is used to construct a Transformer-based runway overrun / runway deviation risk prediction model. It uses the Transformer-based runway overrun / runway deviation risk prediction model to extract runway overrun / runway deviation event features from flight data, and uses historical flight data sequences to predict the distance of the aircraft from the runway centerline and the remaining runway distance at a future time, thereby realizing online risk identification.

[0028] The risk factor identification module is used to obtain key parameters that differ from the warning flight and the normal flight by a preset threshold based on hypothesis testing methods, and output the most similar cause by combining the obtained key parameters with the correspondence between risk factors, thus completing the risk factor identification.

[0029] Preferably, in the obtaining module, the key parameter PV = {v 1 ,v 2 ,...,v d} and risk factors PR={R1,R2,...,R m Specifically, it includes:

[0030] Key parameter PV = {v 1 ,v 2 ,...,v d}, where d is the number of key parameters, v 1 The first key parameter; the key parameter PV consists of two parts, PX and PW. PX is the set of key parameters obtained through flight data, PX = {x 1 ,x 2 ,...,x b}, where b is the number of elements in PX, x 1 PX is the first element in PX; PW is the set of key parameters obtained through meteorological reports and airport pavement data, PW = {w 1 ,w 2 ,...,w c}, where c is the number of elements in PW, and d = b + c, w 1 It is the first element in PW;

[0031] Risk factor PR = {R1, R2, ..., R m}, where m is the number of risk factors, and R1 represents the first risk factor.

[0032] Preferably, the construction module includes: a rating questionnaire unit, a matrix construction unit, and a comparison unit;

[0033] The scoring questionnaire unit is used to assemble an interdisciplinary expert team and design and distribute anonymous scoring questionnaires to assess any risk factor R. j j∈[1,m] and any key parameter v i The degree of correlation of i∈[1,d], and the collection of subjective comments from experts;

[0034] The matrix construction unit is used to, based on the obtained rating questionnaires from each expert, target each risk factor R. j For j∈[1,m], construct the parameter correlation matrix. Where n is the number of experts. For the r-th expert, the key parameter v i and risk factors R j The correlation score was then used; subsequently, the mean μ of the evaluation index was introduced. i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j All anonymous expert ratings and mean μ i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j Subjective comments and opinions are provided again to each expert, and each expert should pay close attention to μ. i,j <3.5, ECI i,j <0.7, IQR i,j The score is >2 and the indicators mentioned in the subjective comments are used to improve and update the original score results. This process is repeated 2 to 3 times.

[0035] The comparison unit is used to determine, based on the final obtained score data, if μ i,j If the risk factor R is ≥.5, then the risk factor R is considered to be ≥.5. j Subject to key parameter v i If the impact exceeds the preset threshold, the corresponding relationship between key parameters and risk factors is finally obtained, denoted as R. j =f(v h ,...,v p ), 1≤h≤p≤m.

[0036] Preferably, the online risk identification module includes: a network construction unit and a risk monitoring unit;

[0037] The network construction unit is used for the flight data sequence [x1, x2, ..., x] of the aircraft within the time interval 0 to t. t ] T The flight parameter sequence is transformed into a high-dimensional feature matrix Z with temporal dependence and multi-parameter correlation by using a 6-layer encoder layer structure. enc =[z1,z2,...,z t ] T The data is then input into a 4-layer Decoder layer for decoding. Based on the global features and historical information provided by the Encoder layer, the future runway state is gradually deduced, and finally the distance d from the runway centerline at the future time t+k is output. t+k and remaining runway distance L t+k ,in This represents the flight data sequence record of the flight at time t. Z represents the recorded value of the first flight data at time t. enc It is a high-dimensional time series feature matrix. Let b be the temporal feature vector of the flight at time t, and b′ be the dimension of the hidden layer of the model.

[0038] The risk monitoring unit is used to set the 95th percentile of the distance from the runway centerline of all flights at time t+k as the maximum value of the centerline deviation threshold range, based on historical flight data. The 5th percentile is set as the minimum value of the range of deviations from the center line from the threshold. The 5th percentile of the remaining runway distance for all flights at time k is set as the remaining runway distance threshold μ. t+k ;when or When L indicates that the aircraft is at risk of veering off the runway at time t+k; when L t+k ≤μ t+k This indicates that the aircraft is at risk of overrunning the runway at time t+k; when the aircraft is at risk of veering off course or overrunning the runway, record the key parameters at the current warning time t and time t+k. We will then continue to monitor the risk status at the next moment.

[0039] Preferably, in the risk factor identification module, key parameters that differ from normal flights exceeding a preset threshold are obtained based on hypothesis testing methods. The process of identifying the most similar trigger, combining the obtained key parameters with the correspondence between risk factors, and then outputting the most similar cause, completes the risk factor identification process as follows:

[0040] When an early warning is issued at time t, the key parameters at time t will be... Key parameters of other regular flights at time t Perform hypothesis testing, where 1, 2, ..., N represent 1 to N different flights. (i∈[1,N]) represents the key parameters of the i-th flight at time t. The key parameters obtained when the difference between the warning flight and the normal flight exceeds the preset threshold are denoted as the trigger parameter R′={v g ,...,v q}, 1≤g≤q≤m; then calculate the causative parameter R′={v g ,...,v q} and each correspondence R j =f(v h ,...,v p The Jaccard similarity of j∈[1,m] is calculated; the final output is the risk factor R with the highest Jaccard similarity. j Complete the identification of risk factors.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0042] (1) Traditional runway overrun / runway deviation detection methods rely on post-flight data to determine whether an overrun / runway deviation event has occurred, and cannot provide online risk warnings. This invention proposes a runway overrun / runway deviation risk prediction model, which constructs a high-performance deep model based on Transformer, using historical flight data sequences as input to predict the distance the aircraft deviates from the runway centerline and the remaining runway distance at future moments. By providing risk warning information a period of time before an overrun / runway deviation event occurs, real-time monitoring of the aircraft's flight status and runway overrun / runway deviation risk warnings are achieved, and flight crews are given time to develop appropriate strategies to prevent runway overrun / runway deviation events from occurring.

[0043] (2) Traditional runway overrun / runway deviation incident causal analysis methods require flight safety experts to manually analyze the precipitating risk types of runway overrun / runway deviation incidents using post-flight data, which is costly. This invention proposes a method for constructing the correspondence between risk factors and key parameters. Based on the Delphi method and integrating multivariate expert experience, it obtains the influence relationships and coupling mechanisms of risk factors and key parameters, and further achieves automated precipitating factor identification by combining anomaly prediction models and hypothesis testing methods. This provides flight safety experts with information on the risk precipitating factors of runway overrun / runway deviation incidents, assisting them in understanding risk patterns and providing reference information during post-flight causal analysis. It also provides flight crews with relevant abnormal parameters, assisting them to take appropriate preventative measures to avoid runway overrun / runway deviation incidents. Attached Figure Description

[0044] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the main steps of embodiments S0-S3 of the present invention;

[0046] Figure 2 This is a schematic diagram of the risk warning and cause identification method for aircraft overrun / deviation from the runway according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1

[0050] like Figure 1 , Figure 2 As shown, the present invention provides a method for risk warning and cause identification of aircraft overrun / runway deviation incidents, comprising the following steps:

[0051] S0: First, refer to the Flight Operation Quality Assurance (FOQA) standards, Standard Operating Procedures (SOPs), and Quick Reference Handbook (QRH) to comprehensively review the key parameters related to takeoff and landing, PV = {v 1 ,v 2 ,...,v d}. Where d is the number of key parameters, v 1 This is the first key parameter. The key parameter PV consists of two parts: PX and PW. PX is the set of key parameters obtained through flight data, PX = {x}. 1 ,x 2 ,...,x b}, where b is the number of elements in PX, x 1 PX is the first element in PX; PW is a set of key parameters obtained through meteorological reports, airport pavement data, etc., PW = {w 1 ,w 2 ,...,w c}, where c is the number of elements in PW, and d = b + c, w 1 This is the first element in the PW (Problem Tree). Subsequently, for events of different stages (takeoff and landing) and different types (runaway and overrun), corresponding risk factor fault trees T are established to further accurately and comprehensively identify the risk factors that cause the aircraft to overrun / vary the runway. The fault tree is constructed based on fault event information related to the aircraft's runway overrun / vary, containing multiple fault events related to the runway overrun / vary and the causal relationships between these events. Specifically, the fault tree is structured according to the top event T, the intermediate events corresponding to the top event PI = {I1, I2, ..., I...} j The sub-events PB = {B1, B2, ..., B} corresponding to the intermediate events. kThe events are distributed hierarchically. PI represents the set of intermediate events corresponding to the top event T, j is the number of elements in PI, and I1 is the first element in PI; PB represents the set of sub-events corresponding to the intermediate event set, k is the number of elements in PB, and B1 is the first element in PI. Events at different levels have causal relationships: the top event T is the effect, and the corresponding intermediate event PI is the cause of the top event; the intermediate event PI is the effect, and the corresponding sub-event PB is the cause of the intermediate event. All basic events in the fault tree are defined as risk factors that cause an aircraft to overrun / deviate from the runway: PR = {R1, R2, ..., R...} m}, where the basic event is the lowest level event in the fault tree, m is the number of risk factors, and R1 represents the first risk factor.

[0052] S1: Establishing key parameters PV={v} based on the modified Delphi method 1 ,v 2 ,...,v d} and risk factors PR={R1,R2,...,R m The correspondence between these parameters is used for the accurate identification and analysis of the risk factors that cause runway over / off-track. The specific steps are as follows:

[0053] (S1.1) Assessment of the degree of parameter-factor correlation based on Subject Matter Expert (SME)

[0054] Assemble an interdisciplinary expert panel (15-20 people), including senior pilots (approximately 40%), aerospace engineering experts (approximately 30%), flight data analysts (approximately 20%), and human factors experts (approximately 10%). Design and distribute anonymous rating questionnaires to assess any risk factor R. j (j∈[1,m]) and any key parameter v i The degree of correlation (i∈[1,d]) and the collection of subjective comments from experts. The rating levels include "1-irrelevant", "2-weakly irrelevant", "3-moderately relevant", "4-relatively relevant", and "5-strongly relevant".

[0055] (S1.2) Update of SME-based scoring data

[0056] Based on the rating questionnaires obtained from each expert in (S1.1), for each risk factor R j (j∈[1,m]), construct the parameter correlation matrix Where n is the number of experts. For the r-th expert, the key parameter v i and risk factors R j The degree of correlation was scored. Then, an evaluation metric was introduced: mean μ. i,jConsensus Index (ECI) i,j Interquartile Range (IQR) i,j ,in For the key parameter v i and risk factors R j The mean score of the degree of correlation, if μ i,j A value ≥3.5 indicates that the key parameter v i and risk factors R j Strong correlation; ECI i,j =1-σ i,j / μ i,j , For the standard deviation of the score, ECI i,j Used to describe the concentration of expert opinions, if ECI i,j A score ≥0.7 indicates a high degree of consensus among expert ratings; IQR i,j for The interquartile range, if IQR i,j A score of ≥2 indicates a significant discrepancy in the ratings.

[0057] All anonymous expert ratings and mean μ i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j Subjective comments and opinions are provided again to each expert, and each expert should pay close attention to μ. i,j <3.5, ECI i,j <0.7, IQR i,j The scoring system is based on a score of ≥2 and the indicators mentioned in the subjective comments. The original scoring results are then improved and updated. This process is repeated 2 to 3 times.

[0058] (S1.3) Confirmation of correspondence based on scoring data

[0059] Based on the final score data, if μ i,j If the value is ≥3.5, then the risk factor R is considered to be... j Subject to key parameter v i The impact is significant, ultimately yielding the correspondence between key parameters and risk factors, denoted as R. j =f(v h ,...,v p ), 1≤h≤p≤m.

[0060] S2: Construct a Transformer-based runway overrun / runway deviation risk prediction model. This model extracts runway overrun / runway deviation event features from flight data and obtains the aircraft's deviation distance from the runway centerline and remaining runway distance at a future time from historical parameter sequences. The specific steps are as follows:

[0061] (S2.1) Establish a Transformer network

[0062] For the flight data sequence of the aircraft in the time interval 0 to t [x1,x2,...,x] t ] T The flight parameter sequence is transformed into a high-dimensional feature matrix Z with temporal dependence and multi-parameter correlation by using a 6-layer encoder layer structure. enc =[z1,z2,...,z t ] T The data is then fed into a four-layer Decoder layer for decoding. Based on the global features and historical information provided by the Encoder layer, the runway state at future time steps are deduced step by step, and finally the distance d from the runway centerline at the future time t+k is output. t+k and remaining runway distance L t+k .in This represents the flight data sequence record of the flight at time t. Z represents the recorded value of the first flight data at time t. enc It is a high-dimensional time series feature matrix. Let b be the temporal feature vector of the flight at time t, b′ be the dimension of the hidden layer of the model, and d be the time series feature vector of the flight at time t. t+k L represents the distance from the runway centerline predicted at time t+k from time t in the future. t+k This represents the remaining runway distance predicted at time t+k from time t.

[0063] Specifically, the distance d from the runway centerline t+k The angle δ between the line connecting the aircraft and the center of the localizer and the runway centerline t+k And the distance between the aircraft and the localizer is determined, d t+k =(L0+SL′) t+k )·tanδ t+k Remaining runway distance L t+k L is calculated by subtracting the distance between the aircraft and the runway threshold from the total runway length. t+k =SL′ t+k Among them, δ t+k Let L' be the angle between the line connecting the aircraft and the center of the localizer at time t+k and the runway centerline, L0 be the distance between the localizer and the end of the runway, S be the runway length, and L' be the angle between the localizer and the end of the runway. t+k Let d be the vertical distance between the aircraft and the entrance at time t+k. t+k >0 represents the maximum distance by which the aircraft's center of gravity deviates from the left side of the runway centerline (d during takeoff, from the takeoff direction; during landing, from the landing direction). t+k <0 represents the maximum distance the aircraft's nose deviates from the right side of the runway centerline (as above).

[0064] A Transformer-based runway overrun / runway deviation risk prediction model is defined as follows:

[0065]

[0066] Among them, This represents the flight data sequence record of the flight at time t. Let f be the time-series feature vector of the flight at time t. In the formula... dec (·) indicates a Decoder layer processing operation, f enc (·) indicates that the Encoder layer is processing the operation, f tansformer (·) represents the computation process of the Transformer network, with the input being the flight data sequence of the aircraft within the time interval 0 to t [x1, x2, ..., xt]. t ] T Obtain the predicted distance d from the runway centerline at the future time t+k. t+k and remaining runway distance L t+k .

[0067] (S2.2) Runway overrun / deviation risk monitoring

[0068] Based on historical flight data, the 95th percentile of the distance from the runway centerline of all flights at time t+k is set as the maximum value of the centerline deviation threshold range. The 5th percentile is set as the minimum value of the range of deviations from the center line from the threshold. The 5th percentile of the remaining runway distance for all flights at time k is set as the remaining runway distance threshold μ. t+k .

[0069] when or When L indicates that the aircraft is at risk of veering off the runway at time t+k; when L t+k ≤μ t+k This indicates that the aircraft is at risk of overrunning the runway at time t+k. When the aircraft is at risk of veering off course or overrunning the runway, record the key parameters at the current warning time t and time t+k. We will then continue to monitor the risk status at the next moment.

[0070] S3: Based on the output of the runway overrun / runway deviation risk prediction model, the event causes are traced using hypothesis testing for online risk detection. The specific steps are as follows:

[0071] When an early warning is issued at time t, the key parameters at time t will be... Key parameters of other regular flights at time t Hypothesis testing was conducted to obtain key parameters showing significant differences between alerted flights and normal flights. Here, 1,2,...,N represent 1 to N different flights. (i∈[1,N]) represents the key parameters of the i-th flight at time t.

[0072] Using QQ graph to verify parameters Does it conform to a normal distribution? If the parameter The variance is σ 2 A normal distribution N(μ,σ) with mean μ. 2 If the value is not significant, then the Z-test is used for significance testing. The Z-test is a parametric test method. First, the Z-value is calculated according to the formula. in σ is the sample mean, μ is the population mean, σ is the population standard deviation, and n is the sample size. The critical value Z is obtained by looking up the standard normal distribution table at a significance level of α = 0.05. α If |Z|>Z α This indicates a significant difference between the data and normal data.

[0073] If parameter If the data does not conform to a normal distribution, a box plot is used for significance testing. The box plot test is a non-parametric method. First, the lower quartile (Q1), upper quartile (Q3), and median of the normal data are calculated, and a box plot is drawn. Then, the interquartile range (IQR) is calculated as IQR = Q3 - Q1, determining the lower bound for outliers as Q1 - 1.5IQR and the upper bound as Q3 + 1.5IQR. Finally, if the data to be tested... or This indicates that there is a significant difference between the data and the normal data.

[0074] Through parameter testing (|Z|>Z) α ) or nonparametric test ( or The key parameters that show significant differences between the warning flights and the normal flights are output, denoted as the causative parameters R′={v g ,...,v q}, 1≤g≤q≤m. Finally, combining the key parameter PV obtained from S1, we get PV={v 1 ,v 2 ,...,v d} and risk factors PR={R1,R2,...,R m The correspondence between} is used to determine the most similar triggering factor for overrun / runway deviation. Specifically, the triggering factor parameter R′={v g ,...,v q} and each correspondence R j =f(v h ,...,vp The Jaccard similarity of (j∈[1,m]) is calculated using the following formula:

[0075]

[0076] Wherein, J(R′,R j The larger the value of ), the more it indicates that the causative parameter R′={v g ,...,v q} and the risk factor R j The corresponding key parameter {v h ,...,v p The more similar the two factors are, the better. The final output is the risk factor R with the highest Jaccard similarity. j Complete the identification of risk factors.

[0077] Example 2

[0078] The present invention also provides a risk warning and cause identification system for aircraft runway overrun / deviation incidents. The system is used to implement the method described in Embodiment 1. The system includes: an acquisition module, a construction module, an online risk identification module, and a risk factor identification module.

[0079] The module is used to obtain a set of key parameters related to aircraft takeoff and landing through flight quality monitoring standards, flight standard operating procedures and flight quick reference manuals; and to obtain a set of risk factors that could cause runway overrun / runway deviation events by building a risk factor fault tree.

[0080] A module for constructing the correspondence between key parameters and risk factors based on the modified Delphi method;

[0081] The online risk identification module is used to build a Transformer-based runway overrun / runway deviation risk prediction model. It uses the Transformer-based runway overrun / runway deviation risk prediction model to extract runway overrun / runway deviation event features from flight data, and uses historical flight data sequences to predict the distance of the aircraft from the runway centerline and the remaining runway distance at a future time, thereby realizing online risk identification.

[0082] The risk factor identification module is used to obtain key parameters that differ from the pre-set threshold between the flight warning and the normal flight based on hypothesis testing methods. It then combines the obtained key parameters with the correspondence between risk factors to output the most similar cause, thus completing the risk factor identification.

[0083] In this embodiment, the key parameter PV = {v} in the acquisition module 1 ,v 2 ,...,v d} and risk factors PR={R1,R2,...,R m Specifically, it includes:

[0084] Key parameter PV = {v 1 ,v 2 ,...,v d}, where d is the number of key parameters, v 1 The first key parameter; the key parameter PV consists of two parts, PX and PW. PX is the set of key parameters obtained through flight data, PX = {x 1 ,x 2 ,...,x b}, where b is the number of elements in PX, x 1 PX is the first element in PX; PW is the set of key parameters obtained through meteorological reports and airport pavement data, PW = {w 1 ,w 2 ,...,w c}, where c is the number of elements in PW, and d = b + c, w 1 It is the first element in PW;

[0085] Risk factor PR = {R1, R2, ..., R m}, where m is the number of risk factors, and R1 represents the first risk factor.

[0086] In this embodiment, the construction module includes: a rating questionnaire unit, a matrix construction unit, and a comparison unit;

[0087] The scoring questionnaire unit is used to assemble an interdisciplinary expert team and design and distribute anonymous scoring questionnaires to assess any risk factor R. j j∈[1,m] and any key parameter v i The degree of correlation of i∈[1,d], and the collection of subjective comments from experts;

[0088] Matrix construction unit, used to analyze each risk factor R based on the obtained expert rating questionnaire. j For j∈[1,m], construct the parameter correlation matrix. Where n is the number of experts. For the r-th expert, the key parameter v i and risk factors R j The correlation score was then used; subsequently, the mean μ of the evaluation index was introduced. i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j All anonymous expert ratings and mean μ i,j Consensus Index (ECI) i,j Interquartile Range (IQR) i,j Subjective comments and opinions are provided again to each expert, and each expert should pay close attention to μ.i,j <3.5, ECI i,j <0.7, IQR i,j The score is >2 and the indicators mentioned in the subjective comments are used to improve and update the original score results. This process is repeated 2 to 3 times.

[0089] The comparison unit is used to determine the score based on the final score data, if μ i,j If the value is ≥3.5, then the risk factor R is considered to be... j Subject to key parameter v i If the impact exceeds the preset threshold, the corresponding relationship between key parameters and risk factors is finally obtained, denoted as R. j =f(v h ,...,v p ), 1≤h≤p≤m.

[0090] In this embodiment, the online risk identification module includes: a network construction unit and a risk monitoring unit;

[0091] Network building blocks are used to construct flight data sequences [x1, x2, ..., xt] for aircraft within the time interval 0 to t. t ] T The flight parameter sequence is transformed into a high-dimensional feature matrix Z with temporal dependence and multi-parameter correlation by using a 6-layer encoder layer structure. enc =[z1,z2,...,z t ] T The data is then fed into a four-layer Decoder layer for decoding. Based on the global features and historical information provided by the Encoder layer, the runway state at future time steps are deduced step by step, and finally the distance d from the runway centerline at the future time t+k is output. t+k and remaining runway distance L t+k .in This represents the flight data sequence record of the flight at time t. Z represents the recorded value of the first flight data at time t. enc It is a high-dimensional time series feature matrix. Let b be the temporal feature vector of the flight at time t, and b′ be the dimension of the hidden layer of the model.

[0092] The risk monitoring unit, based on historical flight data, sets the 95th percentile of the distance from the runway centerline of all flights at time t+k as the maximum value of the centerline deviation threshold range. The 5th percentile is set as the minimum value of the range of deviations from the center line from the threshold. The 5th percentile of the remaining runway distance for all flights at time k is set as the remaining runway distance threshold μ. t+k ;when or When L indicates that the aircraft is at risk of veering off the runway at time t+k; when L t+k ≤μ t+k This indicates that the aircraft is at risk of overrunning the runway at time t+k; when the aircraft is at risk of veering off course or overrunning the runway, record the key parameters at the current warning time t and time t+k. We will then continue to monitor the risk status at the next moment.

[0093] In this embodiment, the risk factor identification module obtains key parameters where the difference between the warning flight and the normal flight exceeds a preset threshold based on hypothesis testing. It then combines the obtained key parameters with the correspondence between risk factors to output the most similar trigger, thus completing the risk factor identification process.

[0094] When an early warning is issued at time t, the key parameters at time t will be... Key parameters of other regular flights at time t Perform hypothesis testing, where 1, 2, ..., N represent 1 to N different flights. (i∈[1,N]) represents the key parameters of the i-th flight at time t. The key parameters obtained when the difference between the warning flight and the normal flight exceeds the preset threshold are denoted as the trigger parameter R′={v g ,...,v q}, 1≤g≤q≤m; then calculate the causative parameter R′={v g ,...,v q} and each correspondence R j =f(v h ,...,v p The Jaccard similarity of j∈[1,m] is calculated; the final output is the risk factor R with the highest Jaccard similarity. j Complete the identification of risk factors.

[0095] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for risk warning and cause identification of aircraft overrun / veergence incidents, characterized in that, The method includes: S0: Obtain the set of key parameters related to aircraft takeoff and landing through flight quality monitoring standards, flight standard operating procedures and flight quick reference manuals; obtain the set of risk factors that may cause runway overrun / runway deviation events by establishing a risk factor fault tree; S1: Construct the correspondence between key parameters and risk factors based on the modified Delphi method; S2: Construct a Transformer-based runway overrun / runway deviation risk prediction model, extract runway overrun / runway deviation event features from flight data using the Transformer-based runway overrun / runway deviation risk prediction model, and use historical flight data sequences to predict the distance of the aircraft from the runway centerline and the remaining runway distance at a future time, so as to achieve online risk identification. S3: Based on hypothesis testing methods, obtain key parameters that show the difference between warning flights and normal flights exceeding a preset threshold. Combine the obtained key parameters with the correspondence between risk factors to output the most similar cause and complete the risk factor identification. In S1, key parameters are established based on the modified Delphi method. Risk factors The correspondence specifically includes: S1.1: Assemble an interdisciplinary expert team and design and distribute anonymous rating questionnaires to assess any risk factor. , With any key parameter , The degree of correlation, and the collection of subjective comments from experts; S1.2: Based on the rating questionnaires obtained from each expert, for each risk factor... , Construct the parameter correlation matrix ,in, For the number of experts, For the first An expert on key parameters and risk factors The correlation score was then used; subsequently, the mean of the evaluation index was introduced. Consensus Index Interquartile range All anonymous expert ratings and averages Consensus Index Interquartile range Subjective comments and opinions are also provided to each expert again, and each expert needs to pay close attention to them. <3.5、 <0.7、 The score is >2 and the indicators mentioned in the subjective comments are used to improve and update the original score results. This process is repeated 2 to 3 times. S1.3: Based on the final scoring data, if > A score of 3.5 indicates that the risk factors are considered to be... Subject to key parameters If the impact exceeds the preset threshold, the corresponding relationship between key parameters and risk factors is finally obtained, denoted as... , .

2. The method according to claim 1, characterized in that, In S0, the key parameters and risk factors Specifically, it includes: Key parameters , For the number of key parameters, This is the first key parameter; key parameter include and Two parts, A set of key parameters obtained through flight data. , for The number of elements in for The first element in; This is a set of key parameters obtained through meteorological reports and airport pavement data. , for The number of elements in, and have , for The first element; Risk factors , The number of risk factors, This represents the first risk factor.

3. The method according to claim 1, characterized in that, In step S2, a Transformer-based runway overrun / runway deviation risk prediction model is constructed. This model is used to extract runway overrun / runway deviation event features from flight data, and the aircraft's deviation distance from the runway centerline and remaining runway distance at a future time are obtained from historical parameter sequences. The method for achieving online risk identification includes: S2.1: For 0~ Flight data sequence of aircraft within a time period A 6-layer encoder structure is used to transform the flight parameter sequence into a high-dimensional feature matrix with temporal dependencies and multi-parameter correlations. The data is then fed into four Decoder layers for decoding. Based on the global features and historical information provided by the Encoder layers, the future runway state is gradually deduced, ultimately outputting the future runway state. Distance from the runway centerline at any given moment and remaining runway distance ,in Representative flights at Flight data sequence records at each moment, Representing the first flight data in The recorded value at any given moment. It is a high-dimensional time series feature matrix. For flights The temporal feature vector at time step, The hidden layer dimension of the model; S2.2: Based on historical flight data, all flights will be... The 95th percentile of the distance from the runway centerline at any given time is set as the maximum value of the threshold range for distance from the centerline. The 5th percentile is set as the minimum value of the range of deviations from the center line from the threshold. ; All flights at The 5th percentile of the remaining runway distance at time t is set as the remaining runway distance threshold. ;when > or < At that time, it indicates that the aircraft was There is always a risk of veering off the track; when < This indicates that the aircraft is There is always a risk of the aircraft overrunning the runway; when there is a risk of the aircraft veering off course or overrunning the runway, record the current warning moment. and Key parameters of time Then, continue to monitor the risk status at the next moment.

4. The method according to claim 1, characterized in that, In step S3, key parameters that differ from normal flights exceeding a preset threshold are obtained based on hypothesis testing. The most similar trigger is then output by combining the obtained key parameters with the correspondence between risk factors. The method for identifying risk factors includes: when When issuing warnings, Key parameters of time Other regular flights Key parameters of time Perform hypothesis testing, where, represent Different flights, Representing the One flight Key parameters at specific times, specifically those indicating discrepancies between warning flights and regular flights exceeding preset thresholds, are denoted as trigger parameters. , Subsequently, the causative parameters were calculated. With each correspondence , The Jaccard similarity; the final output is the risk factor with the highest Jaccard similarity. Complete the identification of risk factors.

5. A risk warning and cause identification system for aircraft runway overrun / deviation incidents, said system being used to implement the method described in any one of claims 1-4, characterized in that, The system includes: an acquisition module, a construction module, an online risk identification module, and a risk factor identification module; The acquisition module is used to obtain a set of key parameters related to aircraft takeoff and landing through flight quality monitoring standards, flight standard operating procedures and flight quick reference manuals; and to obtain a set of risk factors that could cause runway overrun / runway deviation events by establishing a risk factor fault tree. The construction module is used to construct the correspondence between key parameters and risk factors based on the modified Delphi method; The online risk identification module is used to construct a Transformer-based runway overrun / runway deviation risk prediction model. It uses the Transformer-based runway overrun / runway deviation risk prediction model to extract runway overrun / runway deviation event features from flight data, and uses historical flight data sequences to predict the distance of the aircraft from the runway centerline and the remaining runway distance at a future time, thereby realizing online risk identification. The risk factor identification module is used to obtain key parameters that differ from the warning flight and the normal flight by a preset threshold based on hypothesis testing methods, and output the most similar cause by combining the obtained key parameters with the correspondence between risk factors, thus completing the risk factor identification.

6. The system according to claim 5, characterized in that, In the acquisition module, key parameters and risk factors Specifically, it includes: Key parameters , For the number of key parameters, This is the first key parameter; key parameter include and Two parts, A set of key parameters obtained through flight data. , for The number of elements in for The first element in; This is a set of key parameters obtained through meteorological reports and airport pavement data. , for The number of elements in, and have , for The first element; Risk factors , The number of risk factors, This represents the first risk factor.

7. The system according to claim 5, characterized in that, The construction module includes: a rating questionnaire unit, a matrix construction unit, and a comparison unit; The scoring questionnaire unit is used to assemble an interdisciplinary expert team and design and distribute anonymous scoring questionnaires to assess any risk factor. , With any key parameter , The degree of correlation, and the collection of subjective comments from experts; The matrix construction unit is used to analyze each risk factor based on the obtained rating questionnaire from each expert. , Construct the parameter correlation matrix ,in, For the number of experts, For the first An expert on key parameters and risk factors The correlation score was then used; subsequently, the mean of the evaluation index was introduced. Consensus Index Interquartile range All anonymous expert ratings and averages Consensus Index Interquartile range Subjective comments and opinions are also provided to each expert again, and each expert needs to pay close attention to them. <3.5、 <0.7、 The score is >2 and the indicators mentioned in the subjective comments are used to improve and update the original score results. This process is repeated 2 to 3 times. The comparison unit is used to, based on the final obtained score data, if > A score of 3.5 indicates that the risk factors are considered to be... Subject to key parameters If the impact exceeds the preset threshold, the corresponding relationship between key parameters and risk factors is finally obtained, denoted as... , .

8. The system according to claim 5, characterized in that, The online risk identification module includes: a network construction unit and a risk monitoring unit; The network construction unit is used for 0~ Flight data sequence of aircraft within a time period A 6-layer encoder structure is used to transform the flight parameter sequence into a high-dimensional feature matrix with temporal dependencies and multi-parameter correlations. The data is then fed into a four-layer Decoder layer for decoding. Based on the global features and historical information provided by the Encoder layer, the future runway state is gradually deduced, ultimately outputting the future... Distance from the runway centerline at any given moment and remaining runway distance ,in Representative flights at Flight data sequence records at each moment, Representing the first flight data in The recorded value at any given moment. It is a high-dimensional time series feature matrix. For flights The temporal feature vector at time step, The hidden layer dimension of the model; The risk monitoring unit is used to monitor all flights based on historical flight data. The 95th percentile of the distance from the runway centerline at any given time is set as the maximum value of the threshold range for distance from the centerline. The 5th percentile is set as the minimum value of the range of deviations from the center line from the threshold. ; All flights at The 5th percentile of the remaining runway distance at time t is set as the remaining runway distance threshold. ;when > or < At that time, it indicates that the aircraft was There is always a risk of veering off the track; when < This indicates that the aircraft is There is always a risk of the aircraft overrunning the runway; when there is a risk of the aircraft veering off course or overrunning the runway, record the current warning moment. and Key parameters of time Then, continue to monitor the risk status at the next moment.

9. The system according to claim 5, characterized in that, In the risk factor identification module, key parameters are obtained based on hypothesis testing methods to show that the difference between the warning flight and the normal flight exceeds a preset threshold. The process of identifying the most similar cause is then output by combining the obtained key parameters with the correspondence between risk factors. The risk factor identification process includes: when When issuing warnings, Key parameters of time Other regular flights Key parameters of time Perform hypothesis testing, where, represent Different flights, Representing the One flight Key parameters at specific times, specifically those indicating discrepancies between warning flights and regular flights exceeding preset thresholds, are denoted as trigger parameters. , Subsequently, the causative parameters were calculated. With each correspondence , The Jaccard similarity; the final output is the risk factor with the highest Jaccard similarity. Complete the identification of risk factors.

Images