CN111008669A - Deep learning-based heavy landing prediction method - Google Patents

Deep learning-based heavy landing prediction method Download PDF

Info

Publication number
CN111008669A
CN111008669A CN201911255415.2A CN201911255415A CN111008669A CN 111008669 A CN111008669 A CN 111008669A CN 201911255415 A CN201911255415 A CN 201911255415A CN 111008669 A CN111008669 A CN 111008669A
Authority
CN
China
Prior art keywords
landing
flight
model
data
heavy
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.)
Pending
Application number
CN201911255415.2A
Other languages
Chinese (zh)
Inventor
诸彤宇
陆禹成
佟治威
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN201911255415.2A priority Critical patent/CN111008669A/en
Publication of CN111008669A publication Critical patent/CN111008669A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a deep learning-based heavy landing prediction method, which is used for solving the problem that whether an aircraft can generate a heavy landing danger or not according to parameters of an aircraft flight phase. The method mainly comprises a construction method of an autoregressive landing model based on deep learning and a method for detecting the heavy landing by using the autoregressive landing model, wherein the method comprises the following two parts: (1) autoregressive landing model: the model is trained by utilizing a large amount of sample data of normal landing flights based on a deep learning model and is used for capturing the airplane landing state conversion mode. (2) The calculation method comprises the following steps: the time series symbolic representation and landing state dictionary are used to convert the aircraft landing multivariate time series into a symbolic representation sequence using flight process data, such as data taken from a QAR (Quick Access Recorder) as input. And fitting a symbolic representation sequence of the flight to be predicted by using the pre-trained autoregressive landing model, and giving a prediction result of the occurrence probability of the heavy landing danger.

Description

Deep learning-based heavy landing prediction method
Technical Field
The invention relates to the field of civil aviation, in particular to a landing quality prediction method based on flight data, and discloses a method for intelligently detecting landing risks under the condition of big data.
Background
In the civil aviation field, the landing phase is the phase in which the most accidents occur. Studies have shown that 47% of accidents occur during the landing phase. The safety is the life line of the civil aviation industry, and the guarantee of the safety of the civil aviation flight operation is the premise of carrying out all specific civil aviation activities. The landing stage is a high-risk stage of flight operation, the accident rate accounts for 20% of the full-flight stage, the accident death rate accounts for 23% of the full-flight stage, and the accident and unsafe events are obviously higher in incidence rate and death rate than other flight stages. It can be seen that the aircraft landing risk is the largest risk source in civil aviation operation safety. Therefore, the research and implementation of the airplane landing safety risk analysis system have important research significance on civil aviation operation safety.
Unsafe events in the landing phase of an aircraft are mainly: heavy landing, rushing/deviating from the runway, etc. Among them, heavy landing is a landing safety accident that may cause serious damage to the aircraft structure itself and even bring danger to passengers. If the aircraft flight data can be used for predicting whether the aircraft can be landed again or not during landing, a reference index can be given to a pilot to help the pilot to make a decision during landing, so that the occurrence of the landing accident can be reduced.
The QAR is a recorder system which can conveniently and rapidly acquire the operation data of the airplane and can record a plurality of parameters such as the position, the movement, the manipulation, the alarm and the like of the whole flight stage of the airplane. By using the QAR data, the flight parameters of the airplane in the whole flight process can be conveniently acquired.
The existing heavy landing prediction method is mainly based on a statistical machine learning algorithm and a characteristic engineering technology. The disadvantages of this approach are mainly twofold. Firstly, the aircraft landing process is very complex, the prediction effect of the heavy landing depends on the feature structure to a great extent by using a method of expert extraction feature engineering, and the complex landing process is very difficult for the feature engineering. Secondly, the number of heavy landing samples is very small, so that when the model is trained, the situation that positive and negative samples are unbalanced or the number of training samples is small can be met, and the model effect is poor.
Disclosure of Invention
The technical problems solved by the invention are as follows: aiming at the possible occurrence of heavy landing accidents during the landing of the airplane, the heavy landing prediction method based on deep learning is provided, and the precision of heavy landing prediction can be obviously improved.
The technical scheme of the invention is as follows: a deep learning-based heavy landing prediction method comprises the following implementation steps:
(1) acquiring mass QAR flight data, intercepting flight landing stage data, and selecting parameters which are relatively critical to the landing stage. Dividing a multivariate time sequence consisting of an array of a plurality of characteristic landing stages to form a plurality of landing state points, wherein each landing state point is represented as a vector by each selected characteristic;
(2) defining landing state points as recorded values of each moment in an aircraft landing sequence, clustering all the landing state points of all flight samples acquired from a QAR by using a Kmeans method, and enabling the Kmeans clustering to correspond all the landing state points to different landing state categories according to the values of the landing state points; generating a flight landing state dictionary by using a centroid generated by a Kmeans clustering algorithm, mapping each landing state point in a flight to a symbolic representation in the flight landing state dictionary, wherein each flight is composed of a plurality of landing state points, and obtaining a symbolic sequence representation by each flight through mapping;
(3) pre-training is performed based on a large number of landing multivariate time series tokenized sequence data using an autoregressive model based on deep learning. The pre-training mode is that at any time t, the landing state at the previous t-1 times is used for predicting the landing state at the current time. The pre-trained Model is an Aircraft Landing Model (ALM);
(4) and (3) generating a symbolic representation by using a heavy landing training sample and using a data preprocessing mode in the step (2), and finely adjusting on the basis of the model trained in the step (3) to obtain a heavy landing prediction model.
The step (1) is specifically realized by the following steps:
(11) preparing data: the QAR data is extracted and the data between 250 feet from ground height to 0 feet from ground is intercepted as aircraft landing data. Wherein data for a time period 5 seconds before landing occurs is selected as training data.
(12) Feature extraction: based on the recorded data in the QAR data, and the relevant literature, some features that are most important for landing are selected, and are divided into three groups of features, namely, aircraft state class features, environment class features, and driver operation features. Selected characteristic fields and meanings are shown in the following table.
Figure BDA0002310111770000021
Figure BDA0002310111770000031
(13) Marking the vertical load: and marking whether each flight has the heavy landing according to the vertical load field in the QAR data. And recording the vertical load once every 0.125 second, selecting data within 5s before and after the landing, recording 80 vertical loads, and selecting the maximum record value as the final landing vertical load of the flight. And marking the heavy landing according to the size of the landing vertical load, and judging that the heavy landing occurs when the size is larger than 1.5.
The step (2) is specifically realized by the following steps:
(21) each flight is represented as a multivariate time series according to the characteristics selected in step 1. Each second is taken as a landing state, and each landing state can be regarded as a vector consisting of the values of the features selected in step 1 at that time. The landing status vectors for all flights are aggregated.
(22) And clustering all the landing state vectors by using a Kmeans clustering algorithm to generate 900 clustering centers, wherein each clustering center can be expressed into a vector by using the value of each characteristic in the clustering center. These different cluster centers are represented by different symbols.
(23) For each landing state vector, a most suitable centroid is assigned as a symbolic representation of the landing state vector. Each landing state vector of each flight is converted into a symbol corresponding to the centroid, and the landing multivariate time sequence of each flight can be converted into a landing symbol sequence.
The step (3) is specifically realized by the following steps:
(31) the input is a sequence representation of landing symbols for each flight, each of which is mapped to a high-dimensional vector using the embedding layer. The high-dimensional vector corresponding to each landing state symbol is initialized randomly at first and is automatically updated in the training process of the deep learning network. And altitude embedding is introduced, each altitude range corresponds to one altitude embedding, and the altitude embedding is updated along with the training of the deep learning model. And adding the landing state symbol embedding and the altitude embedding as final input of the deep learning model.
(32) And (3) building a one-way LSTM autoregressive model, and predicting the landing state symbol at the current moment by using the landing state symbol at the previous moment at each time step of the model. The LSTM model is pre-trained using a large number of landing symbol sequences. The calculation mode of the LSTM in the training process is as follows:
ft＝σ(Wf·[ht-1,xt]+bf)
it＝σ(Wi·[ht-1,xt]+bi)
Figure BDA0002310111770000032
Figure BDA0002310111770000033
ot＝σ(Wo·[ht-1,xt]+b0)
ht＝ot*tanh(Ct)
wherein, Wf，Wi，WC，Wo，bf，bi，bC，b0For trainable parameters in the model, xtThe symbolic representation corresponding to the landing state at the time t passes through the embedded layer to generate vector representation, and the loss functions used by the other symbols as intermediate variables generated by the LSTM in the calculation process are as follows:
Figure BDA0002310111770000041
wherein T represents the number of landing state points of flight landing data, W, b are parameters of model trainability, and xiThe symbol is corresponding to the landing state at the moment i.
The step (4) is specifically realized by the following steps:
(41) the re-landing prediction dataset is processed into a landing symbolic representation using the same method, and the trained LSTM is fine-tuned. The loss function used is:
Figure BDA0002310111770000042
wherein N represents the total sample size, yiWhether the ith sample is subjected to heavy landing or not is represented, the value of the heavy landing is 1, the value of the non-heavy landing is 0, and p is the probability value for predicting the heavy landing of the last output of the LSTM model;
(42) the refined model may be used for heavy landing prediction. QAR data before landing of a flight is given, preprocessed and input into the finely adjusted model, and a prediction result can be obtained.
Compared with the prior art, the invention has the advantages that: aiming at the possible occurrence of the heavy landing accident during the aircraft landing, the method utilizes flight QAR data to predict the occurrence of the heavy landing accident before the aircraft landing based on a deep learning model, and compared with a general heavy landing prediction method, the method can obviously improve the precision of the heavy landing prediction. Through experiments, the accuracy rate of the heavy landing prediction is 0.618 by using a mode of combining a statistical machine learning method and characteristic engineering, the accuracy rate of the heavy landing prediction by using the method provided by the invention is 0.761, and compared with the accuracy rate of 0.618 of other methods, the accuracy rate effect of the heavy landing prediction by using the method provided by the invention is improved by 14%.
Drawings
FIG. 1 is a flow chart of the method implementation of the present invention.
Detailed Description
The invention is described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the present invention is specifically implemented as follows:
(1) flight data are obtained from the QAR, the data volume is more than 5000, and the samples are used as samples for pre-training a deep autoregressive aircraft landing model. Preprocessing the samples, selecting data with the altitude of below 250 feet, which are not in the takeoff stage of the airplane, according to the altitude parameters, and intercepting and storing the data in the interval. Meanwhile, the aircraft state parameters in the raw data are selected to comprise a pitch angle, a roll angle, a ground speed, a horizontal acceleration, a longitudinal acceleration, a descent rate, an airspeed and the like. The driver operating parameters include roll operation, pitch operation, foot pedal angle, and the like. Thus, the data for each flight is pre-processed to include the parameters and samples within the block. Each record is a vector of the true values of the above parameters recorded at that point, and each vector is referred to as a landing state point.
(2) And (2) clustering all the landing state points obtained in the step (1) by using a Kmeans algorithm, selecting the number of clusters to be 900, and dividing all the landing state points into 900 categories, wherein each category has a clustering mass center which is a vector, and each value in the vector corresponds to the parameter screened in each step (1). And storing the centroid of each clustering class, and storing the centroid into a landing state class-centroid vector mapping according to a key-value form. And mapping each landing state point into a symbol corresponding to the center of mass to which the landing state point belongs according to the stored center of mass. A flight sample is composed of a plurality of landing state points, all the landing state points of the flight sample are mapped into symbols and recombined together according to the original sequence to obtain a symbolic representation of the original flight sample.
(3) And (3) training a deep learning model by using a large number of landing state sequences generated in the step (2). The method is characterized in that each landing state is mapped into a vector, the process is called state embedding, and the vector is trained along with a deep autoregressive aircraft landing model. Meanwhile, altitude embedding is added, the altitude is divided into a plurality of sections at intervals of 10, and each section is represented by a special symbol. The altitude symbolic representation at each time instant is mapped to a vector. The vectors of this portion are also trained with the depth autoregressive aircraft landing model. After the landing state vector representation and the altitude vector representation of each point, the two vectors are added and input into the recurrent neural network. In the recurrent neural network, for a flight landing sample, the sum of the landing state vector and the altitude vector at the corresponding moment is input, the network is updated by using the formula of the LSTM model, and the landing state symbol at the next moment is input as a target value to predict the target value. The prediction method comprises the steps of using the output vector of the LSTM at the previous moment, mapping the output vector of the LSTM at the next moment into 900 scalars by a matrix, bringing the scalars and the signs of the landing state at the moment to be predicted into a cross entropy loss function to obtain prediction loss, and then performing back propagation on the loss to update parameters in the network.
(4) And (4) based on the deep autoregressive aircraft landing model trained in the step (3), recording parameters by using the vertical load from the QAR data, and selecting a sample with the vertical load larger than 1.5 as a heavy landing sample. These heavy landing samples are used for fine tuning based on the aircraft landing model. The fine tuning method comprises the steps of carrying out the same preprocessing mode stated in (1) and (2) on one sample in the heavy landing samples to obtain a symbol sequence of the sample, inputting the symbol sequence into a pre-trained deep autoregressive aircraft landing model, obtaining a vector output by the model at the last moment, mapping the vector into a scalar by using a matrix, bringing information of whether the scalar and the sample are the heavy landing samples into a cross entropy loss function, carrying out back propagation on loss after obtaining the loss, and updating parameters of the model. After the model is trained, the model can be used for re-landing prediction. The method comprises the steps that a certain flight is processed by the preprocessing process in the steps (1) and (2), the processed symbolic sequence is input into a finely adjusted deep autoregressive aircraft landing model, the finely adjusted autoregressive aircraft landing model can directly input the probability that the sample is a heavy landing sample, and if the probability is greater than 0.5, the sample is judged to be the heavy landing sample.
The accuracy rate of the method for predicting the heavy landing reaches 0.761, the accuracy rate of the method for predicting the heavy landing reaches 0.618 by using the previous method, for example, a statistical machine learning method is combined with characteristic engineering, the accuracy rate of the prediction of the heavy landing is improved by 14% compared with the accuracy rate of the prediction of 0.761 provided by the invention, and the effect of the method provided by the invention is integrally improved by 14%. The method is proved to obtain very remarkable results on the aspect of improving the heavy landing prediction effect.
Portions of the invention not described in detail are well within the skill of the art.
Although the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.

Claims (5)

1. A heavy landing prediction method based on deep learning is characterized by comprising the following steps:
step 1: acquiring flight data for recording flight parameters, which is called QAR data for short, wherein the flight data records the data of the whole process from take-off to landing of an airplane, and the final approach data with the altitude of below 250 feet is intercepted from the flight data and is used as flight landing stage data; selecting key parameters in a landing stage, wherein the key parameters comprise pitch angle, roll angle, descent rate, wind speed, wind direction, driver pitch lever operation, driver roll angle operation, driver pedal plate operation, pedal plate angle, ground speed, airspeed, longitudinal acceleration and horizontal acceleration, each flight sample is represented as a multivariate time sequence through interception and key parameter selection operation, each time point is a vector formed by QAR record values of the parameters at the time point, and each vector is called as a landing state point;
step 2: defining landing state points as recorded values of each moment in an aircraft landing sequence, clustering all the landing state points of all flight samples acquired from a QAR by using a Kmeans method, and enabling the Kmeans clustering to correspond all the landing state points to different landing state categories according to the values of the landing state points; generating a flight landing state dictionary by using a centroid generated by a Kmeans clustering algorithm, mapping each landing state point in a flight to a symbolic representation in the flight landing state dictionary, wherein each flight is composed of a plurality of landing state points, and obtaining a symbolic sequence representation by each flight through mapping;
and step 3: building a deep learning autoregressive model by using a recurrent neural network (LSTM), wherein the input of the model is represented by the symbol sequence of the flight obtained in the step 2, the topic structure of the model is LSTM, and the output of the model at each moment is a vector obtained by LSTM coding; the Model is utilized to carry out pre-training in a mode that at any time t, the landing state at the current time is predicted by using the landing states at the previous t-1 times, and the pre-trained Model is an aircraft landing Model ALM (aircraft landing Model, ALM);
and 4, step 4: selecting a sample with a vertical load parameter value larger than 1.5 as a heavy landing training sample according to a vertical load parameter value recorded in QAR data, converting heavy landing sample data into a symbol sequence by using the flight landing state dictionary generated in the step 2, continuing training on the basis of the airplane landing model pre-trained in the step 3 by using the converted flight landing symbol sequence, and finally obtaining a prediction result by using the training method to minimize the heavy landing prediction loss.
2. The deep learning based heavy landing prediction method of claim 1, characterized in that: the step 1 is realized by the following steps:
(11) preparing data: extracting QAR data, intercepting data between 250 feet from the ground height and 0 foot from the ground as aircraft landing data, wherein data of a time period before 5 seconds of landing is selected as training data;
(12) parameter extraction: selecting some characteristics which are most important for landing according to recorded data in QAR data and related documents, and dividing the characteristics into three groups of parameters, namely airplane state parameters, environment parameters and driver operation parameters, wherein the airplane state parameters comprise a pitch angle, a roll angle, a ground speed, a horizontal acceleration, a longitudinal acceleration, a descent rate and an airspeed; the driver operating parameters include roll operation, pitch operation, foot pedal operation, and foot pedal angle; the environmental parameters comprise wind speed and wind direction;
(13) and (3) generating a landing state point: and (3) processing each flight sample in the steps (1) and (2) to obtain a plurality of vectors, wherein each landing site is a vector and contains the value of the parameter selected in the step (2) at the current landing site.
3. The deep learning based heavy landing prediction method of claim 1, characterized in that: the step 2 is realized by the following steps:
(21) clustering operation is carried out on all landing state vectors by using a Kmeans clustering algorithm to generate 900 clustering centroids, the number of the clustering centroids is the optimal clustering centroid number determined by a clustering index, each clustering centroid is a mapping from a symbol to a vector, unique numbers are distributed to the 900 clustering centroids generated by the Kmeans, and the vector representation calculation method of each centroid is as follows: distributing the landing state points in the data to each centroid by a Kmeans algorithm, averaging vectors corresponding to the landing state points divided into the centroids, and obtaining vector representation of the corresponding centroids;
(22) and aiming at each landing state vector, allocating a clustering centroid for the landing state vector, wherein the clustering centroid is used as a symbolic representation of the landing state vector, the centroid is selected by calculating Euclidean distances of all centroids obtained by clustering the landing state point vector and Kmeans, the centroid with the smallest Euclidean distance is selected as the centroid allocated to the landing state point, each landing state vector of each flight is converted into a symbol corresponding to the centroid, and the mapping results of all the landing state points of the same flight form a symbol sequence.
4. The deep learning based heavy landing prediction method of claim 1, characterized in that: the step 3 is realized by the following steps:
(31) inputting a landing symbol sequence representation of each flight, using an embedded layer to map each landing symbol into a high-dimensional vector, wherein the high-dimensional vector corresponding to each landing state symbol is initialized randomly at first and is automatically updated in the training process of the deep learning network; altitude embedding is introduced, each altitude height range corresponds to one altitude embedding, updating is carried out along with training of the deep learning model, and the embedding of the landing state symbol and the embedding of the altitude height are added to serve as the final input of the deep learning model;
(32) building a one-way LSTM autoregressive model, predicting a landing state symbol at the current moment by using the landing state symbol at the previous moment at each time step of the model, converting 50000 landing samples extracted from QAR (quality enhancement Rate) into symbol sequence representation by using a method introduced in 3, training the LSTM model by using the converted data, wherein the calculation mode of the LSTM in the training process is as follows:
ft＝σ(Wf·[ht-1,xt]+bf)
it＝σ(Wi·[ht-1,xt]+bi)
Figure FDA0002310111760000031
Figure FDA0002310111760000032
ot＝σ(Wo·[ht-1,xt]+b0)
ht＝ot*tanh(Ct)
wherein, Wf，Wi，WC，Wo，bf，bi，bC，b0For trainable parameters in the model, xtGenerating vector representation through an embedded layer by using symbolic representation corresponding to the landing state at the time t, wherein other symbols are intermediate variables generated in the calculation process of the LSTM;
the loss function used is:
Figure FDA0002310111760000033
wherein T represents the number of landing state points of flight landing data, W, b are parameters of model trainability, and xiThe symbol is corresponding to the landing state at the moment i.
5. The deep learning based heavy landing prediction method of claim 1, characterized in that: the step 4 is realized by the following steps:
(41) processing the heavy landing prediction data set into a landing symbolic representation by using the same method, and finely adjusting the trained LSTM in a manner that cross entropy loss is optimized on the heavy landing prediction data set, wherein a used loss function is as follows:
Figure FDA0002310111760000034
wherein N represents the total sample size, yiWhether the ith sample is subjected to heavy landing or not is represented, the value of the heavy landing is 1, the value of the non-heavy landing is 0, and p is the probability value for predicting the heavy landing of the last output of the LSTM model;
(42) the finely adjusted deep autoregressive aircraft landing model is used for re-landing prediction, QAR data of a flight before landing is given, the QAR data of the flight before landing is subjected to discretization pretreatment and then is input into the finely adjusted deep autoregressive aircraft landing model, and a prediction result is obtained.
CN201911255415.2A 2019-12-10 2019-12-10 Deep learning-based heavy landing prediction method Pending CN111008669A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911255415.2A CN111008669A (en) 2019-12-10 2019-12-10 Deep learning-based heavy landing prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911255415.2A CN111008669A (en) 2019-12-10 2019-12-10 Deep learning-based heavy landing prediction method

Publications (1)

Publication Number Publication Date
CN111008669A true CN111008669A (en) 2020-04-14

Family

ID=70114328

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911255415.2A Pending CN111008669A (en) 2019-12-10 2019-12-10 Deep learning-based heavy landing prediction method

Country Status (1)

Country Link
CN (1) CN111008669A (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN112687128A (en) * 2020-11-27 2021-04-20 民航成都信息技术有限公司 Prediction method and device for estimated arrival time of flight path and automatic prediction system
CN112793797A (en) * 2021-02-03 2021-05-14 东航技术应用研发中心有限公司 Early warning method and system for landing bounce of airplane
CN112951003A (en) * 2021-02-26 2021-06-11 中国国际航空股份有限公司 Airplane heavy landing detection method, device, equipment and readable storage medium
CN112991575A (en) * 2021-02-08 2021-06-18 北京航空航天大学 Quantitative detection method for atmospheric environment chaos degree in airplane flying process
CN113033621A (en) * 2021-03-05 2021-06-25 南京航空航天大学 Method for identifying unstable approach and inducement thereof of civil aircraft
CN113378451A (en) * 2021-04-16 2021-09-10 北京航空航天大学 Aircraft landing site prediction method, equipment and memory
CN113486938A (en) * 2021-06-28 2021-10-08 重庆大学 Multi-branch time convolution network-based re-landing analysis method and device
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
CN116050842A (en) * 2023-02-01 2023-05-02 东华理工大学 Dynamic control method and system for urban underground engineering construction safety risk
CN116205354A (en) * 2023-02-21 2023-06-02 重庆大学 Method, device and storage medium for predicting flying heavy landing
CN116522771A (en) * 2023-04-21 2023-08-01 重庆大学 Attention mechanism-based bidirectional two-stage interpretable heavy landing prediction method
CN117194982A (en) * 2023-09-06 2023-12-08 中国民航科学技术研究院 Landing load risk early warning method and system for civil airliner and electronic equipment

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107045638A (en) * 2016-12-30 2017-08-15 中国民航管理干部学院 A kind of flight safety affair analytical method based on context-aware model
JP2017162514A (en) * 2017-06-09 2017-09-14 国立研究開発法人宇宙航空研究開発機構 System, method, and program for supporting landing judgement
US20170301247A1 (en) * 2016-04-19 2017-10-19 George Mason University Method And Apparatus For Probabilistic Alerting Of Aircraft Unstabilized Approaches Using Big Data
CN108711005A (en) * 2018-05-14 2018-10-26 重庆大学 Flight risk analysis method based on QAR data and Bayesian network
CN109948540A (en) * 2019-03-19 2019-06-28 四川函钛科技有限公司 Timing QAR parameter attribute extracting method based on curve interpolation and sampling
CN109977517A (en) * 2019-03-19 2019-07-05 北京瑞斯克企业管理咨询有限公司 A kind of personal landing again and group's offline mode comparative analysis method based on QAR parameter curve
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

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170301247A1 (en) * 2016-04-19 2017-10-19 George Mason University Method And Apparatus For Probabilistic Alerting Of Aircraft Unstabilized Approaches Using Big Data
CN107045638A (en) * 2016-12-30 2017-08-15 中国民航管理干部学院 A kind of flight safety affair analytical method based on context-aware model
JP2017162514A (en) * 2017-06-09 2017-09-14 国立研究開発法人宇宙航空研究開発機構 System, method, and program for supporting landing judgement
CN108711005A (en) * 2018-05-14 2018-10-26 重庆大学 Flight risk analysis method based on QAR data and Bayesian network
CN109948540A (en) * 2019-03-19 2019-06-28 四川函钛科技有限公司 Timing QAR parameter attribute extracting method based on curve interpolation and sampling
CN109977517A (en) * 2019-03-19 2019-07-05 北京瑞斯克企业管理咨询有限公司 A kind of personal landing again and group's offline mode comparative analysis method based on QAR parameter curve
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

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
LU，YUCHENG 等: "Pre-training of autoregressive model for aircraft hard landing prediction based on QAR data", 《2019 IEEE 31ST INTERNATIONAL CONFERENCE ON TOOLS WITH ARTIFICIAL INTELLIGENCE (ICTAI)》 *
LU，YUCHENG 等: "Pre-training of autoregressive model for aircraft hard landing prediction based on QAR data", 《2019 IEEE 31ST INTERNATIONAL CONFERENCE ON TOOLS WITH ARTIFICIAL INTELLIGENCE (ICTAI)》, 6 November 2019 (2019-11-06) *
ZHANG, HAOCHI 等: "Aircraft hard landing prediction using LSTM neural network", 《2018 ASSOCIATION FOR COMPUTING MACHINERY ACM》 *
ZHANG, HAOCHI 等: "Aircraft hard landing prediction using LSTM neural network", 《2018 ASSOCIATION FOR COMPUTING MACHINERY ACM》, 23 September 2018 (2018-09-23), pages 1 - 7 *

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN112687128A (en) * 2020-11-27 2021-04-20 民航成都信息技术有限公司 Prediction method and device for estimated arrival time of flight path and automatic prediction system
CN112793797A (en) * 2021-02-03 2021-05-14 东航技术应用研发中心有限公司 Early warning method and system for landing bounce of airplane
CN112991575A (en) * 2021-02-08 2021-06-18 北京航空航天大学 Quantitative detection method for atmospheric environment chaos degree in airplane flying process
CN112951003B (en) * 2021-02-26 2022-11-29 中国国际航空股份有限公司 Airplane heavy landing detection method, device, equipment and readable storage medium
CN112951003A (en) * 2021-02-26 2021-06-11 中国国际航空股份有限公司 Airplane heavy landing detection method, device, equipment and readable storage medium
CN113033621A (en) * 2021-03-05 2021-06-25 南京航空航天大学 Method for identifying unstable approach and inducement thereof of civil aircraft
CN113378451A (en) * 2021-04-16 2021-09-10 北京航空航天大学 Aircraft landing site prediction method, equipment and memory
CN113486938A (en) * 2021-06-28 2021-10-08 重庆大学 Multi-branch time convolution network-based re-landing analysis method and device
CN113486938B (en) * 2021-06-28 2022-11-01 重庆大学 Multi-branch time convolution network-based re-landing analysis method and device
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
CN116050842A (en) * 2023-02-01 2023-05-02 东华理工大学 Dynamic control method and system for urban underground engineering construction safety risk
CN116050842B (en) * 2023-02-01 2023-08-04 东华理工大学 Dynamic control method and system for urban underground engineering construction safety risk
CN116205354A (en) * 2023-02-21 2023-06-02 重庆大学 Method, device and storage medium for predicting flying heavy landing
CN116205354B (en) * 2023-02-21 2023-08-11 重庆大学 Method, device and storage medium for predicting flying heavy landing
CN116522771A (en) * 2023-04-21 2023-08-01 重庆大学 Attention mechanism-based bidirectional two-stage interpretable heavy landing prediction method
CN116522771B (en) * 2023-04-21 2024-01-26 重庆大学 Attention mechanism-based bidirectional two-stage interpretable heavy landing prediction method
CN117194982A (en) * 2023-09-06 2023-12-08 中国民航科学技术研究院 Landing load risk early warning method and system for civil airliner and electronic equipment
CN117194982B (en) * 2023-09-06 2024-02-13 中国民航科学技术研究院 Landing load risk early warning method and system for civil airliner and electronic equipment

Similar Documents

Publication Publication Date Title
CN111008669A (en) Deep learning-based heavy landing prediction method
CN108510741B (en) Conv1D-LSTM neural network structure-based traffic flow prediction method
Ma et al. A hybrid CNN-LSTM model for aircraft 4D trajectory prediction
CN109191922B (en) Large-scale four-dimensional track dynamic prediction method and device
CN113487855B (en) Traffic flow prediction method based on EMD-GAN neural network structure
CN107045638A (en) A kind of flight safety affair analytical method based on context-aware model
US11790767B2 (en) Method, apparatus, device and storage medium for pre-warning of aircraft flight threat evolution
CN107728234A (en) A kind of intensity of lightning value Forecasting Methodology based on atmospheric electric field data
Zhao et al. An incremental clustering method for anomaly detection in flight data
CN114004292B (en) Pilot flat-floating ejector rod behavior analysis method based on flight parameter data unsupervised clustering
CN108304974B (en) Optimization C5.0 and Apriori-based civil aviation NOSHOW prediction and strong factor analysis method
CN110442143B (en) Unmanned aerial vehicle situation data clustering method based on combined multi-target pigeon swarm optimization
CN105956790A (en) Low-altitude flight situation safety evaluation indexes and evaluation method thereof
CN105550473A (en) Heavy landing prediction method based on support vector machine classification
CN103235954A (en) Improved AdaBoost algorithm-based foundation cloud picture identification method
CN115659244A (en) Fault prediction method, device and storage medium
CN110659773A (en) Flight delay prediction method based on deep learning
CN110084413A (en) Safety of civil aviation risk index prediction technique based on PCA Yu depth confidence network
CN117035060A (en) CC-MIDINN-based airplane arrival time prediction method
Herrema et al. A novel machine learning model to predict abnormal Runway Occupancy Times and observe related precursors
CN108762523A (en) Output characters through input method prediction technique based on capsule networks
CN115293639A (en) Battlefield situation studying and judging method based on hidden Markov model
Lu et al. Pre-training of autoregressive model for aircraft hard landing prediction based on QAR data
Zhang et al. Runway visual range prediction based on ensemble learning
CN112488146A (en) Airplane danger identification and trend prediction method based on deep learning

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200414