AUTOMATIC AVIONICS LEARNING

专利摘要:
The document relates to systems and methods for managing the flight of an aircraft, comprising the steps of receiving data (200) from records of the flight of an aircraft; said data comprising data from sensors and / or data from on-board avionics; determining the aircraft state at a point N (220) from the data received (200); determine the state of the plane at point N + 1 (240) from the state of the plane at point N (220), by applying a model learned by machine learning (292). Developments describe the use of flight parameters SEP, FF and N1; unsupervised, offline and / or online machine learning, using a variety of algorithms and neural networks. Software aspects are described.] Figure for the abstract: Fig. 2
公开号:FR3090851A1
申请号:FR1873514
申请日:2018-12-20
公开日:2020-06-26
发明作者:Christophe Pierre；Dorian MARTINEZ；Bastien CRETE
申请人:Thales SA；
IPC主号:

专利说明:

Description
Title of the invention: AUTOMATIC LEARNING IN AVIONICS
Field of the invention
The invention relates to the field of avionics in general. The invention relates in particular to methods and systems for predicting the future state of an aircraft. Prior art
The known prior art approaches (e.g. WO 2017042166, or US9290262) are generally based on a data set modeling the performance of an aircraft.
Different approaches are known for modeling aircraft performance. Works such as "BADA" (acronym for "Base of Aircraft Data" in English) or "safety-line" by EUCASS ("European Conference for Aeronautics and Space Sciences") have limitations. The BADA model is limited in terms of thrust and drag. The EUCASS model only applies to airplanes powered by turbojets (i.e. piloted in NI).
[0004] Technical problems in aeronautics generally involve many different parameters and the optimizations currently developed then converge little, badly or not at all. This absence of convergence (or convergence towards local minima) is frequently observed when the modeled processes are large (and follow different models depending on the observation points).
Approaches based on models integrating physical equations of the aircraft are generally dependent on the quality of the modeling and on the knowledge of the actual behavior of the aircraft. In fact, these methods are generally not robust to the variability of the actual behavior of each aircraft, compared to that of an "average" (modeled) aircraft.
There is an industrial need for advanced methods and systems for optimizing all or part of the operations of an aircraft.
Summary of the invention
The document relates to systems and methods for managing the flight of an aircraft, comprising the steps of receiving data (200) from records of the flight of an aircraft; said data comprising data from sensors and / or data from on-board avionics; determining the aircraft state at a point N (220) from the data received (200); determining the state of the aircraft at point N + l (240) from the state of the aircraft at point N (220), by applying a model learned by machine learning (292). Developments describe the use of SEP, FF and NI flight parameters; unsupervised, offline and / or online machine learning, using a variety of algorithms and neural networks. Software aspects are described. A method for managing the flight of an aircraft is described, comprising the steps of: receiving data from recordings of the flight of an aircraft; said data comprising data from sensors and / or data from on-board avionics; determine the aircraft state at a point N from the data received; determine the state of the plane at point N + l from the state of the plane at point N, by applying a model learned by machine learning. In this embodiment, the learning is carried out from start to finish, ie includes the PERFDB step (performance calculation) and the TRAJ / PRED trajectory calculation step: the output data are determined directly by the learning set implemented on the input data.
In one embodiment, the step of determining the state of the aircraft at point N + 1 from the state of the aircraft at point N comprises the steps of: determining the parameters of SEP, FF and NI flight from airplane state to point N, by applying a model learned by machine learning; and determine the aircraft state at point N + l from the values of the flight parameters SEP, FF and NI by trajectory calculation; in which the value SEP designates the energy available for the climb of the aircraft, the value FF designates the variation in the mass of fuel and the value NI designates the speed of rotation of the first stage of the engine influencing the fuel consumption.
In one embodiment, machine learning is unsupervised. Unsupervised learning aims to find underlying structures from unlabeled (or non-annotated) data. The number and definition of classes are not given a priori. This type of learning includes, for example, deep learning. The associated advantages of this type of learning include the use of high computing power on large accumulated data, the absence of need for human control, the discovery of patterns or patterns or links which are not necessarily understandable by the man but who can be effective.
In one embodiment, the automatic learning is supervised. Advantageously, certain attributes of the data can be known e.g. SEP, N1 or FF. The expected outputs are known, the classes or categories of the records are known (labels, labels). Supervised learning allows manual interventions and therefore can produce effective models, for example converging faster. Conversely, human presuppositions can reduce the space of possibilities (unconstrained in the unsupervised case).
In one embodiment, the automatic learning is carried out offline. The records can be records of past flights (data mining approaches). This embodiment is advantageous in that it makes it possible to reuse existing data (which are numerous and currently underused).
In one embodiment, the automatic learning is carried out online. In one embodiment, machine learning can be performed incrementally or online. From a generic model (type or series of aircraft) known on average, a particular aircraft can be characterized, progressively refined, as it goes on its own flights (by serial number or registration or "tail number"). When the model is known, it is possible to continue learning in data flow (to improve the existing model, without starting from scratch). Offline machine learning learns from a whole data set, while online learning can continue learning ("learning transfer"), on-board, without having to re-ingest the initial data.
It should be noted that the automatic learning implemented in the method according to the invention can include two types of learning: offline learning makes it possible, for example, to configure the generic aircraft model and the online learning then allows you to configure the model that is unique to each particular aircraft. (But offline learning can also be used to specify a specific aircraft.) It is also possible to use only one type of learning (an airline may want to focus only on the aircraft class, while another airline will want to know the precise characteristics of a given aircraft, for example to finely optimize fuel consumption).
In one embodiment, the automatic learning comprises one or more algorithms selected from the algorithms comprising: support vector machines or large margin separators; classifiers; neural networks; decision trees and / or stages of statistical methods such as the Gaussian mixture model, logistic regression, linear discriminant analysis and / or genetic algorithms.
A computer program product is described, said computer program comprising code instructions making it possible to carry out one or more of the steps of the method, when said program is executed on a computer.
A system is described for the implementation of one or more of the steps of the method, the system comprising one or more avionics systems of avionics type of the Flight Management System FMS type and / or electronic flight bag EFB.
In one embodiment, the system further comprises one or more neural networks, chosen from neural networks comprising: an artificial neural network; an acyclic artificial neural network; a recurrent neural network; a forward propagating neural network; a convolutional neural network; and / or a network of generative antagonistic neurons.
Advantageously, the methods according to the invention make it possible to predict the performance of aircraft, and this independently of the models supplied by the manufacturers.
Advantageously, the methods according to the invention allow learning to continue without time limit (e.g. online learning, in particular by reinforcement, using the data streams obtained from commercial flight records).
Advantageously, the method according to the invention can be implemented in embedded systems for calculating and / or predicting trajectories, and in particular in electronic flight bags of the EFB ("Electronic Flight Bag") type. The invention can be implemented in an FMS ("Flight Management System") type computer or in a set of systems interconnecting the FMS with one or more EFBs.
The potential applications of the invention relate to the calculation of trajectories, assistance to an aircraft manufacturer for establishing aircraft performance, the optimization of airline operations of a company, flight simulation, assistance with mission management, assistance with piloting an aircraft, development of avionics systems in the broad sense or even predictive maintenance by modeling the evolution of aircraft performance.
Description of the figures
Various aspects and advantages of the invention will appear in support of the description of a preferred mode of implementation of the invention but not limiting, with reference to the figures below:
[Fig.l] illustrates some of the technical objectives pursued by the invention;
[Fig.2] illustrates the links between the performance model and the predictive trajectory calculation;
[Fig.3] illustrates the coupling between automatic learning and integration in the trajectory prediction;
[Fig. 4] illustrates certain aspects of the invention according to a specific processing chain;
[Fig. 5] illustrates different learning modes according to different embodiments of the invention.
Detailed description of the invention
Different types of automatic learning (or machine) are possible. Machine learning is an area of computing that uses statistical techniques to give computer systems the ability to learn with data (for example, to progressively improve the performance of a specific task), without being explicitly programmed for this purpose.
Machine learning is useful for detecting and recognizing patterns or diagrams or patterns. It is generally easier to collect data (for example, data from a video game or board game) than to explicitly write the program that governs the game in question. In addition, neural networks (hardware realization of machine learning, or software emulation) can be reused to process new data. Machine learning can be performed on particularly large data, i.e. using as much data as possible (e.g. stability, convergence, weak signals, etc.). New data can be added continuously and learning can be refined.
Different learning algorithms can be used, in combination with the characteristics according to the invention. The method may include one or more algorithms among the algorithms comprising: support vector machines or large margin separators (in English "Support Vector Machine", acronym SVM); "boosting" (classifiers); neural networks (in unsupervised learning); decision trees (Random Lorest), statistical methods such as the Gaussian mixture model; logistic regression; linear discriminant analysis; and genetic algorithms.
Machine learning tasks are generally classified into two main categories, depending on whether there is a learning signal or inputs or available feedback or outputs ”.
The expression “supervised learning designates a situation in which examples of inputs and examples of outputs (real or desired) are presented to the computer. Learning then consists of identifying an interlacing of rules that matches the inputs to the outputs (these rules may or may not be understandable for humans).
The expression "semi-supervised learning" designates a situation in which the computer receives only an incomplete set of data: for example, there is missing output data.
The term "reinforcement learning" consists of learning the actions to be taken, from experience, so as to optimize a quantitative reward over time. Through iterative experiences, a decisional behavior (called strategy or policy, which is a function associating in the current state the action to be executed) is determined to be optimal, in that it maximizes the sum of the rewards during time.
The expression “unsupervised learning” (also called deep ap learning or deep learning) designates a situation in which no annotation exists (no wording, no description, etc.), leaving the algorithm for learning only to find one or more structures, between inputs and outputs. Unsupervised learning can be an objective in itself (discovery of structures hidden in the data) or a means of achieving an objective (functionality-based learning).
In computer science, an "online algorithm" (or "online algorithm") is an algorithm which receives its input not at once, but as a data stream, and which must make decisions as and as you go. In the context of machine learning, it is possible to use the terminology of “incremental learning algorithm”.
Not knowing the entirety of the data, an incremental learning algorithm must make choices which may prove to be non-optimal a posteriori. It is possible to perform competitive analyzes by comparing the performance, on the same data, of the incremental learning algorithm and the equivalent having all the data available. Online algorithms include algorithms called K server, Balance !, Balance-Slack, Double Coverage, Equipoise, Handicap, Harmonie, Random-Slack, Tight Span Algorithm, Tree Algorithm and Work Eunction Algorithm. Online algorithms have links to probabilistic and approximate algorithms.
According to the embodiments, the human contribution in the machine learning stages can vary. In some embodiments, machine learning is applied to machine learning itself (reflexive). The entire learning process can be automated, in particular by using several models and comparing the results produced by these models. In most cases, humans are involved in human learning in the loop. The developers or curators are responsible for the maintenance of the clusters of data: ingestion of data, cleaning of the data, discovery of models etc. In some cases, humans do not have to intervene, learning is fully automatic once the data is made available.
Machine learning used in combination with the features according to the invention generally benefits from having large amounts of data. The term "Big data" refers to the collection and analysis of data, carried out on a massive scale. This concept is associated with characteristics of a technical nature which include: volume (eg large collections of data, even if they are redundant), variety (eg many different sources are used), velocity (eg the data is "fresh" ”Or constantly updated in changing or dynamic environments), attesting to a certain veracity (eg the weak signals which are drowned in the noise are not suppressed and can consequently be detected or amplified), to represent in fine a certain value (for example, from a technical and / or business point of view ie business).
In one embodiment, a “policy” learning method can be used. Policy methods are iterative methods alternating between policy evaluation and improvement phases. They are based on the current estimation of the value function (respectively of quality) in the current state to determine the choice of the next action (control), after observation of the new current state and of the received reinforcement signal, the model which has been used is updated. A classic example of this type of method is the SARSA algorithm.
In one embodiment, an "outside politics" learning method can be used. Non-political methods are not sensitive to the way in which actions are selected at all times, but only to the fact of observing a control policy with a sufficient level of exploration. As a result, they can freely observe a different control policy (which may be suboptimal). A classic example of a non-political algorithm is the Q-learning algorithm.
[Fig.l] illustrates some of the technical objectives pursued by the invention.
In terms of optimization, it is customary to use models supplied by the manufacturers 101. These are generally generic, i.e. theoretical, static and poor in data. They relate to an "average" or "idealized" device, difficult to handle or ultimately irrelevant in certain contexts. In other words, there is a need for advanced 102 aircraft models that are "real" i.e. individualized (device by device), dynamic and based on large amounts of data (which are otherwise justly accessible).
In more detail, it is advantageous to be able to specify the performance of a particular device, individually, in particular in terms of optimization (for example for fuel consumption). Depending on the maintenance events carried out on a device, or according to the missions (e.g. load distribution, etc.), these performance data may vary for the same device (different loads, dynamic aspect).
Being able to assess in real time, on board or via remote calculation, the instant performance of an aircraft gives a significant advantage to an airline, which must manage a fleet of aircraft.
The embodiments of the invention described below at least partially meet the needs set out above.
[Fig. 2] illustrates the links between the performance model and the predictive trajectory calculation.
A trajectory comprises a plurality of points, among which the points N and N + 1 (trajectory points, in correspondence with flight plan points, waypoints).
The performance model 210 (known as PERFDB) determines the aircraft state; it includes tables 211 and one or more performance calculators 212. Tables 211 can be estimated (estimation of parameters of a parametric model from real flights).
The data from sensors or from avionics 200 (e.g. aircraft state at point N) are manipulated, as input. At the output, the parameters SEP, NI, FF (220) are determined (directly or indirectly), at said point N.
These output parameters 220 are then used in the trajectory calculation model 230 (integrators and propagators 231, etc.) which predicts the future (or next) state 240 of the aircraft from a present state ( or previous)
Aircraft condition
The state of the aircraft - at a given time - can indeed be characterized (roughly but satisfactorily) by three parameters which are the parameters SEP, FF and NI. These parameters are data measurable by aircraft sensors, and are therefore accessible in flight records.
The acronym SEP, for Specific Excess Power, designates the energy available for the climb of the aircraft, i.e. the ability to climb from the airplane divided by the mass (this parameter is not constant). MS is not measured directly, but is calculated from measurable data (such as altitude, speed, gravitational constant, etc.).
The acronym FF for Euel Elow designates the variation in the mass of fuel.
[Math.l] dm a
The acronym NI refers to the speed of rotation of the first stage of the engine, which has the most influence on fuel consumption. The available power is closely linked to this NI speed.
The SEP, FF and NI parameters are closely linked. In particular, the engine thrust mode and vertical guidance appear to be decisive for the SEP, FF and NI parameters.
An approach (“model-based” in English) can consist in modeling the interdependence between SEP, FF and NI, ie by formulating sets of equations involving these parameters (for example by modeling aerodynamics and / or the engine thrust mode and / or vertical guidance). Particularly effective optimizations (e.g. convergent and fast) can be obtained.
Modelless approaches, based on machine learning.
According to one embodiment of the invention, an advantageous alternative ("model-free" in English) consists in applying automatic learning methods. No prior knowledge is required. In other words, there can be no presupposition of any model, whether aerodynamic or motor, or other: machine learning establishes correspondences between input and output data sets, these data being real data ( because measured directly or determined indirectly).
Various automatic learning methods can be applied, and this at different levels: between 200 and 220 on the one hand (automatic learning 291), and between 200 and 240 on the other hand (automatic learning 292).
[Fig.3] illustrates the coupling between automatic learning and integration in the trajectory prediction.
The integration calculation 330 for predicting the trajectory of the aircraft relates only to the parameters predicted at point N + 1311, obtained by learning 300 from point N. The predicted data 311 and measured 312 serve to continue training 3100 of the Model 300.
The performance model 210 (known as PEREDB) determines the aircraft state; it includes tables 211 and one or more performance calculators 212. Tables 211 can be estimated (estimation of parameters of a parametric model from real flights).
Input is manipulated data from sensors and / or avionics. At the output, the parameters SEP, NI, EL 220 are determined (directly or indirectly).
In one embodiment of the invention, these output parameters are used in the trajectory calculation model 230 (integrators and propagators 231, etc.) which predicts the future (or next) state of the aircraft to from a present (or previous) state
The flight parameters at a point N are received (measured and / or calculated) and then submitted to the learning module which:
- processed the data in batch (unsupervised learning on a heap of data) or
- Processes the data in flow (incremental or online learning see below) In one embodiment, the input data (inputs) are received and / or supplied, as well as the output data (outputs). Automatic learning is performed on the output data and the learning then establishes the "links" between inputs and outputs.
In one embodiment ("differential"), the differences between the data predicted by learning and the data actually measured in flight possibly modify the learned model. The predicted data and / or the measured data are manipulated by the flight computers.
[Fig.4] illustrates certain aspects of the invention according to a specific processing chain.
The sensor measurements and / or calculations and / or other observations are collected in step 410, possibly filtered and formatted in step 420; a corpus (reduced) i.e. reference is defined in step 430 (feedbacks by human expertise and / or machine filtering) and aircraft state tables 440 comprising a plurality of aircraft states are determined. By considering a particular airplane state, the learning module 300 predicts an airplane state at the next point N + 1 from data at the previous point N. The learned model 450 is gradually and / or iteratively refined, possibly validated or modified 460 by an operator, and may optionally be the subject of reports and statistics 470 (for certification authorities, traffic regulation authorities, the manufacturer, an equipment supplier, etc.).
[Fig.5] illustrates different learning modes according to different embodiments of the invention.
In one embodiment, for a given aircraft 500, the properties of which it is desired to know better (eg to determine its flight behavior), the method comprises a first step 512 consisting in modeling the link between (i) the state on the one hand and (ii) the parameters (SEP, FF, NI) on the other. To this end, the model is trained by automatic learning on a large number of flight records relating to the type of aircraft concerned. In a second step, the model is on-board (ie implementation on board) and refined to refine or reinforce or improve it by serial number, ie specifically or specifically to the aircraft considered (each device is unique, slightly different from other devices in the same category or type of device). To this end, the (generic) model is specified by automatic learning carried out on the flight data specific to the aircraft considered. The data can be flight recordings (past) of this aircraft (offline 521) or received live, streaming or streaming (online 522).
This last embodiment is an “end-to-end” modeling: a learning model 300 (neural network type for example) is driven with the airplane state as input, and the SEP, the EL as output. and NI. As a result, the (existing) integrators are exploited with the outputs of this model, in order to predict the future state of the aircraft.
What is learned on an aircraft 520 can be repeated on the scale of a fleet of 530 devices. It is then possible to match the generic model with an average carried out on devices of the same type.
In one embodiment, for a given aircraft 500, the properties of which it is desired to know better (to determine its flight behavior), the method comprises a first step 511 consisting in modeling the link between (i) the state of the aircraft at point N on the one hand and (ii) the state of the aircraft at point N + l on the other hand. This step comprises a step consisting in using the integration / trajectory prediction 230. To this end, the model is trained by automatic learning on a large number of flight records relating to the type of aircraft concerned. In a second step, the model is on-board (ie implementation on board) and refined to refine or reinforce or improve it by serial number, ie specifically or specifically to the aircraft considered (each device is unique, slightly different from other devices in the same category or type of device). To this end, the (generic) model is specified by automatic learning carried out on the flight data specific to the aircraft considered. The data can be flight recordings (past) of this aircraft (offline 521) or received live, streaming or streaming (online 522).
The present invention can be implemented using hardware and / or software elements. It may be available as a computer program product on computer readable media.
Machine learning can correspond to hardware architectures which can be emulated or shnulable by computer (e.g. CPU-GPU), but sometimes not (circuits dedicated to learning may exist).
According to the embodiments, the method according to the invention can be implemented on or by one or more neural networks. A neural network according to the invention can be one or more neural networks chosen from neural networks comprising: a) an artificial neural network (in English "feedforward neural network"; b) an acyclic artificial neural network, eg a multilayer perceptron, thus distinguishing itself from recurrent neural networks; c) a forward propagation neural network; d) a Hopfield neural network (a discrete-time recurrent neural network model whose connection matrix is symmetrical and zero on the diagonal and where the dynamics are asynchronous, a single neuron being updated at each time unit ); e) a recurrent neural network (consisting of interconnected units interacting non-linearly and for which there is at least one cycle in the structure); f) a convolutional neural network (in English “CNN” or “ConvNet” for “Convolutional Neural Networks”, a type of feedforward acyclic artificial neural network, by multilayer stacking of perceptrons) or g) a generative antagonist network (in English “ generative adversarial networks »acronyms GANs, class of unsupervised learning algorithms)
In one embodiment, the learning calculations are carried out offline on a computer on the ground.
Advantageously, if the computer is on board an aircraft and has access to flight data, the model can be trained specifically to model with more precision the particular aircraft on board, by reinforcement methods.
Advantageously, if the computer is on board an aircraft and is connected to a parameter recorder, the flight data can be used in a real-time architecture to improve knowledge of aircraft performance in real time.
In an alternative embodiment, one or more steps of the method according to the invention are implemented in the form of a computer program hosted on a portable computer of the "EFB" type (acronym for "Electronic Flight Bag).
In an alternative embodiment, one or more steps of the method can be implemented within a computer of type "FMS" (acronym for "Flight Management System") or in an FM function of a computer of flight.

权利要求:
Claims (1)
[1" id="c-fr-0001]
Claims [Claim 1] Method for managing the flight of an aircraft, comprising the steps consisting in:- receive data (200) from recordings of the flight of an aircraft; said data comprising data from sensors and / or data from on-board avionics;- determine the aircraft state at a point N (220) from the data received (200);- determine the state of the plane at point N + l (240) from the state of the plane at point N (220), by applying a model learned by machine learning (292) [Claim 2] The method of claim 1, the step of determining the state of the aircraft at point N + 1 (240) from the state of the aircraft at point N (220) comprising the steps of:- determine the flight parameters SEP, FF and NI (220) from the plane state at point N (200), by applying a model learned by automatic learning (291); and- determine the aircraft state at point N + 1 (240) from the values of the flight parameters SEP, FF and N1 by trajectory calculation (230);in which the value SEP designates the energy available for the climb of the aircraft, the value FF designates the variation in the mass of fuel and the value NI designates the speed of rotation of the first stage of the engine influencing the fuel consumption. [Claim 3] The method of claim 1, the machine learning (291, 293) being unsupervised. [Claim 4] The method of claim 1, the machine learning being supervised. [Claim 5] Method according to any one of the preceding claims, the automatic learning being carried out offline. [Claim 6] Method according to any one of the preceding claims, the automatic learning being carried out online. [Claim 7] A method according to any one of the preceding claims, the machine learning comprising one or more algorithms selected from the algorithms comprising: support vector machines or wide margin separators; classifiers; neural networks; decision trees and / or stages of statistical methods such as the Gaussian mixture model, local regression
gistics, linear discriminant analysis and / or genetic algorithms. [Claim 8] Computer program product, said computer program comprising code instructions making it possible to carry out the steps of the method according to any one of claims 1 to 7, when said program is executed on a computer. [Claim 9] System for carrying out the steps of the method according to any one of Claims 1 to 7, the system comprising one or more avionics systems of avionics type of Flight Management System FMS type and / or electronic flight bag EFB. [Claim 10] The system of claim 9, further comprising one or more neural networks, selected from neural networks comprising: an artificial neural network; an acyclic artificial neural network; a recurrent neural network; a forward propagating neural network; a convolutional neural network; and / or a network of generative antagonistic neurons.
1/4

类似技术:

---

公开号 | 公开日 | 专利标题

---

Yu et al.2020|Hyper-parameter optimization: A review of algorithms and applications

---

EP3671392A1|2020-06-24|Automatic learning in avionics

---

EP2696344B1|2019-01-02|Method and system for detecting acoustic events in a given environment

---

Gayathri et al.2014|Software defect prediction system using multilayer perceptron neural network with data mining

---

EP2364467B1|2020-01-01|Method for recognising sequential patterns for a method for fault message processing

---

EP3143563A1|2017-03-22|Distributed model learning

---

FR2963448A1|2012-02-03|METHOD AND SYSTEM FOR ANALYSIS OF FLIGHT DATA RECORDED DURING THE FLIGHT OF AN AIRCRAFT.

---

Saldivar et al.2016|Self-organizing tool for smart design with predictive customer needs and wants to realize Industry 4.0

---

Napoli et al.2014|An agent-driven semantical identifier using radial basis neural networks and reinforcement learning

---

US20170091619A1|2017-03-30|Selective backpropagation

---

CN110245772A|2019-09-17|Analysis core with polymerize

---

WO2015086957A2|2015-06-18|Forecasting maintenance operations to be applied to an engine

---

Chen et al.2019|Predicting air compressor failures using long short term memory networks

---

ElSaid et al.2020|Neuro-evolutionary transfer learning through structural adaptation

---

FR3091762A1|2020-07-17|AUTOMATIC AVIONICS LEARNING

---

LU102710B1|2021-04-08|Input validation method for neural network model by crossing-layer dissection

---

US20200302396A1|2020-09-24|Earning Code Classification

---

Cerutti et al.2018|Learning and reasoning in complex coalition information environments: a critical analysis

---

Bouktif et al.2015|Predicting stock market movement: An evolutionary approach

---

FR3099282A1|2021-01-29|ANALYSIS OF AIRCRAFT TRACKS

---

US11176502B2|2021-11-16|Analytical model training method for customer experience estimation

---

EP3846087A1|2021-07-07|Method and system for selecting a learning model within a plurality of learning models

---

Chinchali et al.2020|Sampling training data for continual learning between robots and the cloud

---

Gurusamy et al.2022|DGSLSTM: Deep Gated Stacked Long Short-Term Memory Neural Network for Traffic Flow Forecasting of Transportation Networks on Big Data Environment

---

EP3913519A1|2021-11-24|Simulation by substitution of a physics laws model with an automatic learning model

同族专利:

---

公开号 | 公开日

---

CN111353256A|2020-06-30|

---

FR3090851B1|2021-03-19|

---

EP3671392A1|2020-06-24|

---

US20200202723A1|2020-06-25|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

US20090112535A1|2007-10-26|2009-04-30|Sensis Corporation|Method of integrating point mass equations to include vertical and horizontal profiles|

---

US9290262B2|2013-02-22|2016-03-22|Thales|Method for creating a vertical trajectory profile comprising multiple altitude levels|

---

WO2017042166A1|2015-09-09|2017-03-16|Thales|Optimising the trajectory of an aircraft|

---

US9542851B1|2015-11-03|2017-01-10|The Boeing Company|Avionics flight management recommender system|

---

US20180259342A1|2017-03-10|2018-09-13|Qualcomm Incorporated|System and method for dead reckoning for a drone|

---

CN112416021B|2020-11-17|2021-12-21|中山大学|Learning-based path tracking prediction control method for rotor unmanned aerial vehicle|

---

CN112610339B|2021-01-13|2021-12-28|南京航空航天大学|Variable cycle engine parameter estimation method based on proper amount of information fusion convolutional neural network|

法律状态:
2019-11-28| PLFP| Fee payment|Year of fee payment: 2 |

---

2020-06-26| PLSC| Publication of the preliminary search report|Effective date: 20200626 |

---

2020-11-25| PLFP| Fee payment|Year of fee payment: 3 |

---

2021-11-26| PLFP| Fee payment|Year of fee payment: 4 |

优先权:

---

申请号 | 申请日 | 专利标题

---

FR1873514A|FR3090851B1|2018-12-20|2018-12-20|AUTOMATIC LEARNING IN AVIONICS|FR1873514A| FR3090851B1|2018-12-20|2018-12-20|AUTOMATIC LEARNING IN AVIONICS|

---

EP19216280.8A| EP3671392A1|2018-12-20|2019-12-13|Automatic learning in avionics|

---

US16/716,160| US20200202723A1|2018-12-20|2019-12-16|Machine learning in avionics|

---

CN201911322849.XA| CN111353256A|2018-12-20|2019-12-20|Machine learning in avionics|

---

FR1915682A| FR3091762B1|2018-12-20|2019-12-31|AUTOMATIC LEARNING IN AVIONICS|