Training method, device and storage medium for flight trainees

Abstract

The present invention provides a training method, device and storage medium for flight trainees. The training method for flight trainees includes the following steps: obtaining a first flight number provided by an instructor; determining a confusion rule corresponding to the instructor and / or the flight trainee, and generating a second flight number that is easily confused with the first flight number based on the confusion rule; and broadcasting an air traffic control training instruction containing the second flight number to the flight trainee. The present invention can extract and train rule features from historical operation data through a conditional generative adversarial network model based on distributed learning, and is used to make intelligent recommendations based on instructor preferences, flight trainees' easy-to-make mistakes, and the accent of the target flight area, and generate similar flight numbers that are easily confused, so as to provide flight trainees with an intelligent air-to-ground communication training environment.

Description

Technical Field

[0001] The present invention relates to the field of aerodynamics, and in particular to a training method for flight trainees, a training device for flight trainees, and a computer-readable storage medium. Background Art

[0002] With the increase in the number of flights, pilots are prone to confuse instructions with similar flight numbers when flying flights, resulting in various errors such as misreading, mishearing or misreading. As a result, the pilot of one flight executes the instructions issued by air traffic control to another flight, resulting in the aircraft flying the wrong direction or route, the aircraft ascending or descending without permission, the aircraft switching frequency incorrectly, and the aircraft mishearing the control instructions and repeating them, interfering with the order of radio calls and other dangerous situations. Therefore, it is very important to train pilots in hearing and vision for similar flight numbers. In daily training, instructors often interfere with flight trainees by simulating similar flight numbers orally. However, this oral method cannot play the air-to-ground calls of similar flight numbers by one-touch triggering, which greatly increases the workload of instructors. In addition, there is currently no training method that can intelligently recommend and generate similar flight numbers that are easy to confuse based on instructor preferences, flight trainees' easy-to-make mistakes, and the accent of the target flight area, and it is impossible to provide intelligent services for flight training.

[0003] In order to solve the above-mentioned defects of the prior art, the art urgently needs an improved training method for flight trainees, which can be used to make intelligent recommendations based on instructor preferences, flight trainees' common error points and the accent of the target flight area, and generate similar flight numbers that are easy to confuse, so as to provide intelligent air traffic control training for flight trainees. Summary of the invention

[0004] A brief summary of one or more aspects is given below to provide a basic understanding of these aspects. This summary is not an exhaustive overview of all conceived aspects, and is neither intended to identify the key or decisive elements of all aspects nor to define the scope of any or all aspects. Its only purpose is to give some concepts of one or more aspects in a simplified form as a prelude to a more detailed description that will be given later.

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a training method for flight trainees, a training device for flight trainees and a computer-readable storage medium, which can extract and train rule features from historical operation data through a conditional generative adversarial network model based on distributed learning, and can be used to make intelligent recommendations based on instructor preferences, flight trainees' common error points and the accent of the target flight area, and generate similar flight numbers that are easy to confuse, so as to provide intelligent air traffic control training for flight trainees.

[0006] Specifically, the above-mentioned flight trainee training method provided according to the first aspect of the present invention includes the following steps: obtaining a first flight number provided by an instructor; determining a confusion rule corresponding to the instructor and / or the flight trainee, and generating a second flight number that is easily confused with the first flight number based on the confusion rule; and broadcasting an air traffic control training instruction containing the second flight number to the flight trainee.

[0007] Furthermore, in some embodiments of the present invention, the step of determining the confusion rules corresponding to the instructor and / or the flight trainee and generating a second flight number that is easily confused with the first flight number based on the confusion rules includes: inputting the first flight number and at least one of the confusion rules into a pre-trained conditional adversarial generative network model one by one, so as to output at least one second flight number corresponding to each of the confusion rules through the conditional adversarial generative network model.

[0008] Furthermore, in some embodiments of the present invention, the step of training the conditional adversarial generative network model includes: obtaining multiple sets of historical operation data of multiple flights. Each set of the historical operation data includes the first identity information of the corresponding instructor, the second identity information of the corresponding flight trainee, the first flight number, the flight phase information, the flight instruction text, and the corresponding second flight number; constructing the conditional adversarial generative network model; and training the conditional adversarial generative network model according to the multiple sets of historical operation data.

[0009] Further, in some embodiments of the present invention, the conditional adversarial generative network model includes a rule determination module and a flight number generation module, and the step of constructing the rule determination module includes: determining corresponding obfuscation rules according to the first flight number and the second flight number in each group of the historical operation data, and constructing an obfuscation rule library; sorting the various obfuscation rules in the obfuscation rule library in descending order according to the number of times the various obfuscation rules are used in the multiple groups of historical operation data, so as to construct a rule determination module based on the obfuscation rule library:

[0010]

[0011] in, For the The obfuscation rule that is used most times during the update cycle of the obfuscation rule library. For the The obfuscation rule that is used the least during the update cycle of the obfuscation rule library. is the number of obfuscation rules in the obfuscation rule base.

[0012] Further, in some embodiments of the present invention, the step of training the conditional adversarial generative network model based on the multiple groups of historical operation data includes: inputting the first identity information, the second identity information, the flight phase information and / or the flight instruction text in each group of the historical operation data into the rule determination module to determine the corresponding confusion rule; inputting the first flight number and the corresponding confusion rule in each group of the historical operation data into the flight number generation module to generate the corresponding flight number prediction value; determining the loss value of the conditional adversarial generative network model based on the difference between the flight number prediction value and the corresponding second flight number in the multiple groups of historical operation data; in response to the loss value being greater than a preset loss threshold, adjusting the parameters of the conditional adversarial generative network model and continuing the training of the conditional adversarial generative network model; and in response to the loss value being less than or equal to the loss threshold, determining that the training of the conditional adversarial generative network model is completed, and obtaining the conditional adversarial generative network model.

[0013] Further, in some embodiments of the present invention, the step of inputting the first identity information, the second identity information, the flight phase information and / or the flight instruction text in each group of the historical operation data into the rule determination module to determine the corresponding confusion rule includes: inputting the first identity information into the rule determination module to determine a first confusion rule corresponding to the instructor's preference; and / or inputting the second identity information into the rule determination module to determine a second confusion rule for the flight student's error-prone points; and / or inputting the flight phase information and / or the flight instruction text into the rule determination module to determine a third confusion rule for the accent of the corresponding target flight area.

[0014] Further, in some embodiments of the present invention, the conditional adversarial generative network model includes multiple local models. Each of the local models corresponds to a local rule base. The step of training the conditional adversarial generative network model according to the multiple groups of historical operation data also includes: determining the corresponding first local rule base according to the first identity information of the instructor, and training the first local model that meets the instructor's preference according to the first confusion rule output by the first local rule base; and / or determining the corresponding second local rule base according to the second identity information of the flight trainee, and training the second local model for the flight trainee's error-prone points according to the second confusion rule output by the second local rule base; and / or determining the corresponding third local rule base according to the flight phase information and / or the flight instruction text, and training the third local model for the accent of the target flight area according to the third confusion rule output by the third local rule base.

[0015] Furthermore, in some embodiments of the present invention, the step of training the conditional adversarial generative network model according to the multiple sets of historical operation data also includes: sending the learning parameters of each of the local models to the global model of the conditional adversarial generative network, and aggregating them using the entropy weighted averaging method; and performing distillation processing on the learning parameters of each of the local models through the global model, and returning the processed learning parameters to each of the local models, so that the learning parameters of each of the local models are equal to those of the global model.

[0016] Further, in some embodiments of the present invention, the step of performing distillation processing on the learning parameters of each local model via the global model includes: calculating the global model The probability distribution to each of the local models The KL divergence function of the probability distribution is:

[0017]

[0018] in, For the The data distribution probability of the local model is is the data distribution probability of the global model, ;

[0019] Using the global model Generator The generated transfer set , calculate each of the local models The output vector gap is minimized to optimize the loss function of the global model:

[0020]

[0021] in, is the behavior identification label, Use a noise vector for the generator and behavior identification labels Generated, and is the log-odds function.

[0022] Furthermore, in some embodiments of the present invention, the step of broadcasting the air traffic control training instructions including the second flight number to the flight trainee includes: obtaining the flight trainee's operation data and / or the repeated voice data; and determining the flight trainee's training performance based on the operation data and / or the repeated voice data.

[0023] In addition, the training device for the flight trainee provided according to the second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and is configured to execute the computer instructions stored in the memory to implement the training method for the flight trainee provided according to the first aspect of the present invention.

[0024] Furthermore, in some embodiments of the present invention, the training device of the flight trainee is connected to a central server. The central server is used to send the learning parameters of the pre-trained conditional adversarial generative network model and aggregate multiple local models. The training device of the flight trainee also includes a sound computer, a host and a local server. The sound computer is used to broadcast the first flight number and the second flight number that is easily confused with the first flight number to the flight trainee. The host is used to collect a variety of historical operation data of the flight and train the conditional adversarial generative network model based on it. The local server is used to obtain the learning parameters of the conditional adversarial generative network model from the central server and send the learning parameters of each of the local models to the central server.

[0025] In addition, the computer-readable storage medium provided in the third aspect of the present invention stores computer instructions, which, when executed by a processor, implement the flight trainee training method provided in the first aspect of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above features and advantages of the present invention can be better understood after reading the detailed description of the embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily drawn to scale, and components with similar related properties or features may have the same or similar reference numerals.

[0027] Figure 1 A schematic structural diagram of a flight trainee training device provided according to some embodiments of the present invention is shown.

[0028] Figure 2 A schematic flow chart of a flight trainee training method provided according to some embodiments of the present invention is shown.

[0029] Figure 3 A schematic diagram of the architecture of a conditional adversarial generative network model provided according to some embodiments of the present invention is shown.

[0030] Figure 4 A schematic diagram of an interface of an obfuscation rule library provided according to some embodiments of the present invention is shown.

[0031] Figure 5 A schematic diagram of an interface for custom obfuscation rules provided according to some embodiments of the present invention is shown.

[0032] Figure 6 A schematic diagram of the architecture of a global model and a local model provided according to some embodiments of the present invention is shown.

[0033] Figure 7 A schematic flow chart of a flight trainee training method provided according to some embodiments of the present invention is shown. DETAILED DESCRIPTION

[0034] The following specific embodiments illustrate the implementation of the present invention, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Although the description of the present invention will be introduced in conjunction with the preferred embodiment, this does not mean that the features of this invention are limited to this implementation. On the contrary, the purpose of introducing the invention in conjunction with the implementation is to cover other options or modifications that may be extended based on the claims of the present invention. In order to provide a deep understanding of the present invention, the following description will include many specific details. The present invention can also be implemented without using these details. In addition, in order to avoid confusion or blurring the focus of the present invention, some specific details will be omitted in the description.

[0035] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0036] In addition, the terms "upper", "lower", "left", "right", "top", "bottom", "horizontal" and "vertical" used in the following description should be understood as the directions shown in the paragraph and the related drawings. Such relative terms are only used for the convenience of description and do not mean that the device described therein must be manufactured or operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0037] It is understood that although the terms "first", "second", "third", etc. may be used herein to describe various components, regions, layers and / or parts, these components, regions, layers and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers and / or parts. Therefore, the first component, region, layer and / or part discussed below may be referred to as a second component, region, layer and / or part without departing from some embodiments of the present invention.

[0038] As mentioned above, in daily training, instructors often use oral methods to simulate similar flight numbers to interfere with flight trainees. However, this oral method cannot play the air-ground calls of similar flight numbers with a one-touch trigger, which greatly increases the workload of instructors. In addition, there is currently no training method that can intelligently recommend and generate similar flight numbers that are easy to confuse based on instructor preferences, flight trainees' easy mistakes, and the accents of the target flight area, and it is impossible to provide intelligent services for flight training.

[0039] In order to overcome the above-mentioned defects of the prior art, the present invention provides a training method for flight trainees, a training device for flight trainees and a computer-readable storage medium, which can extract and train rule features from historical operation data through a conditional generative adversarial network model based on distributed learning, and can be used to make intelligent recommendations based on instructor preferences, flight trainees' common error points and the accent of the target flight area, and generate similar flight numbers that are easy to confuse, so as to provide intelligent air traffic control training for flight trainees.

[0040] In some non-limiting embodiments, the training method for flight trainees provided in the first aspect of the present invention may be implemented based on the training device for flight trainees provided in the second aspect of the present invention. Figure 1 . Figure 1 A schematic structural diagram of a flight trainee training device provided according to some embodiments of the present invention is shown.

[0041] exist Figure 1 In the illustrated embodiment, the training device for the trainee pilot provided in the second aspect of the present invention is configured with a memory (not shown) and a processor 11. The memory includes but is not limited to the computer-readable storage medium provided in the third aspect of the present invention, on which computer instructions are stored. The processor 11 is connected to the memory and is configured to execute the computer instructions stored thereon to implement the training method for the trainee pilot provided in the first aspect of the present invention.

[0042] In addition, in some preferred embodiments, the above-mentioned flight trainee training device provided by the second aspect of the present invention is connected to a central server, and the central server is used to send learning parameters of a pre-trained conditional adversarial generative network model and aggregate multiple local models.

[0043] In addition, Figure 1In the illustrated embodiment, the training device for the flight trainees provided by the second aspect of the present invention further includes a voice computer 12, a host 13 and a local server (not shown). Here, the voice computer 12 is used to broadcast the first flight number and the second flight number that is easily confused with the first flight number to the flight trainees. The host 13 is used to collect a variety of historical operation data of the flight and train the conditional adversarial generation network model based on the data. The local server is used to obtain the learning parameters of the conditional adversarial generation network model from the central server and send the learning parameters of each local model to the central server.

[0044] In addition, Figure 1 In the illustrated embodiment, the training device for the above-mentioned flight trainees provided by the second aspect of the present invention may also optionally include an amplifier 14, an audio output module 15 and a repetition monitoring stimulator 16. Here, the amplifier 14 may be a digital-to-analog amplifier or a power amplifier. The audio output module 15 may be a stereo. The repetition monitoring stimulator 16 is used to implement monitoring and activate the behavior label pasting. Here, the monitoring time is set to 30s. If a repetition instruction is monitored, it is automatically marked as a correct or wrong repetition behavior label and entered into the database for storage. Here, 0-1 variables represent wrong behavior labels and correct behavior labels respectively. 0 is wrong behavior (cur state=False), indicating that the repetition sound is captured within 30s. 1 is correct behavior (cur state=True), indicating that the repetition sound is not captured within 30s.

[0045] The working principle of the training device for flight trainees described above will be described below in conjunction with some embodiments of the training method for flight trainees. Those skilled in the art will appreciate that the embodiments of the training method for flight trainees are only some non-restrictive implementations provided by the present invention, and are intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are convenient for the public to implement, rather than to limit all functions or all working modes of the training device for flight trainees. Similarly, the training device for flight trainees is also only a non-restrictive implementation provided by the present invention, and does not constitute a limitation on the execution subject or execution order of each step in the training method for flight trainees.

[0046] Please refer to the reference for details Figure 2 , Figure 3 And Table 1. Figure 2 A schematic flow chart of a flight trainee training method provided according to some embodiments of the present invention is shown. Figure 3 The schematic diagram of the architecture of the conditional adversarial generative network model provided according to some embodiments of the present invention is shown. Table 1 shows a historical operation data table provided according to some embodiments of the present invention.

[0047] Table 1 Historical operation data table

[0048] teacher Student pilot First flight number Second flight number Second flight number selection time Normal flight instructions Wang Yi Wang Er, Wang San CA1511 CA1517 10:29:06 CA1511, Direct to SOMIA Wang Yi Wang Er, Wang San CA1511 CA1517 10:29:06 null

[0049] like Figure 2 , Figure 3 As shown in Table 1, the processor 11 can pre-train the conditional adversarial generative network model offline. Specifically, in the process of training the conditional adversarial generative network model, the processor 11 can first obtain multiple sets of historical operation data of multiple flights. Here, each set of historical operation data includes the first identity information of the corresponding instructor, the second identity information of the corresponding flight student, the first flight number, the flight phase information, the flight instruction text, and the corresponding second flight number.

[0050] Afterwards, the processor 11 may construct a conditional adversarial generative network model, wherein the conditional adversarial generative network model includes a rule determination module and a flight number generation module.

[0051] Please refer to Figure 4 . Figure 4 A schematic diagram of an interface of an obfuscation rule library provided according to some embodiments of the present invention is shown.

[0052] Specifically, if Figure 4 As shown, the processor 11 may first construct a rule determination module. The processor 11 may determine the corresponding obfuscation rule according to the first flight number and the second flight number in each group of historical operation data, and construct an obfuscation rule library.

[0053] Please further refer to Table 2. Table 2 shows an obfuscation rule library table provided according to some embodiments of the present invention.

[0054] Table 2 Original rule base table

[0055] rule Notes frequency R1|"”"”34 3,4 digits are the same 0 R2|123"” 1, 2, 3 digits are the same 0 R3|12"”4 1, 2, 4 digits are the same 0 R4|1"”34 1, 3, 4 digits are the same 0 R5|"”234 2,3,4 digits are the same 0 R6|1"”"”4 1,4 digits are the same 0 R7|1223-2113 The same value after deleting duplicates 0 R8|User-defined Custom 0

[0056] Afterwards, as shown in Table 2, the processor 11 may sort the various obfuscation rules in the obfuscation rule library in descending order according to the number of times the various obfuscation rules are used in the multiple sets of historical operation data, so as to construct a rule determination module based on the obfuscation rule library:

[0057]

[0058] in, For the The most used obfuscation rules in the obfuscation rule base update cycle. For the The obfuscation rule with the least number of uses in the obfuscation rule base update cycle. is the number of obfuscation rules in the obfuscation rule base.

[0059] Thereafter, the processor may input the first identity information, the second identity information, the flight phase information and / or the flight instruction text in each group of historical operation data into the rule determination module to determine the corresponding obfuscation rule.

[0060] Specifically, the processor 11 may input the first identity information into the rule determination module to determine the first obfuscation rule corresponding to the instructor's preference. Similarly, the processor 11 may input the second identity information into the rule determination module to determine the second obfuscation rule for the error-prone points of the flight trainees. Similarly, the processor 11 may input the flight phase information and / or the flight instruction text into the rule determination module to determine the third obfuscation rule for the accent of the corresponding target flight area.

[0061] Please refer to Figure 5 . Figure 5 A schematic diagram of an interface for custom obfuscation rules provided according to some embodiments of the present invention is shown.

[0062] In addition, Figure 5 In the illustrated embodiment, the processor 11 may also determine a fourth custom confusion rule based on custom text input by the instructor and / or the flight trainee to meet the personalized settings of the instructor and the flight trainee.

[0063] Afterwards, the processor 11 may first determine the objective function of the flight number generation module:

[0064]

[0065] in, is a collection of multiple sets of historical operation data. are the features in multiple sets of historical operation data, To indicate correct or incorrect behavior, is a random noise vector, is a subset of multiple sets of historical operation data, , is the collection of all multiple sets of historical running data, express The corresponding feature space of the historical operation data in represents the potential space, , Represents the output space.

[0066] After that, if Figure 3 As shown, the processor 11 can first set the structure of the conditional adversarial generation network model The generator of the conditional adversarial generation network model, is the discriminator of the conditional adversarial generation network model. Afterwards, the processor 11 can input the first flight number in each group of historical operation data and the corresponding obfuscation rule into the flight number generation module. Using noise vector and behavior identification labels generate , discriminator Receive historical operation data and behavior judgment labels Pseudo samples of . Then, the generator and the discriminator According to the above pseudo sample , generate the corresponding flight number prediction value, and determine the loss value of the conditional adversarial generation network model based on the difference between the flight number prediction value and the corresponding second flight number in multiple groups of historical operation data.

[0067] Thereafter, in response to the loss value being greater than a preset loss threshold, the processor 11 may adjust the parameters of the conditional adversarial generative network model and continue training the conditional adversarial generative network model.

[0068] Thereafter, in response to the loss value being less than or equal to the loss threshold, the processor 11 may determine that the training of the conditional adversarial generative network model is completed and obtain the conditional adversarial generative network model.

[0069] Please refer to Figure 6 . Figure 6 A schematic diagram of the architecture of a global model and a local model provided according to some embodiments of the present invention is shown.

[0070] In addition, Figure 6 In the illustrated embodiment, the above-mentioned conditional adversarial generative network model has multiple local models. Here, each local model corresponds to a local rule base. In the process of training the conditional adversarial generative network model, the processor 11 can also determine the corresponding first local rule base according to the first identity information of the instructor, and train the first local model that meets the instructor's preference according to the first confusion rule output by the first local rule base. Alternatively, the processor 11 can determine the corresponding second local rule base according to the second identity information of the flight trainee, and train the second local model for the flight trainee's error-prone points according to the second confusion rule output by the second local rule base. Alternatively, the processor 11 can determine the corresponding third local rule base according to the flight phase information and / or the flight instruction text, and train the third local model for the accent of the target flight area according to the third confusion rule output by the third local rule base.

[0071] Afterwards, the processor 11 can send the learning parameters of each local model to the global model of the conditional adversarial generative network model and aggregate them using the entropy weighted averaging method.

[0072] Thereafter, the processor 11 may perform distillation processing on the learning parameters of each local model via the global model, and return the processed learning parameters to each local model so that the learning parameters of each local model are equal to those of the global model.

[0073] Afterwards, the processor 11 can calculate the global model The probability distribution to each local model The KL divergence function of the probability distribution is:

[0074]

[0075] in, For the The data distribution probability of the local model is is the data distribution probability of the global model, , so that the data distribution probability of the global model is the same as that of the local model.

[0076] Afterwards, the processor 11 can use the global model Generator The generated transfer set , calculate each local model The output vector gap is calculated and minimized to optimize the loss function of the global model:

[0077]

[0078] in, is the behavior identification label, Using a noise vector for the generator and behavior identification labels Generated, and is the log-odds function.

[0079] In this way, the processor 11 can use the generator to aggregate and identify the historical operating data in the local simulator, and generate samples with similar data distribution, so as to achieve the same distribution of the model parameter settings of the central server and the model parameters in the local server, thereby solving the problem of a large number of simulators and complex training elements. In addition, each local model is independent of each other, and the training of each local model can be decoupled from the global one, so as to adjust the training algorithm and network model structure according to local goals, thereby allowing targeted training of local models according to multiple data sources. In addition, by training local models within each ontology server and only sending the trained learning parameters to the external global model, it is possible to maintain a certain degree of agnosticism for the local training algorithm and model structure to ensure data security.

[0080] In addition, in some preferred embodiments, the processor 11 may also use the potential space to construct an induced distribution G* to alleviate data security during the training process of the conditional adversarial generation network model to prevent data leakage:

[0081]

[0082]

[0083] in, is the expected value of the loss function, is the characteristic function of the noise vector z.

[0084] Afterwards, the processor 11 can construct a generator according to the above induced distribution G* The optimization goal is:

[0085]

[0086] in, The target label for the second flight number of the generator The characteristics generated.

[0087] In this way, by executing the above training steps, the present invention can obtain a conditional generative adversarial network model based on distributed learning, which is used to optimize and update each confusion rule of the confusion rule library according to instructor preferences, flight trainees' common error points and the accent of the target flight area.

[0088] Furthermore, based on the conditional adversarial generative network model obtained by training in the above-mentioned embodiment, the above-mentioned flight trainee training device provided by the present invention can generate at least one second flight number corresponding to the instructor's preferences, the flight trainee's common error points and the accent of the target flight area for the first flight number provided by the instructor, so as to conduct air traffic control training for the flight trainee.

[0089] Please refer to Figure 7 . Figure 7 A schematic flow chart of a flight trainee training method provided according to some embodiments of the present invention is shown.

[0090] like Figure 7 As shown, during the air traffic control training for the flight trainee, the processor 11 may first obtain the first flight number provided by the instructor. Afterwards, the processor 11 may input the first flight number into the pre-trained one to determine the confusion rule corresponding to the instructor and / or the flight trainee, and generate at least one second flight number corresponding to each confusion rule according to the first flight number and the corresponding at least one confusion rule.

[0091] Thereafter, the processor 11 may send the second flight number to the sound computer 12, the amplifier 14, the audio output module 15 and the repeat monitoring activator 16 via the UDP or TCP / IP protocol, and broadcast the air traffic control training (ATC) instructions including the second flight number to the flight trainees.

[0092] Specifically, the processor 11 can obtain the flight trainee's operation data and / or repetition voice data, and determine the flight trainee's training performance based on the above operation data and / or repetition voice data, so as to examine the flight trainee's monitoring ability and repetition ability in the ground-to-air communication scenario.

[0093] In summary, the above-mentioned flight trainee training method, flight trainee training device and computer-readable storage medium provided by the present invention can all extract and train rule features of historical operation data through a conditional generative adversarial network model based on distributed learning, and can be used to make intelligent recommendations based on instructor preferences, flight trainees' common error points and the accent of the target flight area, and generate similar flight numbers that are easy to confuse, so as to provide flight trainees with intelligent air traffic control training, thereby examining the flight trainees' listening and reciting abilities in land-to-air communication scenarios.

[0094] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0095] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein, but should be granted the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for training flight trainees, characterized in that: The following steps are involved: Get the first flight number provided by the instructor; Determine a confusion rule corresponding to the instructor and / or the flight trainee, and input the first flight number and at least one of the confusion rules into a pre-trained conditional adversarial generative network model one by one, so as to generate at least one second flight number that is easily confused with the first flight number through the conditional adversarial generative network model, wherein: The step of training the conditional adversarial generative network model comprises: obtaining multiple groups of historical operation data of multiple flights, each group of the historical operation data respectively includes the first identity information of the corresponding instructor, the second identity information of the corresponding flight student, the first flight number, the flight phase information, the flight instruction text, and the corresponding second flight number; constructing the conditional adversarial generative network model; and training the conditional adversarial generative network model according to the multiple groups of historical operation data, The conditional adversarial generative network model includes a rule determination module and a flight number generation module. The step of constructing the rule determination module includes: determining corresponding obfuscation rules according to the first flight number and the second flight number in each group of the historical operation data, and constructing an obfuscation rule library; sorting the various obfuscation rules in the obfuscation rule library in descending order according to the number of times the various obfuscation rules are used in the multiple groups of historical operation data, so as to construct a rule determination module based on the obfuscation rule library: in, For the The most frequently used obfuscation rules in the obfuscation rule base update cycle, For the The obfuscation rule that is used the least during the update cycle of the obfuscation rule library. is the number of obfuscation rules in the obfuscation rule base; and Broadcast an air traffic control training instruction including the second flight number to the student pilot.

2. The training method according to claim 1, characterized in that: The step of training the conditional adversarial generative network model according to the multiple groups of historical operation data comprises: Inputting the first identity information, the second identity information, the flight phase information and / or the flight instruction text in each group of the historical operation data into the rule determination module to determine the corresponding obfuscation rule; Inputting the first flight number in each group of the historical operation data and the corresponding confusion rule into the flight number generation module to generate a corresponding flight number prediction value; Determining a loss value of the conditional adversarial generative network model according to a difference between the predicted flight number and a corresponding second flight number in the multiple sets of historical operation data; In response to the loss value being greater than a preset loss threshold, adjusting parameters of the conditional adversarial generative network model, and continuing training of the conditional adversarial generative network model; and In response to the loss value being less than or equal to the loss threshold, it is determined that the training of the conditional adversarial generative network model is completed, and the conditional adversarial generative network model is obtained.

3. The training method according to claim 2, characterized in that: The step of inputting the first identity information, the second identity information, the flight phase information and / or the flight instruction text in each group of the historical operation data into the rule determination module to determine the corresponding obfuscation rule comprises: inputting the first identity information into the rule determination module to determine a first obfuscation rule corresponding to the instructor's preference; and / or inputting the second identity information into the rule determination module to determine a second confusion rule for the error-prone points of the flight student; and / or The flight phase information and / or the flight instruction text are input into the rule determination module to determine a third obfuscation rule for the accent of the corresponding target flight area.

4. The training method according to claim 3, characterized in that: The conditional generative adversarial network model includes a plurality of local models, wherein each of the local models corresponds to a local rule base, and the step of training the conditional generative adversarial network model according to the plurality of sets of historical operation data further includes: determining a corresponding first local rule base according to the first identity information of the instructor, and training a first local model that meets the instructor's preference according to a first confusion rule output by the first local rule base; and / or Determine a corresponding second local rule base according to the second identity information of the flight trainee, and train a second local model for the error-prone points of the flight trainee according to the second confusion rule output by the second local rule base; and / or A corresponding third local rule base is determined according to the flight phase information and / or the flight instruction text, and a third local model for the accent of the target flight area is trained according to a third confusion rule output by the third local rule base.

5. The training method according to claim 4, characterized in that: The step of training the conditional adversarial generative network model according to the multiple groups of historical operation data also includes: Sending the learning parameters of each local model to the global model of the conditional adversarial generative network and aggregating them using an entropy weighted averaging method; and The learning parameters of each of the local models are distilled via the global model, and the processed learning parameters are returned to each of the local models so that the learning parameters of each of the local models are equal to those of the global model.

6. The training method according to claim 5, characterized in that: The step of performing distillation processing on the learning parameters of each of the local models via the global model comprises: Calculate the global model The probability distribution to each of the local models The KL divergence function of the probability distribution is: in, For the The data distribution probability of the local model is is the data distribution probability of the global model, ; Using the global model Generator The generated transfer set , calculate each of the local models The output vector gap is minimized to optimize the loss function of the global model: in, is the behavior identification label, Use a noise vector for the generator and behavior identification labels Generated, and is the log-odds function.

7. The training method according to claim 1, characterized in that: The step of broadcasting the air traffic control training instruction including the second flight number to the flight trainee: Acquiring the flight student's operation data and / or recitation voice data; and The training performance of the student pilot is determined based on the operation data and / or the repeated voice data.

8. A training device for flight trainees, characterized in that: include: a memory having computer instructions stored thereon; as well as A processor is connected to the memory and is configured to execute computer instructions stored in the memory to implement the flight student training method as described in any one of claims 1 to 7.

9. The training device according to claim 8, characterized in that The training device of the flight trainee is connected to a central server, and the central server is used to send learning parameters of a pre-trained conditional adversarial generative network model and aggregate multiple local models. The training device of the flight trainee also includes: a voice computer, used for announcing to the flight trainee a first flight number and a second flight number that is easily confused with the first flight number; A host computer, which is used to collect various historical operation data of the flight and train the conditional adversarial generation network model based on the data; and The local server is used to obtain the learning parameters of the conditional adversarial generation network model from the central server and send the learning parameters of each local model to the central server.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by the processor, the flight trainee training method as described in any one of claims 1 to 7 is implemented.

Images