An aviation aircraft satellite communication fault diagnosis and repair method and system based on deep learning

Abstract

The application relates to a kind of deep learning-based aircraft satellite communication fault diagnosis and repair method and system, belong to satellite fault diagnosis technical field.The system includes: obtaining communication satellite topology network, the observation parameter of communication satellite topology network is collected by dual-mode heterogeneous receiving terminal and data archiving is generated constellation measurement dataset;According to constellation measurement dataset and communication satellite topology network, abnormal node positioning information is obtained by star group behavior perception model;According to abnormal node positioning information, abnormal mode reconstruction model is used to obtain inter-satellite communication abnormal mode;According to inter-satellite communication abnormal mode, fault repair is carried out.The deep learning-based aircraft satellite communication fault diagnosis and repair are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of satellite fault diagnosis, and particularly relates to a satellite communication fault diagnosis and repair method and system for an aerial vehicle based on deep learning. BACKGROUND

[0002] With the continuous development of global informatization, digitization and aerospace technology, the traditional single satellite communication system has been difficult to meet the increasingly complex application scenarios and efficient service requirements. Under this background, the construction of multi-satellite communication system has become one of the core directions of modern aerospace technology, and plays an irreplaceable role in many important fields. However, with the significant increase in the number of satellites, the probability of simultaneous failure of multiple satellites also increases, and how to accurately diagnose the fault types of multiple satellites in time has become an important challenge in satellite communication fault repair.

[0003] Traditional fault detection methods usually rely on single satellite fault diagnosis, but in a multi-satellite system, this method often cannot effectively handle the case of multiple satellites failing at the same time, which can easily lead to misdiagnosis or missed diagnosis. In addition, due to the scarcity of fault sample data in satellite systems, and the strong concealment and diversity of fault types, historical operation data cannot provide enough information to support accurate fault classification and positioning. This makes it impossible to identify the fault source and locate the fault type in time in actual application, further increasing the complexity of satellite communication fault repair. Therefore, how to accurately identify the unknown fault type of multi-satellite communication is a problem that needs to be solved. SUMMARY

[0004] To solve the above problems in the prior art, the application provides a satellite communication fault diagnosis and repair method and system for an aerial vehicle based on deep learning,

[0005] The object of the application can be achieved by the following technical solutions:

[0006] A satellite communication fault diagnosis and repair method for an aerial vehicle based on deep learning, comprising:

[0007] Obtaining a communication satellite topology network, collecting observation parameters of the communication satellite topology network through a dual-mode heterogeneous receiving terminal and generating a constellation measurement data set by data archiving;

[0008] Obtaining abnormal node positioning information through a star cluster behavior perception model according to the constellation measurement data set and the communication satellite topology network;

[0009] Obtaining an inter-satellite communication abnormal mode through an abnormal mode reconstruction model according to the abnormal node positioning information, and performing fault repair according to the inter-satellite communication abnormal mode.

[0010] Preferably, the constellation behavior perception model comprises:

[0011] According to the constellation measurement dataset and the communication satellite topology network, an initial network topology, positioning parameters, and a candidate abnormal node are obtained through preliminary fault detection;

[0012] According to the initial network topology, the candidate abnormal node, and the positioning parameters, abnormal node positioning information is obtained through back substitution detection.

[0013] Preferably, the preliminary fault detection specifically comprises:

[0014] The measurement data in the measurement dataset and the predicted pseudo-range are obtained, and a pseudo-range deviation is obtained through residual calculation according to the measurement data and the predicted pseudo-range;

[0015] The pseudo-range deviation is normalized to obtain a normalized pseudo-range deviation;

[0016] According to the normalized pseudo-range deviation, the maximum normalized pseudo-range deviation is obtained through sorting, and the satellite corresponding to the maximum normalized pseudo-range deviation is marked as a candidate abnormal node;

[0017] According to the candidate abnormal node, the maximum correlation coefficient is obtained through correlation calculation and sorting; a preset correlation threshold is set, if the maximum correlation coefficient is greater than the correlation threshold, the satellite corresponding to the maximum correlation is a candidate abnormal node; if the maximum correlation is not greater than the correlation threshold, the satellite corresponding to the maximum correlation is a normal node;

[0018] The initial network topology is obtained by excluding the candidate abnormal node in the communication satellite topology network; and the positioning parameters are obtained according to the initial network topology.

[0019] Preferably, the back substitution detection comprises:

[0020] According to the candidate abnormal node and the positioning parameters, a global detection statistic is obtained through global statistic calculation;

[0021] A global threshold is preset, and a final candidate abnormal node-free seat is obtained through recovery judgment according to the global detection statistic and the global threshold.

[0022] The recovery judgment comprises judging whether the global detection statistic is less than the global threshold, if yes, the candidate abnormal node is recovered to a normal node; if no, the candidate abnormal node is still a candidate abnormal node.

[0023] According to the final candidate abnormal node-free seat, the abnormal node positioning information is obtained through least square estimation.

[0024] Preferably, the abnormal mode reconstruction model comprises:

[0025] obtaining failure satellite telemetry data, and obtaining latent space representation according to the failure satellite telemetry data through a compression model;

[0026] a probability model;

[0027] obtaining the inter-satellite communication anomaly pattern according to the distribution inner likelihood rate through a scoring model.

[0028] Preferably, the compression model comprises:

[0029] obtaining failure satellite telemetry reduced dimension data by dimensionality reduction on the failure satellite telemetry data;

[0030] obtaining the latent space representation according to the failure satellite telemetry reduced dimension data through a low dimension representation optimization model.

[0031] Preferably, the low dimension representation optimization model comprises:

[0032] obtaining sample initial low dimension representation according to the failure satellite telemetry reduced dimension data through a reconstruction loss function minimization;

[0033] The reconstruction loss function is expressed as:

[0034] ,

[0035] wherein, L rec the reconstruction loss function is represented as, N the number of failure satellite telemetry reduced dimension data is represented as, i the i-th failure satellite telemetry reduced dimension data is represented as, x i the failure satellite telemetry reduced dimension is represented as, Attention( ,a i ) the attention mechanism is represented as, the data after the reconstruction loss function is represented as, a i the attention weight is represented as, λ is a regularization parameter, j the j-th weight is represented as, M the number of weights is represented as, w j the weight is represented as;

[0036] obtaining the latent space representation according to the sample initial low dimension representation through a contrastive loss function calculation.

[0037] Preferably, the probability model comprises:

[0038] a component type probability is obtained from the latent space representation by an artificial neural network;

[0039] a complete mixture component parameter is obtained from the component type probability by a maximization computation, the complete mixture component parameter including a weight parameter, a mean vector parameter, a covariance matrix parameter;

[0040] the in-distribution likelihood is obtained from the complete mixture component parameter and the latent space representation by a probability computation;

[0041] the probability computation expression is:

[0042] ,

[0043] wherein, P(x) the in-distribution likelihood, c denotes the cth complete mixture component parameter, C denotes the number of complete mixture component parameters, is a weight, denotes the determinant of the covariance matrix of the cth complete mixture component, tr denotes a trace operation, x denotes the latent space representation, μ c is a mean vector, T denotes a transposition computation.

[0044] Preferably, the scoring model includes:

[0045] an out-of-distribution likelihood value is obtained from the in-distribution likelihood by a scoring computation;

[0046] the scoring computation expression is:

[0047] ,

[0048] wherein, Score(x out ) the out-of-distribution likelihood value, c denotes the cth complete mixture component parameter, C denotes the number of complete mixture component parameters, is a weight, denotes the determinant of the covariance matrix of the cth complete mixture component, x in denotes the latent space representation, μ c is a mean vector, T denotes a transposition computation, the out-of-distribution likelihood value being obtained from the in-distribution likelihood by a negative logarithm computation;

[0049] a preset score threshold, and determining the inter-satellite communication abnormal mode according to the score threshold and the out-of-distribution likelihood value;

[0050] The score determination includes determining whether the out-of-distribution likelihood value is greater than the score threshold. If yes, the inter-satellite communication abnormal mode is of an unknown type. If no, the inter-satellite communication abnormal mode is of a known type.

[0051] A deep learning-based aviation aircraft satellite communication fault diagnosis and repair system, comprising a data acquisition module, a fault detection module, and a type diagnosis module:

[0052] The data acquisition module is configured to acquire a communication satellite topology network, collect observation parameters of the communication satellite topology network through a dual-mode heterogeneous receiving terminal, and generate a constellation measurement data set by archiving data;

[0053] The fault detection module is configured to obtain abnormal node positioning information according to the constellation measurement data set and the communication satellite topology network through a star cluster behavior perception model;

[0054] The type diagnosis module is configured to obtain an inter-satellite communication abnormal mode according to the abnormal node positioning information through an abnormal mode reconstruction model, and perform fault repair according to the inter-satellite communication abnormal mode;

[0055] The star cluster behavior perception model comprises:

[0056] According to the constellation measurement data set and the communication satellite topology network, an initial network topology, positioning parameters, and candidate abnormal nodes are obtained through preliminary fault detection;

[0057] According to the initial network topology, the candidate abnormal nodes, and the positioning parameters, the abnormal node positioning information is obtained through back substitution detection;

[0058] The preliminary fault detection specifically includes:

[0059] The measurement data in the measurement data set and the predicted pseudo-range are obtained, and the pseudo-range deviation is calculated according to the measurement data and the predicted pseudo-range;

[0060] The pseudo-range deviation is normalized to obtain a normalized pseudo-range deviation;

[0061] According to the normalized pseudo-range deviation, the maximum normalized pseudo-range deviation is obtained by sorting, and the beacon corresponding to the maximum normalized pseudo-range deviation is marked as a candidate abnormal node;

[0062] According to the candidate abnormal node, a maximum correlation coefficient is calculated and sorted by correlation; a preset correlation threshold is set, if the maximum correlation coefficient is greater than the correlation threshold, the satellite corresponding to the maximum correlation is the candidate abnormal node; if the maximum correlation is not greater than the correlation threshold, the satellite corresponding to the maximum correlation is a normal node;

[0063] The initial network topology is obtained by excluding the candidate abnormal node in the communication satellite topology network; and the positioning parameter is obtained according to the initial network topology;

[0064] The back substitution detection includes:

[0065] According to the candidate abnormal node and the positioning parameter, a global detection statistic is calculated by a global statistic;

[0066] A global threshold is preset, and the final candidate abnormal node is obtained by recovery judgment according to the global detection statistic and the global threshold;

[0067] The recovery judgment includes judging whether the global detection statistic is less than the global threshold, if yes, the candidate abnormal node is recovered to a normal node; if no, the candidate abnormal node is still a candidate abnormal node;

[0068] According to the final candidate abnormal node, the abnormal node positioning information is obtained by least square estimation.

[0069] The beneficial effects of the present application are:

[0070] (1) Through the star cluster behavior perception model and deep learning technology, the position of the faulty satellite can be accurately identified from the satellite measurement data, and the determination of the abnormal node positioning information is further optimized through the back substitution detection, thereby improving the accuracy and reliability of the fault detection.

[0071] (2) Through low-dimensional representation and probability model calculation of the fault satellite telemetry data, the system can efficiently diagnose the inter-satellite communication abnormal mode. Combined with the contrast loss function optimization and component type probability analysis of deep learning, the system can determine the fault type in a short time, improve the efficiency and scalability of diagnosis, and provide timely decision support for subsequent repair work.

[0072] (3) Through the distribution outside likelihood value calculated by the scoring model, the system can detect unknown faults, and combined with the preset scoring threshold, unknown obstacles can be identified and corresponding repair measures can be taken, thereby improving the accuracy and efficiency of satellite communication fault repair. BRIEF DESCRIPTION OF DRAWINGS

[0073] In order to facilitate the understanding of those skilled in the art, the present application will be further described below with reference to the accompanying drawings.

[0074] Figure 1 A flowchart of a deep learning-based aircraft satellite communication fault diagnosis and repair method of the present application. DETAILED DESCRIPTION

[0075] To further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purposes, the specific embodiments, structures, features and effects according to the present application are described in detail below in combination with the drawings and preferred embodiments.

[0076] Please refer to Figure 1 A deep learning-based aircraft satellite communication fault diagnosis and repair method, comprising:

[0077] Obtaining a communication satellite topology network, collecting observation parameters of the communication satellite topology network through a dual-mode heterogeneous receiving terminal and generating constellation measurement data sets by data archiving;

[0078] Obtaining abnormal node positioning information through a constellation behavior perception model according to the constellation measurement data sets and the communication satellite topology network;

[0079] Obtaining an inter-satellite communication abnormal mode through an abnormal mode reconstruction model according to the abnormal node positioning information, and performing fault repair according to the inter-satellite communication abnormal mode.

[0080] Specifically, the dual-mode heterogeneous receiving terminal includes GPS L1 and BDS B1; the constellation measurement data includes pseudo-range; the constellation represents a satellite network, and the satellites in the satellite network cooperate with each other according to specific orbits and configurations.

[0081] In this embodiment, the constellation is composed of 24 satellites distributed on 6 orbital planes, 4 satellites per plane, with an angular interval of 60 degrees between satellites to ensure global coverage. All satellites are located in medium earth orbits and have a relative distance of 3600 kilometers. Each satellite is responsible for specific tasks, including global navigation positioning, communication, weather monitoring, disaster warning, and precise time synchronization.

[0082] Specifically, the constellation behavior perception model includes:

[0083] S201: Obtaining an initial network topology, positioning parameters, and candidate abnormal nodes through preliminary fault detection according to the constellation measurement data sets and the communication satellite topology network;

[0084] S202: Obtaining the abnormal node positioning information through back substitution detection according to the initial network topology, the candidate abnormal nodes, and the positioning parameters.

[0085] Specifically, the preliminary fault detection specifically includes:

[0086] S201-1: Obtain measurement data in the measurement data set and predicted pseudo-range, and calculate pseudo-range deviation from the measurement data and the predicted pseudo-range;

[0087] S201-2: Obtain normalized pseudo-range deviation by normalizing the pseudo-range deviation;

[0088] S201-3: Obtain the maximum normalized pseudo-range deviation from the normalized pseudo-range deviation by sorting, and mark the satellite corresponding to the maximum normalized pseudo-range deviation as a candidate abnormal node;

[0089] S201-4: Obtain the maximum correlation coefficient from the candidate abnormal node by correlation calculation and sorting; preset a correlation threshold, if the maximum correlation coefficient is greater than the correlation threshold, the satellite corresponding to the maximum correlation is a candidate abnormal node; if the maximum correlation is not greater than the correlation threshold, the satellite corresponding to the maximum correlation is a normal node;

[0090] S201-5: Obtain the initial network topology by excluding the candidate abnormal node in the communication satellite topology network; obtain the positioning parameter according to the initial network topology.

[0091] Specifically, the back substitution detection includes:

[0092] S202-1: Obtain global detection statistics from the candidate abnormal node and the positioning parameter by global statistics calculation;

[0093] The global statistics calculation expression is:

[0094] ,

[0095] Wherein, S is the global detection statistics, T denotes the transpose operation, denotes the positioning parameter, is the pseudo-range deviation, ρ denotes the measurement data, l denotes the distance between the receiver position and the satellite position, c denotes the speed of light, δt s is the satellite clock error, δt u is the receiver clock error, is the ionospheric delay, is the tropospheric delay;

[0096] S202-2: preset a global threshold, and obtaining a final non-candidate abnormal node seat through recovery judgment according to the global detection statistic and the global threshold;

[0097] The recovery judgment comprises judging whether the global detection statistic is less than the global threshold. If yes, the candidate abnormal node is recovered to a normal node; and if no, the candidate abnormal node is still a candidate abnormal node.

[0098] S202-3: obtaining the abnormal node positioning information through least square estimation according to the final non-candidate abnormal node seat.

[0099] Specifically, the abnormal mode reconstruction model comprises:

[0100] S301: obtaining fault satellite telemetry data, and obtaining a potential space representation through a compression model according to the fault satellite telemetry data;

[0101] S302: a probability model;

[0102] S303: obtaining the inter-satellite communication abnormal mode through a scoring model according to the distribution internal likelihood.

[0103] Specifically, the compression model comprises:

[0104] S301-1: obtaining fault satellite telemetry dimension reduction data by dimension reduction on the fault satellite telemetry data;

[0105] S301-2: obtaining the potential space representation through a low-dimensional representation optimization model according to the fault satellite telemetry dimension reduction data.

[0106] Specifically, the low-dimensional representation optimization model comprises:

[0107] obtaining a sample initial low-dimensional representation by minimizing a reconstruction loss function according to the fault satellite telemetry dimension reduction data;

[0108] The reconstruction loss function is expressed as:

[0109] ,

[0110] wherein, L rec the reconstruction loss function is represented as, N the number of fault satellite telemetry dimension reduction data is represented as, i the i-th fault satellite telemetry dimension reduction data is represented as, x i the fault satellite telemetry dimension reduction is represented as, Attention( ,a i )denotes an attention mechanism, denotes data passed through a reconstruction loss function, a i denotes an attention weight, λ is a regularization parameter, j denotes the j-th weight, M denotes the number of weights, w j denotes a weight;

[0111] The latent space representation is calculated according to the initial low-dimensional representation of the sample through a contrastive loss function.

[0112] Specifically, the probability model comprises:

[0113] S302-1: obtaining a component type probability through an artificial neural network according to the latent space representation;

[0114] S302-2: obtaining a complete mixed component parameter through maximization calculation according to the component type probability;

[0115] S302-3: obtaining the intra-distribution likelihood rate through probability calculation according to the complete mixed component parameter and the latent space representation;

[0116] The probability calculation expression is:

[0117] ,

[0118] wherein, P(x) is the intra-distribution likelihood rate, c denotes the c-th complete mixed component parameter, C denotes the number of complete mixed component parameters, is a weight, denotes the determinant of the covariance matrix of the c-th complete mixed component, tr denotes a trace operation, x denotes the latent space representation, μ c is a mean vector, T denotes a transpose calculation;

[0119] Specifically, the complete mixed component parameter comprises a weight parameter, a mean vector parameter, and a covariance matrix parameter.

[0120] In this embodiment, the artificial neural network comprises three layers of hidden layers, the first layer of hidden layers has 128 neurons; the second layer of hidden layers has 64 neurons; the third layer of hidden layers has 32 neurons; the hidden layers use a ReLU activation function; and the output layer uses a softmax activation function.

[0121] In particular, the scoring model comprises:

[0122] S303-1: calculating a distribution-out likelihood value according to the distribution-in likelihood rate through scoring calculation;

[0123] The scoring calculation expression is:

[0124]

[0125] wherein, Score(x out ) is the distribution-out likelihood value, c denotes the cth complete mixture component parameter, C denotes the number of complete mixture component parameters, is a weight, denotes the determinant of the covariance matrix of the cth complete mixture component, x in denotes the latent space representation, μ c is a mean vector, T denotes a transpose calculation, and the distribution-out likelihood value is calculated by negative logarithm calculation on the distribution-in likelihood rate;

[0126] S303-2: presetting a scoring threshold, and judging the inter-satellite communication abnormal mode through scoring according to the distribution-out likelihood value and the scoring threshold;

[0127] The scoring judgment comprises judging whether the distribution-out likelihood value is greater than the scoring threshold. If yes, the inter-satellite communication abnormal mode is an unknown type; if no, the inter-satellite communication abnormal mode is a known type.

[0128] In this embodiment, the system comprises a sensor, an inferrer, an executor, an HMI interface, a communication controller, and a ground receiver. The sensor monitors satellite environment information and state data in real time and stores them in a database; the inferrer analyzes satellite faults based on a constellation behavior perception model and an abnormal mode reconstruction model, identifies fault types and positions, and transmits the results to the executor. After receiving the fault information, the executor feeds it back to the communication controller, which sends the message to the ground receiver and transmits the repair results to the executor to guide subsequent repair operations. The HMI interface displays satellite state, fault information, and repair progress through a human-machine interface for operators to monitor and make decisions.

[0129] An aviation aircraft satellite communication fault diagnosis and repair system based on deep learning comprises a data acquisition module, a fault detection module, and a type diagnosis module.

[0130] ​The data collection module is configured to acquire a communication satellite topology network, collect observation parameters of the communication satellite topology network through a dual-mode heterogeneous receiving terminal, and generate a constellation measurement data set by archiving data;

[0131] The fault detection module is configured to obtain abnormal node positioning information through a constellation behavior perception model according to the constellation measurement data set and the communication satellite topology network;

[0132] The type diagnosis module is configured to obtain an inter-satellite communication abnormal mode through an abnormal mode reconstruction model according to the abnormal node positioning information, and perform fault repair according to the inter-satellite communication abnormal mode;

[0133] The constellation behavior perception model comprises:

[0134] The preliminary fault detection is configured to obtain an initial network topology, positioning parameters, and a candidate abnormal node according to the constellation measurement data set and the communication satellite topology network;

[0135] The back substitution detection is configured to obtain the abnormal node positioning information according to the initial network topology, the candidate abnormal node, and the positioning parameters;

[0136] The preliminary fault detection specifically comprises:

[0137] The measurement data and the predicted pseudo-range in the measurement data set are acquired, and a pseudo-range deviation is calculated according to the measurement data and the predicted pseudo-range;

[0138] The pseudo-range deviation is normalized to obtain a normalized pseudo-range deviation;

[0139] The maximum normalized pseudo-range deviation is obtained by sorting according to the normalized pseudo-range deviation, and the satellite corresponding to the maximum normalized pseudo-range deviation is marked as a candidate abnormal node;

[0140] The maximum correlation coefficient is obtained by correlation calculation and sorting according to the candidate abnormal node; a preset correlation threshold is set, if the maximum correlation coefficient is greater than the correlation threshold, the satellite corresponding to the maximum correlation is a candidate abnormal node; if the maximum correlation is not greater than the correlation threshold, the satellite corresponding to the maximum correlation is a normal node;

[0141] The initial network topology is obtained by excluding the candidate abnormal node in the communication satellite topology network; and the positioning parameters are obtained according to the initial network topology;

[0142] The back substitution detection comprises:

[0143] The global detection statistic is calculated according to the candidate abnormal node and the positioning parameters through a global statistic calculation;

[0144] a preset global threshold, and a final non-candidate abnormal node seat is obtained according to the global detection statistic and the global threshold through the recovery judgment;

[0145] The recovery judgment comprises judging whether the global detection statistic is less than the global threshold, if yes, the candidate abnormal node is recovered to a normal node; if no, the candidate abnormal node is still a candidate abnormal node.

[0146] The abnormal node positioning information is obtained according to the final non-candidate abnormal node seat through least square estimation.

[0147] The computer storage medium of the embodiment of the application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component.

[0148] The computer readable signal medium can include a data signal propagating in a baseband or as part of a carrier wave propagating through a transmission medium, in which the computer readable program code is embodied. Such a propagating data signal can take many forms, including but not limited to electro-magnetic, optical or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can send, propagate or transmit a program for use by or in connection with an instruction execution system, apparatus or component.

[0149] The computer readable media on which the program code can be carried by any suitable medium, including but not limited to wireless, wired, optical fiber cable, RF, and the like, or any suitable combination of these. Computer program code for carrying out operations of the present application can be written in one or more programming languages, or combinations of languages, including object oriented, such as Java, Smalltalk, C++, and conventional procedural, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0150] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed as above with the preferred embodiment, it is not intended to limit the present application, and any person skilled in the art can make some changes or modifications to the equivalent embodiments within the scope of the technical solution of the present application, without departing from the technical solution of the present application. Any simple modification, equivalent change and modification of the above embodiments, which does not depart from the technical solution of the present application, and is based on the technical essence of the present application, still belongs to the scope of the technical solution of the present application.

Claims

1. A deep learning-based method for diagnosing and repairing satellite communication faults of an aerial vehicle, characterized in that, The method comprises the following steps: acquiring a communication satellite topology network, collecting observation parameters of the communication satellite topology network by a dual-mode heterogeneous receiving terminal, and archiving data to generate constellation measurement data sets; obtaining abnormal node positioning information by a constellation behavior perception model according to the constellation measurement data sets and the communication satellite topology network; obtaining an inter-satellite communication abnormal mode by an abnormal mode reconstruction model according to the abnormal node positioning information; performing fault repair according to the inter-satellite communication abnormal mode; the constellation behavior perception model comprises: obtaining an initial network topology, positioning parameters, and candidate abnormal nodes by preliminary fault detection according to the constellation measurement data sets and the communication satellite topology network; obtaining the abnormal node positioning information by back substitution detection according to the initial network topology, the candidate abnormal nodes, and the positioning parameters; the preliminary fault detection specifically comprises: acquiring measurement data and predicted pseudo-range in the measurement data sets, and obtaining pseudo-range deviation by residual calculation according to the measurement data and the predicted pseudo-range; obtaining normalized pseudo-range deviation by normalizing the pseudo-range deviation; obtaining the maximum normalized pseudo-range deviation by sorting according to the normalized pseudo-range deviation, and marking the satellite corresponding to the maximum normalized pseudo-range deviation as a candidate abnormal node; obtaining the maximum correlation coefficient by correlation calculation and sorting according to the candidate abnormal node; a preset correlation threshold is set, if the maximum correlation coefficient is greater than the correlation threshold, the satellite corresponding to the maximum correlation is a candidate abnormal node; if the maximum correlation is not greater than the correlation threshold, the satellite corresponding to the maximum correlation is a normal node; obtaining the initial network topology by excluding the candidate abnormal node in the communication satellite topology network; and obtaining the positioning parameters according to the initial network topology; the back substitution detection comprises: obtaining global detection statistics by global statistics calculation according to the candidate abnormal node and the positioning parameters; a global threshold is preset, and the final candidate abnormal node-free seat is obtained by recovery judgment according to the global detection statistics and the global threshold; the recovery judgment comprises judging whether the global detection statistics is less than the global threshold, if yes, the candidate abnormal node is restored to a normal node; if no, the candidate abnormal node is still a candidate abnormal node; the abnormal node positioning information is obtained by least square estimation according to the final candidate abnormal node-free seat.

2. The deep learning based aviation aircraft satellite communication fault diagnosis and repair method according to claim 1, characterized in that, the abnormal mode reconstruction model comprises: acquiring fault satellite telemetry data, and obtaining potential space representation by a compression model according to the fault satellite telemetry data; a probability model; the inter-satellite communication abnormal mode is obtained by a scoring model according to the distribution likelihood.

3. The deep learning based aviation aircraft satellite communication fault diagnosis and repair method according to claim 2, characterized in that, the compression model comprises: obtaining fault satellite telemetry dimension reduction data by dimension reduction on the fault satellite telemetry data; the potential space representation is obtained by a low-dimensional representation optimization model according to the fault satellite telemetry dimension reduction data.

4. The deep learning based aviation aircraft satellite communication fault diagnosis and repair method according to claim 3, characterized in that, the low-dimensional representation optimization model comprises: obtaining sample initial low-dimensional representation by minimizing a reconstruction loss function according to the fault satellite telemetry dimension reduction data; the reconstruction loss function is represented as: ， wherein, L rec represents the reconstruction loss function, N represents the number of fault satellite telemetry dimension reduction data, i represents the i-th fault satellite telemetry dimension reduction data, x i represents the fault satellite telemetry dimension reduction, Attention( ,a i ) represents the attention mechanism, represents the data subjected to the reconstruction loss function, a i represents the attention weight, λ is a regularization parameter, j represents the j-th weight, M represents the number of weights, w j represents the weight; The latent space representation is calculated according to the initial low-dimensional representation of the sample through a contrast loss function.

5. The deep learning based aviation aircraft satellite communication fault diagnosis and repair method according to claim 3, characterized in that, The probability model comprises: A component type probability is obtained through an artificial neural network according to the latent space representation; Complete mixed component parameters including weight parameters, mean vector parameters and covariance matrix parameters are calculated through maximization according to the component type probability; The in-distribution likelihood is calculated through probability according to the complete mixed component parameters and the latent space representation; The probability calculation expression is: ， wherein, P(x) is the likelihood within the distribution, c denotes the c-th complete mixture component parameter, C denotes the number of complete mixture component parameters, is a weight, denotes the determinant of the covariance matrix of the c-th complete mixture component, tr denotes the trace operation, x denotes the latent space representation, μ c is the mean vector, T denotes the transposition operation.

6. The deep learning based aviation aircraft satellite communication fault diagnosis and repair method according to claim 2, characterized in that, The scoring model comprises: An out-of-distribution likelihood value is calculated through scoring according to the in-distribution likelihood; The scoring calculation expression is: ， wherein, Score(x out ) is the out-of-distribution likelihood value, c denotes the c-th complete mixture component parameter, C denotes the number of complete mixture component parameters, is a weight, denotes the determinant of the covariance matrix of the c-th complete mixture component, x in denotes the latent space representation, μ c is the mean vector, T denotes a transposed calculation, the out-of-distribution likelihood value being calculated by a negative logarithm calculation on the in-distribution likelihood rate; A preset scoring threshold, the out-of-distribution likelihood value and the scoring threshold are used to determine the inter-satellite communication abnormal mode through scoring; The scoring determination comprises determining whether the out-of-distribution likelihood value is greater than the scoring threshold, if yes, the inter-satellite communication abnormal mode is an unknown type, and if no, the inter-satellite communication abnormal mode is a known type.

7. A deep learning based aerial vehicle satellite communication fault diagnosis and repair system, characterized in that, The system comprises a data collection module, a fault detection module and a type diagnosis module; The data collection module is configured to acquire a communication satellite topology network, collect observation parameters of the communication satellite topology network through a dual-mode heterogeneous receiving terminal, and generate constellation measurement data sets through data archiving; The fault detection module is configured to acquire abnormal node positioning information through a constellation behavior perception model according to the constellation measurement data sets and the communication satellite topology network; The type diagnosis module is configured to acquire an inter-satellite communication abnormal mode through an abnormal mode reconstruction model according to the abnormal node positioning information; Fault repair is performed according to the inter-satellite communication abnormal mode; The constellation behavior perception model comprises: An initial network topology, positioning parameters and candidate abnormal nodes are acquired through preliminary fault detection according to the constellation measurement data sets and the communication satellite topology network; The abnormal node positioning information is acquired through back substitution detection according to the initial network topology, the candidate abnormal nodes and the positioning parameters; The preliminary fault detection specifically comprises: Measurement data in the measurement data sets and predicted pseudo-range are acquired, and pseudo-range deviation is calculated through residual calculation according to the measurement data and the predicted pseudo-range; Normalized pseudo-range deviation is obtained through normalization processing of the pseudo-range deviation; The maximum normalized pseudo-range deviation is obtained through sorting according to the normalized pseudo-range deviation, and a satellite corresponding to the maximum normalized pseudo-range deviation is marked as a candidate abnormal node; The maximum correlation coefficient is obtained through correlation calculation and sorting according to the candidate abnormal nodes; a preset correlation threshold is provided, if the maximum correlation coefficient is greater than the correlation threshold, a satellite corresponding to the maximum correlation coefficient is a candidate abnormal node, and if the maximum correlation is not greater than the correlation threshold, a satellite corresponding to the maximum correlation is a normal node; The initial network topology is obtained by excluding the candidate abnormal node in the communication satellite topology network; and the positioning parameters are obtained according to the initial network topology; The back substitution detection comprises: calculating a global detection statistic by a global statistic according to the candidate abnormal node and the positioning parameter; presetting a global threshold, and judging recovery according to the global detection statistic and the global threshold to obtain a final no-candidate-abnormal-node seat; the recovery judgment comprises judging whether the global detection statistic is less than the global threshold, yes, the candidate abnormal node is recovered to a normal node; no, the candidate abnormal node is still a candidate abnormal node; obtaining the abnormal node positioning information by least square estimation according to the final no-candidate-abnormal-node seat.

Images