Airborne navigation integrity fault detection and identification method fusing multi-base augmentation information

Abstract

The application discloses a kind of fusion multi-base enhanced information airborne navigation integrity fault detection and identification method, belong to satellite navigation and aviation operation safety technical field.The method obtains satellite-based enhancement, ground-based enhancement, airborne enhancement, satellite navigation observation and auxiliary navigation information and carries out standardization preprocessing;For ionospheric anomaly and deception jamming, propagation anomaly feature and interference anomaly feature are constructed;In the sliding time window, the data layer fusion of the correlation variance contribution rate method is carried out to the same kind of information, and the fusion feature is constructed as the different kind of time sequence feature with different kind of information;Cross attention based Transformer network is used to fuse the features of different time sequence features, form interactive fusion representation, output integrity fault detection result and fault type identification result.The application improves the reliability of airborne navigation integrity fault detection and fault type identification ability in complex environment, and is suitable for Beidou aviation navigation and other multi-base enhanced cooperative navigation scene.

Description

Technical Field

[0001] This invention relates to a method for detecting airborne navigation integrity faults. In particular, it relates to a method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information. Background Technology

[0002] To provide high-security aviation navigation services, in addition to core constellations such as GPS, BeiDou, GLONASS, and GALILEO, modern global navigation satellite systems (GNSS) also extensively rely on various augmentation systems. Augmentation systems improve the accuracy, integrity, continuity, and availability of satellite navigation and positioning by correcting, monitoring, and constraining the pseudorange observations of the core constellations, thereby providing aircraft with high-precision, high-integrity positioning, navigation, and timing (PNT) services. Typical augmentation systems include airborne augmentation systems (ABAS), ground-based augmentation systems (GBAS), and satellite-based augmentation systems (SBAS).

[0003] Modern airborne navigation receivers typically integrate global satellite navigation and its augmentation functions into multi-mode receivers (MMRs) to support traditional precision approaches and satellite-based navigation throughout the entire flight phase. Depending on the equipment configuration and airworthiness approval status, MMRs can support various augmented navigation modes such as SBAS, GBAS, and ABAS. SBAS mode primarily utilizes wide-area differential correction information and integrity parameters broadcast by satellite-based augmentation systems to form satellite-based augmented navigation solutions and integrity assessments. GBAS mode primarily utilizes local differential correction information and integrity parameters broadcast by airport ground-based augmentation systems to form ground-based augmented navigation solutions and integrity assessments. ABAS mode primarily relies on airborne redundant observations and autonomous integrity monitoring algorithms to form airborne augmented navigation solutions and integrity assessments. Integrity assessment is essentially accomplished through fault detection and identification, integrity parameter calculation, and alarm logic. Existing MMRs typically perform integrity assessments based on correction information, integrity parameters, or airborne redundant observations provided by a single augmentation source, forming a single-base augmentation architecture characterized by independent operation of a single augmentation mode. The architecture can meet the airborne integrity assessment requirements under normal conditions, but under complex environmental conditions such as ionospheric anomalies and deception interference, different enhancement modes may have different perception, estimation and response results for the same navigation error source, reducing the reliability of fault detection and identification results.

[0004] On the one hand, ionospheric anomalies cause spatial non-uniformity and temporal non-stationarity in the propagation delay of satellite navigation signals, leading to deviations or even mismatches in the perception and response results of SBAS, GBAS, and ABAS to the same error source. This propagation anomaly results in inconsistencies in the correction information, integrity parameters, or fault indications provided by different augmentation systems. On the other hand, spoofing interference, by forging or forwarding satellite navigation signals, can cause SBAS, GBAS, and ABAS to exhibit different "information distortion modes," resulting in a single augmentation mode often only being able to perceive a portion of the effects of spoofing interference. Most existing airborne integrity fault detection and identification methods are designed independently around a single base augmentation source, lacking a mechanism for the coordinated utilization of complementary information from SBAS, GBAS, ABAS, and other auxiliary navigation equipment, and also lacking technical means to uniformly map the inconsistencies of multi-base augmentation information into a basis for fault detection and identification. Therefore, in complex environments such as ionospheric anomalies and spoofing interference, they are prone to missed detections, false detections, or misjudgments of fault types, making it difficult to meet the aircraft's requirements for high-reliability integrity monitoring and operational safety assurance. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and to address the problems of insufficient detection reliability, weak fault type identification ability, and difficulty in the coordinated use of multi-base augmentation information in airborne navigation integrity fault detection and identification in complex environments such as ionospheric anomalies and deception interference. This invention provides an airborne navigation integrity fault detection and identification method that integrates multi-base augmentation information, thereby improving the reliability of airborne integrity fault detection and identification in complex environments.

[0006] The technical solution adopted in this invention is: a method for detecting and identifying airborne navigation integrity faults by integrating multi-base augmentation information, characterized by comprising the following steps:

[0007] 1) Information acquisition and preprocessing, including: acquiring the raw multi-source input information of the airborne receiver at the current moment, and performing unified and standardized preprocessing on the raw multi-source input information to obtain standardized multi-source input information;

[0008] 2) Construction of environmental impact characteristics

[0009] Within the sliding time window, environmental impact characteristics are constructed from standardized multi-source input information, including: propagation anomaly characteristics and interference anomaly characteristics;

[0010] 3) Integration of similar information

[0011] Within the sliding time window, data fusion based on the correlation variance contribution rate method is performed on similar information in the outputs of steps 1) and 2) to form similar information fusion features.

[0012] 4) Construction of heterogeneous temporal features

[0013] Within the sliding time window, the similar information fusion features output in step 3) are grouped with other heterogeneous information output in steps 1) and 2) to construct three groups of heterogeneous time series features.

[0014] Other anomalous information in the outputs of steps 1) and 2) includes: satellite navigation observation information, auxiliary navigation information, and interference anomaly characteristics;

[0015] 5) Feature fusion based on cross-attention

[0016] A Transformer network based on cross-attention is used to construct the input sequence and fuse the heterogeneous temporal features output in step 4) to obtain an interactive fusion representation;

[0017] 6) Fault detection and fault type identification

[0018] The interactive fusion representation output from step 5) The fault detection result and the fault type identification result are generated through a dual-task output layer consisting of a fault detection model and a fault type identification model.

[0019] The airborne navigation integrity fault detection and identification method integrating multi-base augmentation information of the present invention has the following beneficial effects:

[0020] First, this invention comprehensively utilizes satellite-based augmentation, ground-based augmentation, airborne augmentation, satellite navigation observation, and auxiliary navigation information, overcoming the limitations of a single augmentation mode in terms of incomplete fault perception and response in complex environments, and improving the reliability of airborne navigation integrity fault detection.

[0021] Second, this invention integrates differential correction information, integrity parameters, navigation solutions, alarm information, and multi-observation consistency disruption features of propagation anomalies and interference anomalies provided by the multi-base augmentation system as similar information for data layer fusion. The fusion weight is determined based on the correlation variance contribution rate method, which can reduce the impact of single information source anomalies on detection results. At the same time, it can characterize propagation error mismatch caused by ionospheric anomalies and signal observation anomalies caused by deception interference, thereby improving the robustness of multi-base augmentation information utilization.

[0022] Third, this invention explicitly models the cross-layer interaction relationships between enhanced information fusion features, observation and auxiliary information features, and anomaly features through a Transformer network based on cross-attention. It can learn the correlation weights between different information sources and improve the ability to identify ionospheric anomalies, deception interference, compound faults, and other enhanced information anomalies.

[0023] Fourth, the present invention outputs integrity fault detection results and fault type identification results, which can provide a basis for airborne navigation alarms, navigation mode switching and operational risk assessment in complex environments, and is applicable to Beidou aviation navigation and other multi-base augmented cooperative navigation application scenarios. Attached Figure Description

[0024] Figure 1 This is a flowchart of the airborne navigation integrity fault detection and identification method integrating multi-base augmentation information according to the present invention;

[0025] Figure 2 This is a block diagram of the multi-base enhanced feature fusion structure based on the cross-attention Transformer network in this invention. Detailed Implementation

[0026] The following detailed description of the airborne navigation integrity fault detection and identification method of the present invention, which integrates multi-base augmentation information, is provided in conjunction with embodiments and accompanying drawings.

[0027] like Figure 1 As shown, the airborne navigation integrity fault detection and identification method of the present invention, which integrates multi-base augmentation information, includes the following steps performed in sequence:

[0028] 1) Information Acquisition and Preprocessing

[0029] The system acquires the raw multi-source input information from the airborne receiver at the current moment and performs unified and standardized preprocessing on the raw multi-source input information to obtain standardized multi-source input information. Specifically:

[0030] (11) Acquire the raw multi-source input information of the airborne receiver, wherein the raw multi-source input information consists of multi-base enhancement and multi-source navigation information, specifically including:

[0031] a) Satellite-based augmentation information, including: wide-area differential correction information, satellite-based augmentation integrity parameters, satellite-based augmentation navigation solutions, and satellite-based augmentation alarm information;

[0032] b) Foundation augmentation information, including: local differential correction information, foundation augmentation integrity parameters, foundation augmentation navigation solution, and foundation augmentation alarm information;

[0033] c) Airborne augmentation information, including: airborne augmentation navigation solution, airborne augmentation integrity parameters, and airborne augmentation alarm information;

[0034] d) Satellite navigation observation information, including: acquisition parameters consisting of the amplitude of the relevant main peak, the amplitude of the relevant maximum side peak, and the half-power width of the relevant main peak; tracking loop parameters consisting of the outputs of the code tracking loop (DLL), phase-locked loop (PLL), and frequency-locked loop (FLL) discriminators; and pseudorange, carrier phase, Doppler, carrier-to-noise ratio, satellite ephemeris, and satellite geometric distribution parameters.

[0035] e) Auxiliary navigation information, including: inertial navigation system position, speed, attitude information, barometric altitude, radio altitude, airspeed, track angle, and flight phase markings.

[0036] (12) Perform unified preprocessing on the original multi-source input information to form standardized multi-source input information. The preprocessing includes: time synchronization, resampling, coordinate unification, missing value completion, outlier removal or pruning, and dimension normalization.

[0037] 2) Construction of environmental impact characteristics

[0038] Within the sliding time window, environmental impact features are constructed from standardized multi-source input information, including propagation anomaly features and interference anomaly features. In this embodiment, the sliding time window length is set to 120 sampling times. Details are as follows:

[0039] (21) The propagation anomaly features include: enhancement correction difference features, pseudorange residual time gradient features, and inter-satellite residual spatial gradient features.

[0040] The enhanced correction difference feature is composed of the mean, root mean square, and maximum and minimum difference of the wide-area differential correction information and local differential correction information difference of all satellites within the sliding time window;

[0041] The pseudorange residual time gradient feature is composed of the mean, root mean square, and maximum absolute value of the pseudorange residual time gradients of all satellites within a sliding time window. Within the sliding time window, the theoretical pseudorange is calculated from the satellite-based augmentation navigation solution, the ground-based augmentation navigation solution, the airborne augmentation navigation solution, and the satellite ephemeris. The difference between these calculations and the pseudoranges in the satellite navigation observation information is used to obtain the pseudorange residual. The pseudorange residual time gradient is then calculated. Based on this, the mean, root mean square, and maximum absolute value of the pseudorange residual time gradients of all satellites under different augmentation modes are calculated to form the pseudorange residual time gradient feature.

[0042] The inter-satellite residual spatial gradient feature is composed of the mean, root mean square, and maximum absolute value of the pseudorange residual differences between all satellites at the same time within a sliding time window. Within the sliding time window, the pseudorange residual differences of different satellites under different enhancement modes are calculated at the same time. Based on this, the mean, root mean square, and maximum absolute value of the pseudorange residual differences are calculated to form the inter-satellite residual spatial gradient feature.

[0043] (22) The interference anomaly features include: correlation peak distortion features, tracking loop anomaly features, and multi-observation consistency disruption features. Among them,

[0044] The aforementioned correlation peak distortion characteristics are calculated from the acquisition parameters in the satellite navigation observation information, including peak ratio and peak width; wherein, the peak ratio is the average of the ratio of the correlation main peak amplitude to the correlation maximum side peak amplitude of all satellites, and the peak width is the average of the correlation main peak half-power width of all satellites.

[0045] The aforementioned tracking loop anomaly features are calculated from the root mean square of the differences between adjacent time points output by the code tracking loop (DLL), phase-locked loop (PLL), and frequency-locked loop (FLL) discriminators in the satellite navigation observation information.

[0046] The aforementioned multi-observation consistency violation characteristics are calculated from pseudorange, carrier phase, and Doppler data in satellite navigation observation information. These include inconsistencies between pseudorange change rate and Doppler data, and between carrier phase change rate and Doppler data. Within a sliding time window, the pseudorange change rate is calculated from the difference in pseudorange at adjacent times, and the range change rate is obtained by multiplying Doppler by the carrier wavelength. The absolute mean and root mean square of the differences between the pseudorange change rates and range change rates of all satellites are calculated to form the inconsistency between pseudorange change rate and Doppler data. Similarly, the carrier phase change rate is calculated from the difference in carrier phase at adjacent times, and the absolute mean of the sum of the carrier phase change rates and Doppler data of all satellites is calculated to form the inconsistency between carrier phase change rate and Doppler data.

[0047] 3) Integration of similar information

[0048] Within the sliding time window, data fusion based on the correlation variance contribution rate method is performed on similar information in the outputs of steps 1) and 2) to form similar information fusion features.

[0049] The aforementioned similar information refers to repetitive descriptions of the same physical object, the same navigation state, or the same fault state, including: wide-area differential correction information and local differential correction information, integrity parameters, navigation solutions, and alarm information in different enhancement information, as well as propagation anomaly characteristics; the aforementioned correlation variance contribution rate method includes correlation calculation, variance contribution rate calculation, fusion weight calculation, and data layer fusion; wherein,

[0050] (31) Correlation calculation

[0051] Calculate the correlation matrix for each type of information separately. :

[0052] (1);

[0053] in, Indicates the first in a class of information The information source and the first The correlation coefficient between information sources The dimension of the same type of information.

[0054] (32) Calculation of variance contribution rate

[0055] Calculate the first of each type of information separately Variance contribution rate of each information source :

[0056] (2);

[0057] in, For the first The variance of each information source within the current time window For the first The variance of each information source within the current time window.

[0058] (33) Calculation of fusion weights

[0059] Based on correlation and variance contribution rate, construct the first class of information for each class. The fusion weight of each information source :

[0060] (3);

[0061] in, For the first The variance contribution rate of each information source For each type of information, the first The average relevance between an information source and other information sources For each type of information, the first The average relevance of an information source to other information sources:

[0062] (4);

[0063] 34) Data Layer Fusion

[0064] By weighted summation of various types of information, we obtain: differential correction information fusion, integrity parameter fusion, navigation solution fusion, alarm information fusion, and propagation anomaly feature fusion, ultimately forming a fusion feature of similar information.

[0065] 4) Construction of heterogeneous temporal features

[0066] Within the sliding time window, the similar information fusion features output in step 3) are grouped with other heterogeneous information output in steps 1) and 2) to construct three groups of heterogeneous time-series features.

[0067] Other anomalous information in the output of steps 1) and 2) includes: satellite navigation observation information, auxiliary navigation information, and interference anomaly characteristics.

[0068] The grouping involves constructing three groups of heterogeneous time-series features based on information source and functional attributes: enhanced information fusion features, observation and auxiliary information features, and anomaly features; among which, Real-time augmented information fusion features This includes differential correction information fusion, integrity parameter fusion, navigation solution fusion, and alarm information fusion; Features of observation and auxiliary information at different times This includes satellite navigation observation information and auxiliary navigation information; Anomalous characteristics of time This includes the fusion of propagation anomaly features and interference anomaly features.

[0069] 5) Feature fusion based on cross-attention

[0070] like Figure 2 As shown, a Transformer network based on cross-attention is used to construct the input sequence and fuse the heterogeneous temporal features output in step 4) to obtain an interactive fusion representation.

[0071] 51) Transformer Network Input Sequence Construction

[0072] The heterogeneous temporal features of each group are normalized and concatenated, and then mapped using a unified linear projection. Preset dimensions of time Embedded vector :

[0073] (5);

[0074] Where: projection matrix , for Time of the first Normalized vectors of heterogeneous temporal features and concatenated vectors for The dimension of the bias vector , and The parameters are learned end-to-end during the Transformer network training. In this embodiment, the preset dimension... The value is set to 64.

[0075] Subsequently, the embedding vector Overlay positional encoding vectors ,get:

[0076] (6);

[0077] in, for The input vector of the Transformer network at each time step; Time position encoding vector The odd and even elements are:

[0078] (7);

[0079] Ultimately, it forms within the sliding time window. Time-based Transformer network input sequence :

[0080] (8);

[0081] in, The length of the sliding time window. for Constantly enhance the information fusion sequence, for Time-based observation and auxiliary information sequence, for Time-based anomaly feature sequence.

[0082] 52) Cross-attention fusion

[0083] The three sets of Transformer network input sequences are passed through a cross-attention layer and a feature fusion layer to form an interactive fusion representation.

[0084] The cross-attention layer employs a two-way, eight-head cross-attention mechanism to construct multi-path interaction relationships between the enhanced information fusion sequence, the observation and auxiliary information sequences, and the anomaly feature sequence, including the interaction feature representation (MHA) of the enhanced information fusion sequence and the observation and auxiliary information sequence. 1←2 Enhanced Information Fusion Sequence—Anomaly Feature Sequence Interaction Feature Representation (MHA) 1←3 In each interaction, an 8-head cross-attention mechanism is used to weight the query vector, key vector, and value vector to obtain the corresponding interaction feature representation: MHA. 1←2 MHA 1←3 In MHA 1←2 In the computation, the query vector is an enhanced information fusion sequence, while both the key vector and value vector are sequences of observation and auxiliary information; in MHA 1←3 In the calculation, the query vector is an enhanced information fusion sequence, while the key vector and value vector are both abnormal feature sequences.

[0085] The feature fusion layer includes residual and normalization processing and gated weighted fusion; firstly, the interactive feature representation MHA output by the cross-attention layer is processed. 1←2 and MHA 1←3 ,as well as Real-time Enhanced Information Fusion Sequence After residual and normalization processing, a standardized interactive feature representation of the enhanced information fusion sequence—observation and auxiliary information sequence—is obtained. Standardized interactive feature representation of enhanced information fusion sequence—abnormal feature sequence Then on , and enhanced information fusion sequence Perform gated weighted fusion to obtain Moment-based interactive fusion representation :

[0086] (9);

[0087] in, , , The fusion weights are adaptively determined through a gating mechanism.

[0088] 6) Fault detection and fault type identification

[0089] The interactive fusion representation output from step 5) The fault detection result and the fault type identification result are generated through a dual-task output layer consisting of a fault detection model and a fault type identification model.

[0090] The fault detection model consists of a fully connected layer, a GELU nonlinear activation layer, a normalization layer, and a binary classification output layer. It outputs the detection result of whether an integrity fault exists in the current interactive fusion representation. This includes: the interaction fusion representation is processed by the fully connected layer, the GELU nonlinear activation layer, and the normalization layer. Linear mapping and nonlinear transformations are performed to enhance the fault detection model's ability to represent complex boundaries and weak anomaly features, outputting a fault detection representation for subsequent processing. A binary classification output layer with a Softmax activation function and a decision criterion is used to determine the presence of a fault in the fault detection representation. The Softmax activation function outputs the fault probability corresponding to the fault detection representation, and the decision criterion determines whether an integrity fault exists based on a threshold decision criterion. During fault detection model training, a binary classification cross-entropy loss function is constructed using the interactive fusion representation and the corresponding integrity fault label. Through backpropagation and parameter updates, the mapping relationship between the interactive fusion representation and integrity faults is learned, obtaining the parameters of each layer of the fault detection model and the detection threshold. When applying the fault detection model, the trained parameters are used to determine the presence of an integrity fault in the current input interactive fusion representation: if the fault probability is greater than or equal to the detection threshold, an integrity fault is determined to exist; otherwise, it is determined to be normal.

[0091] The fault type identification model consists of a classification network structure comprising a fully connected layer, a GELU nonlinear activation layer, and a multi-classification output layer. It is used to generate fault type identification results, including: an interactive fusion representation formed by the fully connected layer and the GELU nonlinear activation layer. Linear mapping and nonlinear transformations are performed to enhance the fault type identification model's ability to represent complex boundaries and weak anomaly features, outputting a fault identification representation for subsequent processing. A multi-class output layer with a Softmax activation function and a decision criterion is used to determine the fault type of the representation. The Softmax activation function outputs the identification probability for each fault type, and the decision criterion determines the fault type based on the maximum probability criterion. Fault types include ionospheric anomaly faults, spoofing interference faults, combined ionospheric anomaly and spoofing interference faults, and other enhanced information anomaly faults. During training, a multi-class cross-entropy loss function is constructed using the interactive fusion representation and corresponding integrity fault type labels. Through backpropagation and parameter updates, the differential distributions of different fault types in enhanced information fusion, observation and auxiliary information, and anomaly features are learned, obtaining the parameters of each layer of the fault type identification model. When applying the fault type identification model, the trained parameters are used to identify the fault type of the current input interactive fusion representation. The identification probabilities of each fault type are calculated and compared, and the fault type with the highest probability is taken as the final fault type identification result.

Claims

1. A method for detecting and identifying airborne navigation integrity faults by integrating multi-base augmentation information, characterized in that, Includes the following steps: 1) Information acquisition and preprocessing, including: acquiring the raw multi-source input information of the airborne receiver at the current moment, and performing unified and standardized preprocessing on the raw multi-source input information to obtain standardized multi-source input information; 2) Construction of environmental impact characteristics Within the sliding time window, environmental impact characteristics are constructed from standardized multi-source input information, including: propagation anomaly characteristics and interference anomaly characteristics; 3) Integration of similar information Within the sliding time window, data fusion based on the correlation variance contribution rate method is performed on similar information in the outputs of steps 1) and 2) to form similar information fusion features. 4) Construction of heterogeneous temporal features Within the sliding time window, the similar information fusion features output in step 3) are grouped with other heterogeneous information output in steps 1) and 2) to construct three groups of heterogeneous time series features. Other anomalous information in the outputs of steps 1) and 2) includes: satellite navigation observation information, auxiliary navigation information, and interference anomaly characteristics; 5) Feature fusion based on cross-attention A Transformer network based on cross-attention is used to construct the input sequence and fuse the heterogeneous temporal features output in step 4) to obtain an interactive fusion representation; 6) Fault detection and fault type identification The interactive fusion representation output from step 5) The fault detection result and the fault type identification result are generated through a dual-task output layer consisting of a fault detection model and a fault type identification model.

2. The airborne navigation integrity fault detection and identification method according to claim 1, characterized in that, In step 1): (11) The original multi-source input information is composed of multi-base enhancement and multi-source navigation information, specifically including: a. Satellite-based augmentation information, including wide-area differential correction information, satellite-based augmentation integrity parameters, satellite-based augmentation navigation solutions, and satellite-based augmentation alarm information; b. Foundation augmentation information, including: local differential correction information, foundation augmentation integrity parameters, foundation augmentation navigation solution, and foundation augmentation alarm information; c. Airborne augmentation information, including: airborne augmentation navigation solution, airborne augmentation integrity parameters, and airborne augmentation alarm information; d. Satellite navigation observation information, including: acquisition parameters consisting of the amplitude of the relevant main peak, the amplitude of the relevant maximum side peak, and the half-power width of the relevant main peak; tracking loop parameters consisting of the outputs of the code tracking loop, phase-locked loop, and frequency-locked loop discriminator; and pseudorange, carrier phase, Doppler, carrier-to-noise ratio, satellite ephemeris, and satellite geometric distribution parameters. e. Auxiliary navigation information, including: inertial navigation system position, speed, attitude information, barometric altitude, radio altitude, airspeed, track angle, and flight phase markings; (12) The preprocessing includes: time synchronization, resampling, coordinate unification, missing value completion, outlier removal or pruning, and dimension normalization.

3. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 1, characterized in that, In step 2): (21) The propagation anomaly features include: enhancement correction difference features, pseudorange residual temporal gradient features, and inter-satellite residual spatial gradient features; wherein, The enhanced correction difference feature is composed of the mean, root mean square, and maximum and minimum difference of the wide-area differential correction information and local differential correction information difference of all satellites within the sliding time window; The pseudorange residual time gradient feature is composed of the mean, root mean square, and maximum absolute value of the pseudorange residual time gradients of all satellites within a sliding time window. Within the sliding time window, the theoretical pseudorange is calculated from the satellite-based augmentation navigation solution, the ground-based augmentation navigation solution, the airborne augmentation navigation solution, and the satellite ephemeris. The difference between these calculations and the pseudoranges in the satellite navigation observation information is used to obtain the pseudorange residual. The pseudorange residual time gradient is then calculated. Based on this, the mean, root mean square, and maximum absolute value of the pseudorange residual time gradients of all satellites under different augmentation modes are calculated to form the pseudorange residual time gradient feature. The inter-satellite residual spatial gradient feature is composed of the mean, root mean square, and maximum absolute value of the pseudorange residual difference between all satellites at the same time within a sliding time window. Within the sliding time window, the pseudorange residual difference between different satellites under different enhancement modes at the same time is calculated, and on this basis, the mean, root mean square, and maximum absolute value of the pseudorange residual difference are calculated to form the inter-satellite residual spatial gradient feature. (22) The interference anomaly features include: correlation peak distortion features, tracking loop anomaly features, and multi-observation consistency disruption features; among which, The aforementioned correlation peak distortion characteristics are calculated from the acquisition parameters in the satellite navigation observation information, including peak ratio and peak width; wherein, the peak ratio is the average of the ratio of the correlation main peak amplitude to the correlation maximum side peak amplitude of all satellites, and the peak width is the average of the correlation main peak half-power width of all satellites; The aforementioned tracking loop anomaly features are calculated from the root mean square of the differences between adjacent time points output by the code tracking loop, phase-locked loop, and frequency-locked loop discriminators in the satellite navigation observation information. The aforementioned multi-observation consistency violation characteristics are calculated from pseudorange, carrier phase, and Doppler data in satellite navigation observation information. These characteristics include the inconsistency between pseudorange change rate and Doppler data, and the inconsistency between carrier phase change rate and Doppler data. Within a sliding time window, the pseudorange change rate is calculated from the difference in pseudorange at adjacent times, and the range change rate is obtained by multiplying Doppler by the carrier wavelength. The absolute mean and root mean square of the differences between the pseudorange change rates and range change rates of all satellites are calculated to form the inconsistency between pseudorange change rate and Doppler data. Similarly, the carrier phase change rate is calculated from the difference in carrier phase at adjacent times, and the absolute mean of the sum of the carrier phase change rates and Doppler data of all satellites is calculated to form the inconsistency between carrier phase change rate and Doppler data.

4. The airborne navigation integrity fault detection and identification method according to claim 1, characterized in that, The similar information mentioned in step 3) refers to repeated descriptions of the same physical object, the same navigation state, or the same fault state, including: wide-area differential correction information and local differential correction information, integrity parameters, navigation solutions and alarm information in different enhancement information, and propagation anomaly characteristics; the correlation variance contribution rate method includes correlation calculation, variance contribution rate calculation, fusion weight calculation and data layer fusion.

5. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 4, characterized in that, (31) The correlation calculation mentioned above is to calculate the correlation matrix of each type of information separately. : （1）； in, Indicates the first in a class of information The information source and the first The correlation coefficient between information sources The dimension of the same type of information; (32) The variance contribution rate calculation mentioned above is to calculate the variance contribution rate of each type of information separately. Variance contribution rate of each information source : （2）； in, For the first The variance of each information source within the current time window For the first The variance of each information source within the current time window; (33) The fusion weight calculation mentioned above is based on the correlation and variance contribution rate to construct the first fusion weight of each type of information. The fusion weight of each information source : （3）； in, For the first The variance contribution rate of each information source For each type of information, the first The average relevance between an information source and other information sources For each type of information, the first The average relevance of an information source to other information sources: （4）； (34) The data layer fusion is to perform weighted summation of various similar information to obtain: differential correction information fusion, integrity parameter fusion, navigation solution fusion, alarm information fusion and propagation anomaly feature fusion, and finally form similar information fusion features.

6. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 1, characterized in that, The grouping mentioned in step 4) involves constructing three groups of heterogeneous time-series features based on information source and functional attributes, including: enhanced information fusion features, observation and auxiliary information features, and anomaly features; among which, Real-time augmented information fusion features This includes differential correction information fusion, integrity parameter fusion, navigation solution fusion, and alarm information fusion; Features of observation and auxiliary information at different times This includes satellite navigation observation information and auxiliary navigation information; Anomalous characteristics of time This includes the fusion of propagation anomaly features and interference anomaly features.

7. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 1, characterized in that, Step 5) describes the construction of the Transformer network input sequence, which includes: The heterogeneous temporal features of each group are normalized and concatenated, and then mapped using a unified linear projection. Preset dimensions of time Embedded vector : （5）； Where: projection matrix ,for Time of the first Normalized vectors of heterogeneous temporal features and concatenated vectors for The dimension of the bias vector , and The parameters are learned end-to-end during the training of the Transformer network; Subsequently, the embedding vector Overlay positional encoding vectors ,get: （6）； in, for The input vector of the Transformer network at each time step; Time position encoding vector The odd and even terms are: （7）； Ultimately, it forms within the sliding time window. Time-based Transformer network input sequence : （8）； in, The length of the sliding time window. for Constantly enhance the information fusion sequence, for Time-based observation and auxiliary information sequence, for Time-based anomaly feature sequence.

8. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 1, characterized in that, The cross-attention fusion described in step 5) involves passing the three sets of Transformer network input sequences through a cross-attention layer and a feature fusion layer to form an interactive fused representation; The cross-attention layer employs a two-way, eight-head cross-attention mechanism to construct multi-path interaction relationships between the enhanced information fusion sequence, the observation and auxiliary information sequences, and the anomaly feature sequence, including the interaction feature representation (MHA) of the enhanced information fusion sequence and the observation and auxiliary information sequence. 1←2 Enhanced Information Fusion Sequence—Anomaly Feature Sequence Interaction Feature Representation (MHA) 1←3 In each interaction, an 8-head cross-attention mechanism is used to weight the query vector, key vector, and value vector to obtain the corresponding interaction feature representation: MHA. 1←2 MHA 1←3 In MHA 1←2 In the computation, the query vector is an enhanced information fusion sequence, while both the key vector and value vector are sequences of observation and auxiliary information; in MHA 1←3 In the calculation, the query vector is an enhanced information fusion sequence, while both the key vector and the value vector are sequences of anomalous features; The feature fusion layer includes residual and normalization processing and gated weighted fusion; firstly, the interactive feature representation MHA output by the cross-attention layer is processed. 1←2 and MHA 1←3 ,as well as Real-time Enhanced Information Fusion Sequence After residual and normalization processing, a standardized interactive feature representation of the enhanced information fusion sequence—observation and auxiliary information sequence—is obtained. Standardized interactive feature representation of enhanced information fusion sequence—abnormal feature sequence Then on , and enhanced information fusion sequence Perform gated weighted fusion to obtain Moment-based interactive fusion representation : （9）； in, , , The fusion weights are adaptively determined through a gating mechanism.

9. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 1, characterized in that, The fault detection model described in step 6) consists of a fully connected layer, a GELU nonlinear activation layer, a normalization layer, and a binary classification output layer. It is used to output the detection result of whether an integrity fault exists in the current interactive fusion representation. This includes: the interaction fusion representation is processed by the fully connected layer, the GELU nonlinear activation layer, and the normalization layer. Linear mapping and nonlinear transformations are performed to enhance the fault detection model's ability to express complex boundaries and weak anomaly features, outputting a fault detection representation for subsequent processing. A binary classification output layer with a Softmax activation function and a decision criterion is used to determine the presence of a fault in the fault detection representation. The Softmax activation function outputs the fault probability corresponding to the fault detection representation, while the decision criterion determines whether an integrity fault exists based on a threshold decision criterion. During fault detection model training, a binary cross-entropy loss function is constructed using the interactive fusion representation and the corresponding integrity fault label. Through backpropagation and parameter updates, the mapping relationship between the interactive fusion representation and integrity faults is learned, obtaining the parameters of each layer of the fault detection model and the detection threshold. When applying the fault detection model, the trained parameters are used to determine the presence of an integrity fault in the current input interactive fusion representation: if the fault probability is greater than or equal to the detection threshold, an integrity fault is determined to exist; otherwise, it is determined to be normal.

10. The method for detecting and identifying airborne navigation integrity faults by fusing multi-base augmentation information according to claim 1, characterized in that, The fault type identification model described in step 6) consists of a classification network structure comprising a fully connected layer, a GELU nonlinear activation layer, and a multi-classification output layer. This network is used to generate fault type identification results, including: interactive fusion representations by the fully connected layer and the GELU nonlinear activation layer. Linear mapping and nonlinear transformation are performed to enhance the fault type identification model's ability to express complex boundaries and weak anomaly features, outputting a fault identification representation for subsequent processing. A multi-classification output layer with a Softmax activation function and a decision criterion is used to determine the fault type of the representation. The Softmax activation function outputs the identification probability of each fault type, and the decision criterion determines the fault type based on the maximum probability criterion. Fault types include ionospheric anomaly faults, deception interference faults, combined ionospheric anomaly and deception interference faults, and other enhanced information anomaly faults. During training, a multi-class cross-entropy loss function is constructed using the interactive fusion representation and corresponding integrity fault type labels. Through backpropagation and parameter updates, the differential distributions of different fault types in enhanced information fusion, observation and auxiliary information, and anomaly features are learned, obtaining the parameters of each layer of the fault type identification model. When applying the fault type identification model, the trained parameters are used to identify the fault type of the current input interactive fusion representation. The identification probabilities of each fault type are calculated and compared, and the fault type with the highest probability is taken as the final fault type identification result.

Images