Ionospheric scintillation-multipath coupling adaptive integrity enhancement method and system based on deep learning

Abstract

This invention belongs to the field of satellite navigation and avionics technology, and proposes a deep learning-based adaptive integrity enhancement method and system for ionospheric scintillation-multipath coupling. The method includes: S1, signal data acquisition and preprocessing; S2, disturbance feature extraction and state identification; S3, generation of a comprehensive disturbance intensity index; S4, calculation of the adaptive Sigma expansion coefficient; S5, integrity calculation; and S6, integrity enhancement and adaptive optimization. This invention improves the integrity accuracy and reliability of navigation systems in complex signal environments, ensuring that the airborne equipment of the Global Navigation Satellite System (GLS) meets airworthiness standards during the CAT II / III precision approach phase.

Description

Technical Field

[0001] This invention belongs to the field of satellite navigation and avionics technology, and in particular relates to an ionospheric scintillation-multipath coupling adaptive integrity enhancement method and system based on deep learning. Background Technology

[0002] During the precision approach and landing phases of civil aviation, aircraft have extremely high requirements for the reliability, accuracy, and safety of their navigation systems. Navigation integrity, as a key indicator for ensuring flight safety, requires the system to be able to detect and warn of potential navigation signal anomalies within a limited time (generally 6 to 300 seconds).

[0003] Traditional integrity monitoring methods often employ receiver autonomous integrity monitoring (RAIM), differential monitoring (GBAS integrity monitoring), fixed Sigma inflation based on statistical assumptions, and some enhanced integrity estimation methods based on statistical detection. These methods generally rely on the assumption of linear Gaussian noise and depend on preset thresholds to distinguish abnormal states.

[0004] However, with the large-scale application of Global Navigation Satellite Systems (GNSS) in civil aviation, factors such as ionospheric scintillation, multipath interference, signal fading, and antenna pattern effects exhibit strong nonlinear and non-stationary coupling characteristics in complex environments, resulting in significant limitations of traditional static algorithms in the following aspects: Static threshold failure: Traditional integrity algorithms cannot adjust the error model in real time according to the disturbance characteristics during different flight phases, such as low-altitude approach and urban reflection environment, leading to an increase in false alarm rate.

[0005] Missing environmental features: Existing methods only consider the geometric factor GDOP and measurement noise, without modeling the dynamic characteristics of the ionosphere and the effects of multipath coupling.

[0006] Insufficient model generalization: When ionospheric scintillation and multipath interference occur simultaneously, the errors are superimposed nonlinearly, and traditional statistical models cannot accurately capture their changing patterns.

[0007] Airworthiness performance is limited: Under complex weather and channel conditions, the probability of missed integrity checks is much higher than the 1×10⁻⁶ requirement of the FAA AC-20-138G standard. -9 It cannot support CAT II / III level operation.

[0008] The technical reasons for the above limitations include: 1. Integrity modeling assumptions are overly idealistic: Existing methods generally assume that navigation signal errors follow a Gaussian independent distribution and that the error characteristics remain stable over time. In complex environments such as ionospheric scintillation, multipath interference, and antenna pattern variations, this assumption no longer holds, leading to significant deviations between error estimates and actual conditions, thus affecting the accuracy of protection limits (VPL / HPL).

[0009] 2. Fixed expansion factor, lacking dynamic adaptability: Existing Sigma-Inflation models typically use empirical coefficients or fixed proportions to statically inflate the standard deviation. Under different flight phases and different signal environments (such as reflections from buildings around airports, enhanced ionospheric activity), the error distribution characteristics differ significantly. A fixed expansion factor cannot adapt to changes, easily leading to overly conservative or insufficient protection limits.

[0010] 3. Lack of joint modeling for ionospheric and multipath coupling effects: Existing techniques often treat ionospheric delay and multipath interference separately, without considering their interaction in the time and spatial domains. In real-world environments, these two types of errors often exhibit nonlinear superposition characteristics. Failure to establish a joint model can lead to inaccurate estimation of error propagation paths, thereby affecting the reliability of integrity assessment.

[0011] 4. Lack of real-time dynamic adjustment mechanism for integrity monitoring: Existing algorithms mostly use preset thresholds for anomaly detection during operation, and cannot adjust the monitoring sensitivity in real time according to signal status, flight attitude, or environmental disturbances. When signal quality changes rapidly, the false alarm rate is prone to increase or missed detections occur, reducing the system's safety margin and availability.

[0012] 5. Complex implementation and low verification efficiency in airworthiness engineering: Most current integrity algorithms rely on offline simulation verification, and algorithm parameters need to be set manually, making it impossible to automatically adapt to data from multiple scenarios. In an airborne environment, the algorithm debugging and verification cycle is long, which is not conducive to airworthiness compliance assessment and system engineering integration.

[0013] In summary, existing integrity monitoring technologies cannot simultaneously meet the requirements of real-time performance, adaptability, and integrity accuracy in ionospheric scintillation-multipath coupling scenarios. They still have a high false alarm rate and missed detection risk in complex signal environments, which limits their application in high-level precision approach (CAT II / III) conditions. Summary of the Invention

[0014] The purpose of this invention is to propose an ionospheric scintillation-multipath coupling adaptive integrity enhancement method and system based on deep learning, which improves the integrity accuracy and reliability of navigation systems in complex signal environments and ensures that airborne equipment of the Global Navigation Satellite Landing System (GLS) meets airworthiness standards during the CAT II / III precision approach phase.

[0015] To achieve the above objectives, the technical solution of the present invention is implemented as follows: A deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling includes: S1. Signal Data Acquisition and Preprocessing: Acquired signal data includes basic observation data of the Global Navigation Satellite System (GNSS), ionospheric disturbance characteristic data, multipath and spacecraft characteristic data; preprocessing of signal data includes synchronization and normalization. S2. Disturbance feature extraction and state recognition: Short-time Fourier transform is performed on the signal data to obtain ionospheric branches and multipath branches; The ionospheric perturbation fusion features are extracted using a Physically Aware Long Short-Term Memory (PA-LSTM) structure for ionospheric branches. ;in This represents the fused feature vector of ionospheric perturbations. The output of the time series characteristics of ionospheric perturbation at the current moment is adjusted respectively. Current state of the memory unit With amplitude flicker index The weighting coefficients of the three; The multipath branch is given a multi-source feature set from different satellites or observation paths as input, and a temporal convolutional filter based on perceptual physics is used to extract the multipath dynamic features. ;in Represents the dynamic feature vector of multipath. This represents a multi-source feature set from different satellites or observation paths. This represents the fused feature map of the convolutional layer output; The feature fusion layer under physical constraints fuses two high-dimensional features, namely ionospheric perturbation fusion feature and multipath dynamic feature, to obtain perturbation feature; ; in, Representing perturbation features; perturbation fusion dynamic attention weights and multipath dynamic attention weights It is obtained through adaptive learning via backpropagation during training. This represents the rate of change of the signal-to-noise ratio. This is a comprehensive characteristic quantity of the satellite's elevation angle. This is the physical adjustment coefficient; Then, the perturbation feature state vector is generated. : ; Among them, disturbance characteristics After nonlinear mapping transformation, three perturbation correlation quantities are obtained: Ionospheric disturbance intensity index This reflects the amplitude of phase fluctuations caused by ionospheric scintillation; Multipath interference intensity index This characterizes the energy variation and attitude dependence of the reflected signal; Signal health score This is a normalized signal stability index, reflecting the overall integrity of the navigation signal; Based on the joint judgment of the three disturbance-related quantities, the disturbance level (Level) is obtained. S3. Generate a comprehensive disturbance intensity index: The disturbance characteristics are calculated by adaptive weighting of disturbance level and time smoothing to generate a comprehensive disturbance intensity index for integrity modeling. ; ; in This represents the comprehensive disturbance index. These are weighting coefficients that are dynamically adjusted according to the disturbance level. This represents the comprehensive disturbance intensity index, where n is the index of the current time and N is the size of the sliding window; S3. Generate a comprehensive disturbance intensity index: The disturbance characteristics are calculated by adaptive weighting of disturbance level and time smoothing to generate a comprehensive disturbance intensity index for integrity modeling. S4. Perform adaptive Sigma expansion coefficient calculation: dynamically estimate the Sigma expansion coefficient under the current environment by comprehensively considering the disturbance intensity index; S5. Perform integrity calculation: Based on the calculated adaptive Sigma expansion coefficient, dynamically calculate and evaluate the overall integrity level at the current moment to obtain an integrity health score; S6. Integrity Enhancement and Adaptive Optimization: The observation reliability is dynamically weighted based on the integrity and health score; the navigation state covariance matrix is ​​dynamically adjusted based on the adaptive Sigma expansion coefficient; thereby correcting the navigation state estimation model.

[0016] Furthermore, step S6 also includes: introducing a disturbance compensation term correction. For integrity enhancement and correction in the navigation solution process: ;in, These are the original pseudorange or carrier observation values. For observations corrected for integrity, It is obtained by jointly estimating the disturbance fusion index and the current disturbance level, and is used to correct the ranging bias caused by signal distortion.

[0017] Furthermore, the preprocessing in step S1 includes: all signal data are synchronized by interpolation and resampling using GNSS time as a unified reference; then, the signal data are normalized to construct a feature matrix.

[0018] Furthermore, in step S4, a disturbance-driven dynamic response mapping mechanism is introduced during the Sigma calculation process. An adaptive expansion ratio adjustment model is established through the joint input of the comprehensive disturbance intensity index and the disturbance level label Level, which is used to dynamically estimate the Sigma expansion coefficient under the current environment.

[0019] Furthermore, in step S5, a multi-source integrity quantification model is constructed using the comprehensive disturbance intensity index, the disturbance level label Level, and the adaptive Sigma expansion coefficient as inputs, and an integrity health scoring function is defined.

[0020] Furthermore, in step S6, a dynamically weighted observation confidence update strategy was designed: ; in, This represents the observation weight matrix at the current time. For standard weights, The disturbance sensitivity coefficient, To comprehensively assess the intensity of the disturbance, The score is given for the integrity of sexual health.

[0021] In another aspect, this invention proposes a deep learning-based adaptive integrity enhancement system for ionospheric scintillation-multipath coupling, comprising: Signal data acquisition and preprocessing module: Signal data includes basic observation data of Global Navigation Satellite System (GNSS), ionospheric disturbance characteristic data, multipath and spacecraft characteristic data; preprocessing of signal data includes synchronization and normalization; The disturbance feature extraction and state recognition module performs short-time Fourier transform on the signal data to obtain ionospheric branches and multipath branches; The ionospheric perturbation fusion features are extracted using a Physically Aware Long Short-Term Memory (PA-LSTM) structure for ionospheric branches. ;in This represents the fused feature vector of ionospheric perturbations. The output of the time series characteristics of ionospheric perturbation at the current moment is adjusted respectively. Current state of the memory unit With amplitude flicker index The weighting coefficients of the three; The multipath branch is given a multi-source feature set from different satellites or observation paths as input, and a temporal convolutional filter based on perceptual physics is used to extract the multipath dynamic features. ;in Represents the dynamic feature vector of multipath. This represents a multi-source feature set from different satellites or observation paths. This represents the fused feature map of the convolutional layer output; The feature fusion layer under physical constraints fuses two high-dimensional features, namely ionospheric perturbation fusion feature and multipath dynamic feature, to obtain perturbation feature; ; in, Representing perturbation features; perturbation fusion dynamic attention weights and multipath dynamic attention weights It is obtained through adaptive learning via backpropagation during training. This represents the rate of change of the signal-to-noise ratio. This is a comprehensive characteristic quantity of the satellite's elevation angle. This is the physical adjustment coefficient; Then, the perturbation feature state vector is generated. : ; Among them, disturbance characteristics After nonlinear mapping transformation, three perturbation correlation quantities are obtained: Ionospheric disturbance intensity index This reflects the amplitude of phase fluctuations caused by ionospheric scintillation; Multipath interference intensity index This characterizes the energy variation and attitude dependence of the reflected signal; Signal health score This is a normalized signal stability index, reflecting the overall integrity of the navigation signal; Based on the joint judgment of the three disturbance-related quantities, the disturbance level (Level) is obtained. Integrated Disturbance Modeling Module: The module calculates the integrated disturbance intensity index for integrity modeling by adaptively weighting the disturbance characteristics by disturbance level and smoothing over time. ; ; in This represents the comprehensive disturbance index. These are weighting coefficients that are dynamically adjusted according to the disturbance level. This represents the comprehensive disturbance intensity index, where n is the index of the current time and N is the size of the sliding window; Adaptive Sigma Expansion Coefficient Calculation Module: Dynamically estimates the Sigma expansion coefficient under the current environment by comprehensively considering the disturbance intensity index; Integrity Calculation and Alarm Module: Based on the calculated adaptive Sigma expansion coefficient, the module dynamically calculates and evaluates the overall integrity level at the current moment to obtain an integrity health score. Integrity Enhancement and Adaptive Optimization Module: Based on the integrity and health score, the observation reliability is dynamically weighted; the navigation state covariance matrix is ​​dynamically adjusted based on the adaptive Sigma expansion coefficient; thereby correcting the navigation state estimation model.

[0022] Furthermore, the integrity enhancement and adaptive optimization module also includes: introducing a perturbation compensation term for correction. For integrity enhancement and correction in the navigation solution process: ;in, These are the original pseudorange or carrier observation values. For observations corrected for integrity, It is obtained by jointly estimating the disturbance fusion index and the current disturbance level, and is used to correct the ranging bias caused by signal distortion.

[0023] The present invention has the following beneficial effects: Compared with existing navigation signal integrity enhancement methods, the proposed deep learning-based ionospheric scintillation-multipath coupling adaptive integrity enhancement method has significant technical advantages in terms of structural design, algorithm adaptability, integrity response speed, and model interpretability. This method combines multi-source disturbance feature fusion with a physical perception deep learning model to achieve dynamic evaluation and adaptive optimization of navigation signal integrity, thereby significantly improving the reliability and robustness of the navigation system in complex disturbance environments.

[0024] First, at the structural level, this invention introduces a physically perturbation-constrained gated temporal modeling structure (Physically Aware Long Short-Term Memory Unit - PA-LSTM) to replace the fixed gating mechanism of traditional LSTM. This allows the network to dynamically adjust the memory update rate and forgetting ratio based on the flicker intensity and phase perturbation level of the signal. Simulation results show that, under the same input dimensionality, the PA-LSTM model has approximately 23% fewer parameters than the traditional two-layer LSTM, improves computational efficiency by approximately 17%, and increases perturbation recognition accuracy by approximately 11.4%. This lightweight structure with enhanced physical constraints effectively reduces the risk of model overfitting and improves the generalization ability to complex perturbation signals.

[0025] Secondly, regarding perturbation identification and feature fusion, the proposed dual-branch multi-source fusion structure (I-Branch and M-Branch) can simultaneously characterize ionospheric scintillation and multipath interference features, and achieves adaptive feature weighting through physically constrained fusion layers. Compared with traditional CNN-RNN cascaded networks, this method improves the accuracy of signal perturbation level identification by approximately 12.7% and reduces the false positive rate by approximately 19.3% in multipath-dominated scenarios. This improvement enables the algorithm to have a stronger ability to distinguish ionospheric-multipath coupling phenomena.

[0026] Furthermore, regarding integrity quantification and alarm mechanisms, this invention introduces an adaptive Sigma expansion and integrity scoring model. By comprehensively considering disturbance intensity indicators and system uncertainties, it achieves continuous signal health assessment. Experiments show that in GNSS scintillation datasets (based on COSPAS-SARSAT and SCIDA stations), when the S4 exponent is greater than 0.6 and σφ is higher than 0.4 radians, this method can still maintain an integrity health score Hscore ≥ 0.72, while the traditional fixed threshold algorithm drops to below 0.55, effectively avoiding false alarms and missed alarms.

[0027] In the integrity enhancement and optimization phase, this invention introduces an optimization strategy based on perturbation weight adjustment and adaptive covariance expansion to achieve dynamic correction of the navigation state estimation process. Experimental results show that under moderate ionospheric perturbation (S4≈0.5), the system pseudorange positioning error is reduced from 1.86 m to 1.12 m; under strong perturbation (S4≈0.8), the error is reduced by approximately 38%, and the signal integrity recovery time is shortened by approximately 27%. These results verify the rapid adaptation and integrity recovery capabilities of this method in highly dynamic environments.

[0028] Regarding model interpretability and engineering feasibility, the physical perturbation adjustment channels (parameters γ1, γ2, Σ(t)) of the PA-LSTM model can be directly mapped and associated with ionospheric scintillation parameters, allowing the network's internal response to be interpreted through physical indices. Unlike traditional black-box deep models, this method provides a clear perturbation input-response mechanism, which can be directly applied to spaceborne navigation receivers or ground-based augmentation systems. The computational latency of this invention does not exceed 36 ms, meeting the real-time requirements of aviation integrity applications.

[0029] In summary, this invention, by combining deep temporal modeling, physical disturbance perception, and integrity adaptive optimization, significantly outperforms existing technologies in terms of structural simplification, identification accuracy, alarm stability, and integrity recovery speed. The algorithm has a clear structure, strong embeddability, and can be widely applied in scenarios such as aviation, surveying, unmanned systems, and precision time synchronization, possessing significant engineering value for improving the integrity assurance level and anti-interference robustness of GNSS signals. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention.

[0031] Figure 2 This is a schematic diagram of the disturbance feature extraction and state recognition process in Embodiment 1 of the present invention.

[0032] Figure 3 This is a schematic diagram of the ionospheric branched physical sensing long short-term memory unit PA-LSTM structure in Embodiment 1 of the present invention.

[0033] Figure 4 This is a schematic diagram of the temporal convolutional filter structure for multipath branch sensing physics in Embodiment 1 of the present invention.

[0034] Figure 5 This is a schematic diagram of the KAN architecture in the adaptive Sigma expansion coefficient calculation of Embodiment 1 of the present invention.

[0035] Figure 6 This is a schematic diagram of the system architecture of Embodiment 2 of the present invention. Detailed Implementation

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0037] The design concept of this invention is to address the problems of existing navigation integrity monitoring algorithms in complex environments such as ionospheric scintillation, multipath interference, and changes in flight attitude, which include static integrity modeling, inaccurate error estimation, and lack of dynamic adaptive adjustment mechanisms. This invention proposes an algorithm that can identify ionospheric and multipath coupling disturbance characteristics in real time and adaptively adjust the Sigma Inflation coefficient according to the degree of disturbance. This aims to improve the integrity accuracy and reliability of navigation systems in complex signal environments, ensuring that GLS airborne equipment meets airworthiness standards during the CAT II / III precision approach phase.

[0038] Based on the above design concept, the present invention will be further described below with reference to specific embodiments and accompanying drawings.

[0039] Example 1: The proposed method in this embodiment is an adaptive integrity enhancement method for ionospheric scintillation-multipath coupling based on deep learning, such as... Figure 1 As shown, it includes the following steps: Step S1, Signal Data Acquisition and Preprocessing: Acquire signal data including basic observation data of the Global Navigation Satellite System (GNSS), ionospheric disturbance characteristic data, multipath and spacecraft characteristic data; preprocessing of signal data includes synchronization and normalization.

[0040] The data collected in this step mainly includes: 1. GNSS basic observation data: pseudorange observations carrier phase Signal-to-noise ratio Satellite elevation angle With azimuth .

[0041] 2. Ionospheric disturbance characteristics: amplitude scintillation index Phase flicker index ionospheric delay

[0042] 3. Multipath and Plateau Characteristics: Multipath Residuals Flight attitude angle (roll) , looking up ,yaw Antenna installation azimuth angle .

[0043] The subscript "i" in the above data represents the i-th satellite or the i-th observation path, and is used to distinguish parallel observation data from multiple satellites, multiple paths, and multiple channels.

[0044] To ensure consistency of characteristics, all signal data are in GNSS time. To unify the benchmark, synchronization is achieved through interpolation and resampling. Subsequently, each observation is normalized to construct a feature matrix: ; This feature matrix is ​​used for perturbation feature extraction and state recognition in step S2. It extracts dynamic features reflecting ionospheric and multipath interference. After identification and classification by a deep learning model, it outputs high-dimensional perturbation feature results for subsequent integrity parameter adaptive steps.

[0045] Step S2, Disturbance Feature Extraction and State Recognition: Perform short-time Fourier transform on the signal data to obtain the ionospheric branch and multipath branch. Extract ionospheric disturbance fusion features from the ionospheric branch and extract multipath dynamic features related to the aircraft attitude change and signal geometric propagation path from the multipath branch. Fuse the two to obtain the disturbance features and output the disturbance feature state vector.

[0046] In this step, to capture the temporal dependence and spatial coupling characteristics of disturbances from different sources, the feature matrix obtained in the previous step is... As the input signal, a short-time Fourier transform (STFT) is performed on the input signal to convert the time signal into a time-frequency spectrum form, capturing the transient high-frequency disturbance component such as "ionospheric scintillation"; allowing subsequent models to see the frequency distribution of the disturbance, rather than just looking at time fluctuations.

[0047] The specific process is as follows: Figure 2 As shown, after short-time Fourier transform, the ionospheric branch (I-Branch) and the multipath branch (M-Branch) are obtained.

[0048] I. For the ionospheric branch (I-Branch): With the input signal For the ionospheric branch input, a Physically Aware Long Short-Term Memory (PA-LSTM) structure is used to extract the temporal fluctuation features of ionospheric scintillation. This unit dynamically adjusts the forgetting and input weights through a perturbation gating mechanism to achieve adaptive response to scintillation signals of different intensities, mainly used to identify the temporal correlation under the superposition of slow trends and high-frequency perturbations. The ionospheric delay in the input... In the PA-LSTM structure, it does not directly participate in the gating operation alone, but rather serves as the amplitude flicker index. and phase flicker index The physical constraint terms participate in the gating coefficient adjustment.

[0049] Instead of drawing extra attention, let the disturbance indicators... The gating coefficient is directly adjusted (γ is the physical adjustment weight). In other words, the degree of opening and closing of each gate is dynamically adjusted according to changes in ionospheric disturbance.

[0050] like Figure 3 The diagram shows the structure of a Physically Aware Long Short-Term Memory (PA-LSTM) unit, where: Indicates the input signal at the current moment; This represents the final ionospheric perturbation time series characteristics output; This indicates a dynamic fusion of history and new memories; This indicates the hidden state at the previous moment. In this embodiment, it represents the output of the ionospheric disturbance time series characteristics at the previous moment. This indicates the state of the memory unit at the previous moment; This indicates the hidden state at the current moment; in this embodiment, it represents the output of the ionospheric perturbation timing characteristics at the current moment. Indicates the current state of the memory unit; (1) Forget Gate: Calculate the forget gate opening based on the current input signal and the previous hidden state. : ; in, This is the perturbation adjustment coefficient, used to dynamically adjust the memory retention ratio based on the amplitude flicker intensity.

[0051] Forget gate bias term; Forget gate weight matrix; The Sigmoid function takes values ​​in the range (0, 1).

[0052] When ionospheric scintillation intensifies Larger , Increase gating sensitivity to help forget old memories more quickly, thus responding to sudden disturbances.

[0053] (2) Input Gate: ; Input the gate opening degree to determine the proportion of new information written; Input gate weight matrix; Input gate bias; Phase disturbance adjustment coefficient; When the phase disturbance increases, the input gate opening... Increased size allows new features to enter memory units more quickly.

[0054] (3) Candidate Memory Generation: ; Candidate memory states; Candidate layer weight matrix; : Bias term; Hyperbolic tangent function; Candidate memory terms reflect the potential perturbation response characteristics of the input signal at the current moment.

[0055] (4) Cell State Update: ; The current state of the memory unit; The state of the memory unit in the previous moment; “·” indicates element-wise multiplication.

[0056] This process integrates old memories with new disturbance characteristics, forming a dynamic response to changes over time.

[0057] (5) Output Gate & Hidden State: The output gate controls the propagation intensity of the disturbance response characteristics at the current moment and is coupled with the phase flicker index to form a dynamic adjustment mechanism. Its calculation expression is as follows: ; Output gate control factor, used to adjust the transmission ratio of disturbance information; Output gate weight matrix; Output gate bias term; : Disturbance adjustment term, used to introduce dynamic correction of the output gate by the phase flicker intensity; The nonlinear mapping result of the current state of the memory cell reflects the cumulative effect of the disturbance.

[0058] The output gate adjustment process achieves joint adaptive control of disturbance and memory state. When phase flicker is severe ( When the disturbance is relatively large, the output gate automatically weakens the propagation of unstable features, thereby suppressing the transmission of invalid information; when the disturbance tends to stabilize, the output gate opening increases, enhancing long-term trend memory, enabling the network to maintain stronger time dependence in a stable environment.

[0059] To further characterize the comprehensive temporal features of ionospheric perturbations, this invention introduces a physical fusion weighting mechanism based on the hidden layer output, defining the perturbation fusion feature vector as follows: ; These are the weighting coefficients for adjusting the output of the ionospheric disturbance timing characteristics at the current moment, the state of the memory cell at the current moment, and the amplitude flicker index, respectively.

[0060] The final output ionospheric perturbation fused feature vector As a high-confidence time-series feature input, it provides a foundation for subsequent fusion modeling.

[0061] II. For multipath branches (M-Branch): Multipath branching is used to extract multipath dynamic features related to carrier attitude changes and signal geometric propagation paths, using multi-source feature sets from different satellites or observation paths: For input.

[0062] To characterize the time-varying patterns of multipath interference, the multipath branch adopts a temporal convolutional filter structure based on perceptual physics, such as... Figure 4 As shown, a physical modulation layer is embedded on the basis of the traditional dilated causal convolutional network to adaptively correct the weights of the convolutional kernel through perturbation, thereby enhancing the sensitivity of the convolutional layer to changes in geometric structure and fluctuations in signal energy.

[0063] In the actual implementation, the input signal first passes through the physical modulation layer, and the corresponding convolution kernel coefficients are adjusted according to the signal-to-noise ratio change rate at the current moment. With elevation gradient Dynamic adjustment: ; in, For standard convolution weights, Let be the perturbation mapping function. This is the adjustment coefficient, used to control the modulation intensity.

[0064] This represents the time-varying convolution weights modulated by physical perturbation features, used to dynamically correct the response amplitude of the convolution kernel under different perturbation intensities.

[0065] The physically modulated signal enters the dilated causal convolutional layer to extract energy coupling features at different time scales; then, the signal is suppressed by the normalization layer (WeightNorm) and the nonlinear activation function (ReLU) to enhance the structural change features of the signal.

[0066] In the convolution output stage, this branch introduces a gated convolution layer, which uses the sigmoid control factor of the gate unit to achieve feature selection and preservation of important information. ; in, The main channel convolution weight matrix, These are the gated convolution weight matrices, which control the main feature extraction and gated modulation, respectively. The fused feature map output by the convolutional layer comprehensively reflects the joint response of multipath dynamic features in the main and auxiliary channels.

[0067] This structure can adaptively adjust the convolution response amplitude according to different signal incident angles and multipath delay characteristics, enhancing the model's ability to focus on features during multipath-dominated periods. Finally, the output is added to the main channel via residual connections to obtain: ; Output multipath dynamic feature vector It comprehensively reflects the time correlation, energy distribution characteristics, and coupling relationship with geometric parameters of multipath reflection signals, providing a stable dynamic representation input for subsequent feature fusion layers.

[0068] III. Feature Fusion: The feature fusion layer under physical constraints will integrate multipath dynamic feature vectors. Fusion feature vectors with ionospheric perturbations When fusing two high-dimensional features and calculating the attention weights, physical quantities (such as signal elevation and signal-to-noise ratio change rate) are introduced. ) as a regulating factor; The final fused perturbation characteristics are as follows: ; in The perturbation features are represented, and the perturbation is fused with dynamic attention weights. and multipath dynamic attention weights All of these are adaptively learned by the network during training via backpropagation. Specifically, the model introduces learnable weight parameters during the feature fusion stage and, based on the integrity labels of the training samples (such as different perturbation levels), automatically adjusts the distribution of weights over time by minimizing the prediction error loss function. Therefore, and This reflects the dynamic allocation of the model's attention to ionospheric and multipath features under different perturbation environments.

[0069] The comprehensive characteristic quantity of satellite elevation angle is derived from the elevation angle of each satellite. Calculated by weighted average or normalization, it is used to characterize the strength of multipath geometric relationships; it is used to describe the strength of multipath geometric relationships. This is a physical adjustment coefficient used to balance the contributions of depth features and physical constraints.

[0070] The main function of the physical constraint attention mechanism is to prevent the network from learning blindly, but to learn with the knowledge of physics.

[0071] The high-dimensional feature vector output by the fusion layer After nonlinear mapping transformation, three perturbation correlation quantities are obtained: Ionospheric disturbance intensity index This reflects the amplitude of phase fluctuations caused by ionospheric scintillation; Multipath interference intensity index This characterizes the energy variation and attitude dependence of the reflected signal; Signal health score , is a normalized signal stability index that reflects the overall integrity of the navigation signal.

[0072] Based on the above three disturbance-related quantities, disturbance state identification and level classification are performed. The level rules are based on the joint judgment of signal health score, ionospheric disturbance intensity index, and multipath interference intensity index, and are defined as follows: Table 1: Status Level Table

[0073] Finally, step S2 outputs the perturbation feature state vector: ; The perturbation feature state vector comprehensively reflects the dynamic characteristics of the ionosphere and multipath perturbations, as well as the overall integrity and health status of the signal, and serves as the input basis for integrity calculation and Sigma adaptive adjustment in subsequent steps.

[0074] Step S3: Generate a comprehensive disturbance intensity index: The disturbance characteristics are calculated by adaptive weighting of disturbance level and time smoothing to generate a comprehensive disturbance intensity index for integrity modeling.

[0075] This step receives the perturbation characteristic state vector output from step S2. By adaptively weighting the disturbance level and calculating with time smoothing, a comprehensive disturbance intensity index that can be used for integrity modeling is generated.

[0076] Based on the disturbance level label (Level), different weighting coefficients are assigned to the ionosphere and multipath characteristics to reflect the dominant influence of different disturbance sources. The formula for calculating the comprehensive disturbance index is as follows: ; in, These are weighting coefficients that are dynamically adjusted according to the disturbance level. For example, when the level is high (severe disturbance), the weighting coefficient is increased. To amplify the impact of the ionosphere and multipath propagation on overall risk, a moving average smoothing method was used on the results to suppress short-term fluctuations. ; This formula is used to calculate the weighted average over a sliding window of N, where n is the index of the current time step, representing the average over the past N time steps. N is the size of the sliding window, which determines the amount of historical data used during smoothing.

[0077] Final output The comprehensive disturbance intensity index characterizes the combined intensity of ionospheric and multipath disturbances experienced by the navigation signal at the current moment. It serves as the input for the next step (adaptive Sigma expansion coefficient calculation) and is used to dynamically correct the sensitivity and margin of the navigation system's integrity protection limit.

[0078] Step S4: Perform adaptive Sigma expansion coefficient calculation: Dynamically estimate the Sigma expansion coefficient under the current environment by comprehensively considering the disturbance intensity index.

[0079] To achieve adaptive adjustment of the integrity error distribution by the disturbance, this step introduces a disturbance-driven dynamic response mapping mechanism during the Sigma calculation. Unlike the traditional linear weighting method, this embodiment does not directly superimpose various disturbance characteristics, but rather synthesizes the disturbance intensity index... Combined with the disturbance level label (Level), an adaptive inflation ratio adjustment model is established to dynamically estimate the Sigma inflation coefficient under the current environment. Its calculation function is defined as follows: ; in, This represents the performance benchmark of the navigation system under conditions free from external disturbances, typically the default strength state of the ionosphere and multipath signals. This value is used to dynamically adjust the overall strength of the current navigation system and provides a benchmark for subsequent system optimization.

[0080] This represents the adaptive Sigma expansion coefficient.

[0081] The nonlinear mapping function obtained through training is implemented using a learnable parameterized neural network structure, such as a multi-layer nonlinear learning unit based on the KAN architecture, as shown in Figure 5. This unit has environmental adaptability and can automatically adjust the expansion amplitude according to the temporal changes of the perturbation features, realizing a dynamic adjustment process from experience-dependent to data-driven.

[0082] The core principle of this step lies in automatically estimating the optimal Sigma expansion adjustment amount at each moment based on the comprehensive disturbance intensity index and historical operational data. By introducing a nonlinear learning layer of a neural network, it can sense the dominance of different disturbance sources (ionospheric scintillation, multipath interference) and perform weighted correction on the expansion ratio in the corresponding direction. When ionospheric disturbances dominate, the vertical expansion adjustment can be automatically enhanced to improve the integrity of the system and suppress ionospheric anomalies; when multipath interferences dominate, the weight in that direction is adaptively reduced to maintain the overall steady-state response, thereby avoiding excessive amplification of the error distribution.

[0083] Step S5: Perform integrity calculation: Based on the calculated adaptive Sigma expansion coefficient, dynamically calculate and evaluate the overall integrity level at the current moment to obtain an integrity health score.

[0084] After calculating the adaptive Sigma expansion coefficient, this step dynamically assesses the overall integrity level of the navigation system at the current moment and issues real-time alarms based on the assessment results. The core function of this step is to comprehensively consider factors such as the overall disturbance intensity index, the adaptive Sigma expansion coefficient, and the disturbance level to quantitatively describe the health status of the navigation signal, outputting an integrity health score and alarm flags for subsequent integrity assurance and safety control.

[0085] Firstly, based on the comprehensive disturbance intensity index Disturbance Level and Adaptive Sigma Expansion Coefficient As input, a multi-source integrity quantification model is constructed. Considering the dynamic changes of navigation signals under the combined effects of ionospheric scintillation, multipath interference, and system noise expansion, this invention defines an integrity and health scoring function as follows: ; in, To balance the influence of various factors, a weighted parameter is used. This model integrates the influence of different disturbance sources through an exponential decay and linear combination method, achieving a continuous characterization of navigation signal health. When ionospheric disturbances or multipath interference intensify, the overall disturbance intensity index is adjusted. Increase, or adapt the Sigma expansion coefficient Improvement will lead to A significant decrease, characterizing a reduction in the level of system integrity.

[0086] To achieve integrity monitoring and rapid response to anomalies, this step further sets integrity alarm thresholds. And determine the status of navigation signals in real time based on the health score: ; When the health score is below the threshold When this occurs, an integrity alarm signal will be automatically triggered. This indicates a decrease in the reliability of the current navigation signal. The threshold value varies depending on the application scenario. The sensitivity and stability can be adjusted by adaptively setting historical statistics or empirical parameters to ensure a balance between sensitivity and stability under different environments.

[0087] Through the above modeling, this step realizes a real-time assessment and alarm triggering mechanism for navigation signal integrity under multi-source disturbance environments. Compared with traditional fixed threshold or single-source monitoring methods, this step can dynamically adjust the risk sensitivity according to the signal disturbance intensity and system adaptive expansion, making the alarm response more robust and physically consistent. The final output... This represents the overall integrity level of the navigation system at the current moment. These are the corresponding integrity status indicators, and together they provide a basis for ensuring the integrity of the navigation system and for adaptive adjustments.

[0088] Step S6, Integrity Enhancement and Adaptive Optimization: Based on the integrity and health score, the observation reliability is dynamically weighted; the navigation state covariance matrix is ​​dynamically adjusted based on the adaptive Sigma expansion coefficient; thereby correcting the navigation state estimation model.

[0089] The signal health score output in the previous step With integrity warning sign Building upon the previous steps, this step establishes an adaptive optimization mechanism to enhance the integrity and stabilize the performance of the navigation system under disturbed environments. This step serves as the closed-loop decision-making mechanism for the entire method, dynamically adjusting data weights, filtering parameters, and model confidence levels to reduce positioning errors and integrity degradation caused by ionospheric scintillation and multipath coupling.

[0090] First, the output of the previous step is received, and then combined with historical state information to comprehensively evaluate the current signal quality. When High (e.g., above the preset threshold) When ), maintain normal navigation status; when Reduce or When an alarm is triggered, the integrity enhancement mechanism is automatically activated, and the navigation state estimation model is corrected through adaptive weight adjustment and estimation covariance amplification strategies.

[0091] In this mechanism, a dynamically weighted observation confidence update strategy is designed in this step: ; in, This represents the observation weight matrix at the current time. For standard weights, This is the disturbance sensitivity coefficient. When the disturbance increases or the integrity decreases, the weighting factor automatically decays, thus suppressing the impact of unreliable observations on the navigation system estimation. Simultaneously, the state covariance matrix of the navigation system... Based on the adaptive Sigma expansion coefficient Dynamic adjustment: ; in, This is the adjusted covariance matrix. When external disturbances cause an increase in measurement noise, It automatically relaxes covariance constraints, enhancing the robustness of the filtering model; when the signal environment is stable, Gradually restore to 1, thus maintaining the accuracy of the estimate.

[0092] In addition, to further suppress pseudorange and carrier observation anomalies caused by ionospheric scintillation, this step introduces a perturbation compensation term correction. For integrity enhancement and correction in the navigation solution process: ; in, These are the original pseudorange or carrier observation values. For observations corrected for integrity, It is obtained by jointly estimating the adaptive Sigma expansion coefficient and the current disturbance level, and is used to correct the ranging error caused by signal distortion.

[0093] Through the aforementioned dynamic weighting and covariance adjustment mechanism, this step achieves real-time adaptive optimization of the navigation state estimation process, ensuring stable solutions and reliable results even under strong disturbances. When integrity alarms continue to be triggered, source-level switching or multi-mode combination strategies can be further triggered to achieve optimal fusion of multi-source navigation data, thereby ensuring continuous availability and integrity maintenance.

[0094] In summary, this step introduces a physical disturbance-driven adaptive optimization mechanism on the basis of the traditional navigation state estimation framework. Through the integrity feedback closed loop, it realizes dynamic weight adjustment, noise adaptive expansion and observation correction under disturbance conditions, thereby significantly improving the robustness and integrity assurance capability of the navigation system in complex space environments.

[0095] Example 2: This embodiment proposes an ionospheric scintillation-multipath coupling adaptive integrity enhancement system based on deep learning, such as... Figure 6 As shown, it includes: Signal data acquisition and preprocessing module: Signal data includes basic observation data of Global Navigation Satellite System (GNSS), ionospheric disturbance characteristic data, multipath and spacecraft characteristic data; preprocessing of signal data includes synchronization and normalization; The disturbance feature extraction and state recognition module performs short-time Fourier transform on the signal data to obtain ionospheric branches and multipath branches; The ionospheric perturbation fusion features are extracted using a Physically Aware Long Short-Term Memory (PA-LSTM) structure for ionospheric branches. ;in This represents the fused feature vector of ionospheric perturbations. The output of the time series characteristics of ionospheric perturbation at the current moment is adjusted respectively. Current state of the memory unit With amplitude flicker index The weighting coefficients of the three; The multipath branch is given a multi-source feature set from different satellites or observation paths as input, and a temporal convolutional filter based on perceptual physics is used to extract the multipath dynamic features. ;in Represents the dynamic feature vector of multipath. This represents a multi-source feature set from different satellites or observation paths. This represents the fused feature map of the convolutional layer output; The feature fusion layer under physical constraints fuses two high-dimensional features, namely ionospheric perturbation fusion feature and multipath dynamic feature, to obtain perturbation feature; ; in, Representing perturbation features; perturbation fusion dynamic attention weights and multipath dynamic attention weights It is obtained through adaptive learning via backpropagation during training. This represents the rate of change of the signal-to-noise ratio. This is a comprehensive characteristic quantity of the satellite's elevation angle. This is the physical adjustment coefficient; Then, the perturbation feature state vector is generated. : ; Among them, disturbance characteristics After nonlinear mapping transformation, three perturbation correlation quantities are obtained: Ionospheric disturbance intensity index This reflects the amplitude of phase fluctuations caused by ionospheric scintillation; Multipath interference intensity index This characterizes the energy variation and attitude dependence of the reflected signal; Signal health score This is a normalized signal stability index, reflecting the overall integrity of the navigation signal; Based on the joint judgment of the three disturbance-related quantities, the disturbance level (Level) is obtained. Integrated Disturbance Modeling Module: The module calculates the integrated disturbance intensity index for integrity modeling by adaptively weighting the disturbance characteristics by disturbance level and smoothing over time. ; ; in This represents the comprehensive disturbance index. These are weighting coefficients that are dynamically adjusted according to the disturbance level. This represents the comprehensive disturbance intensity index, where n is the index of the current time and N is the size of the sliding window; Adaptive Sigma Expansion Coefficient Calculation Module: Dynamically estimates the Sigma expansion coefficient under the current environment by comprehensively considering the disturbance intensity index; Integrity Calculation and Alarm Module: Based on the calculated adaptive Sigma expansion coefficient, the module dynamically calculates and evaluates the overall integrity level at the current moment to obtain an integrity health score. Integrity Enhancement and Adaptive Optimization Module: Based on the integrity and health score, the observation reliability is dynamically weighted; the navigation state covariance matrix is ​​dynamically adjusted based on the adaptive Sigma expansion coefficient; thereby correcting the navigation state estimation model.

[0096] The integrity enhancement and adaptive optimization module also includes: introducing a perturbation compensation term correction. For integrity enhancement and correction in the navigation solution process: ;in, These are the original pseudorange or carrier observation values. For observations corrected for integrity, It is obtained by jointly estimating the disturbance fusion index and the current disturbance level, and is used to correct the ranging bias caused by signal distortion.

[0097] The preprocessing in the signal data acquisition and preprocessing module includes: all signal data are synchronized by interpolation and resampling using GNSS time as a unified reference; then, each signal data is normalized to construct a feature matrix.

[0098] In the adaptive Sigma expansion coefficient calculation module, a disturbance-driven dynamic response mapping mechanism is introduced during the Sigma calculation process. By combining the comprehensive disturbance intensity index and the disturbance level label Level as inputs, an adaptive expansion ratio adjustment model is established to dynamically estimate the Sigma expansion coefficient under the current environment.

[0099] In the integrity calculation and alarm module, a multi-source integrity quantification model is constructed using the comprehensive disturbance intensity index, disturbance level label Level, and adaptive Sigma expansion coefficient as inputs, and an integrity health scoring function is defined.

[0100] In the integrity enhancement and adaptive optimization module, a dynamically weighted observation confidence update strategy is designed: ; in, This represents the observation weight matrix at the current time. For standard weights, The disturbance sensitivity coefficient, To comprehensively assess the intensity of the disturbance, The score is given for the integrity of sexual health.

[0101] The ionospheric scintillation-multipath coupling adaptive integrity enhancement system proposed in this embodiment can achieve the ionospheric scintillation-multipath coupling adaptive integrity enhancement method based on deep learning as described in Embodiment 1, and has the same technical effect as Embodiment 1.

[0102] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling, characterized in that, include: S1. Signal Data Acquisition and Preprocessing: Acquired signal data includes basic observation data of the Global Navigation Satellite System (GNSS), ionospheric disturbance characteristic data, multipath and spacecraft characteristic data; preprocessing of signal data includes synchronization and normalization. S2. Disturbance feature extraction and state recognition: Short-time Fourier transform is performed on the signal data to obtain ionospheric branches and multipath branches; The ionospheric perturbation fusion features are extracted using a Physically Aware Long Short-Term Memory (PA-LSTM) structure for ionospheric branches. ;in This represents the fused feature vector of ionospheric perturbations. The output of the time series characteristics of ionospheric perturbation at the current moment is adjusted respectively. Current state of the memory unit With amplitude flicker index The weighting coefficients of the three; The multipath branch is given a multi-source feature set from different satellites or observation paths as input, and a temporal convolutional filter based on perceptual physics is used to extract the multipath dynamic features. ;in Represents the dynamic feature vector of multipath. This represents a multi-source feature set from different satellites or observation paths. This represents the fused feature map of the convolutional layer output; The feature fusion layer under physical constraints fuses two high-dimensional features, namely ionospheric perturbation fusion feature and multipath dynamic feature, to obtain perturbation feature; ； in, Representing perturbation features; perturbation fusion dynamic attention weights and multipath dynamic attention weights It is obtained through adaptive learning via backpropagation during training. This represents the rate of change of the signal-to-noise ratio. This is a comprehensive characteristic quantity of the satellite's elevation angle. This is the physical adjustment coefficient; Then, the perturbation feature state vector is generated. : ; Among them, disturbance characteristics After nonlinear mapping transformation, three perturbation correlation quantities are obtained: Ionospheric disturbance intensity index This reflects the amplitude of phase fluctuations caused by ionospheric scintillation; Multipath interference intensity index This characterizes the energy variation and attitude dependence of the reflected signal; Signal health score This is a normalized signal stability index, reflecting the overall integrity of the navigation signal; Based on the joint judgment of the three disturbance-related quantities, the disturbance level (Level) is obtained. S3. Generate a comprehensive disturbance intensity index: The disturbance characteristics are calculated by adaptive weighting of disturbance level and time smoothing to generate a comprehensive disturbance intensity index for integrity modeling. ； ； in This represents the comprehensive disturbance index. These are weighting coefficients that are dynamically adjusted according to the disturbance level. This represents the comprehensive disturbance intensity index, where n is the index of the current time and N is the size of the sliding window; S4. Perform adaptive Sigma expansion coefficient calculation: dynamically estimate the Sigma expansion coefficient under the current environment by comprehensively considering the disturbance intensity index; S5. Perform integrity calculation: Based on the calculated adaptive Sigma expansion coefficient, dynamically calculate and evaluate the overall integrity level at the current moment to obtain an integrity health score; S6. Integrity Enhancement and Adaptive Optimization: The observation reliability is dynamically weighted based on the integrity and health score; the navigation state covariance matrix is ​​dynamically adjusted based on the adaptive Sigma expansion coefficient; thereby correcting the navigation state estimation model.

2. The deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling according to claim 1, characterized in that, Step S6 also includes: introducing a disturbance compensation term correction. For integrity enhancement and correction in the navigation solution process: ;in, These are the original pseudorange or carrier observation values. For observations corrected for integrity, It is obtained by jointly estimating the disturbance fusion index and the current disturbance level, and is used to correct the ranging bias caused by signal distortion.

3. The deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling according to claim 1, characterized in that, The preprocessing in step S1 includes: all signal data are synchronized by interpolation and resampling using GNSS time as a unified reference; then, each signal data is normalized to construct a feature matrix.

4. The deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling according to claim 1, characterized in that, In step S4, a disturbance-driven dynamic response mapping mechanism is introduced into the Sigma calculation process. An adaptive expansion ratio adjustment model is established through the joint input of the comprehensive disturbance intensity index and the disturbance level label Level, which is used to dynamically estimate the Sigma expansion coefficient under the current environment.

5. The deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling according to claim 1, characterized in that, In step S5, a multi-source integrity quantification model is constructed using the comprehensive disturbance intensity index, the disturbance level label Level, and the adaptive Sigma expansion coefficient as inputs, and an integrity health scoring function is defined.

6. The deep learning-based adaptive integrity enhancement method for ionospheric scintillation-multipath coupling according to claim 1 or 2, characterized in that, In step S6, a dynamically weighted observation confidence update strategy was designed: ； in, This represents the observation weight matrix at the current time. For standard weights, The disturbance sensitivity coefficient, To comprehensively assess the intensity of the disturbance, The score is given for the integrity of sexual health.

7. A deep learning-based adaptive integrity enhancement system for ionospheric scintillation-multipath coupling, characterized in that, include: Signal data acquisition and preprocessing module: Signal data includes basic observation data of Global Navigation Satellite System (GNSS), ionospheric disturbance characteristic data, multipath and spacecraft characteristic data; preprocessing of signal data includes synchronization and normalization; The disturbance feature extraction and state recognition module performs short-time Fourier transform on the signal data to obtain ionospheric branches and multipath branches; The ionospheric perturbation fusion features are extracted using a Physically Aware Long Short-Term Memory (PA-LSTM) structure for ionospheric branches. ;in This represents the fused feature vector of ionospheric perturbations. The output of the time series characteristics of ionospheric perturbation at the current moment is adjusted respectively. Current state of the memory unit With amplitude flicker index The weighting coefficients of the three; The multipath branch is given a multi-source feature set from different satellites or observation paths as input, and a temporal convolutional filter based on perceptual physics is used to extract the multipath dynamic features. ;in Represents the dynamic feature vector of multipath. This represents a multi-source feature set from different satellites or observation paths. This represents the fused feature map of the convolutional layer output; The feature fusion layer under physical constraints fuses two high-dimensional features, namely ionospheric perturbation fusion feature and multipath dynamic feature, to obtain perturbation feature; ； in, Representing perturbation features; perturbation fusion dynamic attention weights and multipath dynamic attention weights It is obtained through adaptive learning via backpropagation during training. This represents the rate of change of the signal-to-noise ratio. This is a comprehensive characteristic quantity of the satellite's elevation angle. This is the physical adjustment coefficient; Then, the perturbation feature state vector is generated. : ; Among them, disturbance characteristics After nonlinear mapping transformation, three perturbation correlation quantities are obtained: Ionospheric disturbance intensity index This reflects the amplitude of phase fluctuations caused by ionospheric scintillation; Multipath interference intensity index This characterizes the energy variation and attitude dependence of the reflected signal; Signal health score This is a normalized signal stability index, reflecting the overall integrity of the navigation signal; Based on the joint judgment of the three disturbance-related quantities, the disturbance level (Level) is obtained. Integrated Disturbance Modeling Module: The module calculates the integrated disturbance intensity index for integrity modeling by adaptively weighting the disturbance characteristics by disturbance level and smoothing over time. ； ； in This represents the comprehensive disturbance index. These are weighting coefficients that are dynamically adjusted according to the disturbance level. This represents the comprehensive disturbance intensity index, where n is the index of the current time and N is the size of the sliding window; Adaptive Sigma Expansion Coefficient Calculation Module: Dynamically estimates the Sigma expansion coefficient under the current environment by comprehensively considering the disturbance intensity index; Integrity Calculation and Alarm Module: Based on the calculated adaptive Sigma expansion coefficient, the module dynamically calculates and evaluates the overall integrity level at the current moment to obtain an integrity health score. Integrity Enhancement and Adaptive Optimization Module: Based on the integrity and health score, the observation reliability is dynamically weighted; the navigation state covariance matrix is ​​dynamically adjusted based on the adaptive Sigma expansion coefficient; thereby correcting the navigation state estimation model.

8. The deep learning-based adaptive integrity enhancement system for ionospheric scintillation-multipath coupling according to claim 7, characterized in that, The integrity enhancement and adaptive optimization module also includes: introducing a perturbation compensation term correction. For integrity enhancement and correction in the navigation solution process: ;in, These are the original pseudorange or carrier observation values. For observations corrected for integrity, It is obtained by jointly estimating the disturbance fusion index and the current disturbance level, and is used to correct the ranging bias caused by signal distortion.

Images