High-precision continuous positioning method based on inertial navigation GPS

Abstract

The application is suitable for the field of satellite navigation technology, and provides a GPS high-precision continuous positioning method based on inertial navigation, which comprises the following steps: collecting vehicle motion characteristics, environment semantics and ionospheric scintillation probability index, fusing to obtain comprehensive observation untrustworthiness, generating a unified control parameter vector through hierarchical decision, dynamically adjusting the observation fusion strategy of the integrated navigation filter, switching the filter coupling mode and outputting the continuous positioning result. The application realizes adaptive regulation and control by comprehensively interfering with the ground and space, isolates the GPS distortion observation, relies on the inertial navigation to ensure continuous and stable positioning, and significantly improves the GPS positioning accuracy and continuity in complex scenes.

Description

Technical Field

[0001] This invention belongs to the field of satellite navigation technology, and particularly relates to a GPS high-precision continuous positioning method based on inertial navigation. Background Technology

[0002] The increasing demands for positioning accuracy and continuity in fields such as intelligent vehicles and autonomous driving have led to the mainstream application of GPS and inertial navigation combined positioning, widely used in urban roads and cross-regional driving scenarios. Combined navigation, using Kalman filtering as its core fusion method and leveraging the continuous calculation capabilities of inertial navigation to compensate for the susceptibility of GPS signals to interference, has become a key technology for ensuring vehicle positioning reliability. However, the ability to adapt to interference in complex scenarios remains a core challenge for the industry's development.

[0003] Current mainstream integrated navigation and positioning technologies directly couple and fuse GPS observation data with inertial navigation data, and complete state estimation and measurement updates through Kalman filtering with fixed parameters. Some improved schemes simply adjust the filter parameters based on single vehicle motion characteristics or local environmental information, without constructing a multi-dimensional interference perception and hierarchical control mechanism.

[0004] Existing integrated navigation and positioning methods cannot comprehensively identify the dual interference from the ground-based building environment and the space ionosphere. The observation information fusion strategy lacks dynamic adaptive adjustment capabilities, and in complex interference scenarios, it is prone to filter divergence, positioning accuracy drift, and positioning interruption, making it difficult to achieve high-precision continuous positioning of GPS. Summary of the Invention

[0005] The purpose of this invention is to provide a GPS high-precision continuous positioning method based on inertial navigation, which aims to solve the technical problems existing in the prior art as identified in the background art.

[0006] This invention is implemented as follows: a GPS high-precision continuous positioning method based on inertial navigation, the method comprising:

[0007] Collect multi-dimensional scene data during vehicle operation. The multi-dimensional scene data includes vehicle motion feature data output by the inertial navigation system, environmental semantic information, and ionospheric scintillation probability index.

[0008] The risk coefficient of the urban canyon pattern is calculated based on the vehicle motion feature data and the environmental semantic information, and the ionospheric scintillation threat coefficient is calculated based on the ionospheric scintillation probability index. The risk coefficient of the urban canyon pattern and the ionospheric scintillation threat coefficient are multiplied and fused to obtain the comprehensive observation unreliability.

[0009] Based on the comprehensive observation unreliability, a hierarchical decision scheduling is performed to generate a unified control parameter vector containing pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command.

[0010] Based on the unified control parameter vector, the observation information fusion strategy in the integrated navigation filter is dynamically adjusted. The adjusted pseudorange observation noise variance scaling factor and Doppler velocity observation gain weight are applied to the filter measurement update stage. The coupling relationship between the filter state prediction and measurement update is switched according to the filter coupling mode command, and continuous positioning results are output.

[0011] As a further aspect of the present invention, the collection of multi-dimensional scene data during vehicle operation specifically includes:

[0012] The vehicle's longitudinal acceleration, lateral acceleration, and heading angle change rate are collected in real time by an inertial navigation system. The longitudinal acceleration change frequency is calculated based on the longitudinal acceleration, and the heading angle change rate amplitude is calculated based on the heading angle change rate. The longitudinal acceleration change frequency and the heading angle change rate amplitude are then subjected to time-domain statistical feature extraction to generate vehicle motion feature data.

[0013] By matching the vehicle's current location with a high-precision vehicle map, information on road grade, roadside building density, and the proportion of space under viaducts is obtained, and the information on road grade, roadside building density, and the proportion of space under viaducts is used as environmental semantic information.

[0014] By receiving the total electron content data of the ionosphere broadcast by the ionospheric grid model, the current UTC time and vehicle latitude and longitude coordinates are obtained. The current UTC time, the vehicle latitude and longitude coordinates and the total electron content data of the ionosphere are input into the ionospheric scintillation probability prediction model, and the ionospheric scintillation probability prediction model outputs the ionospheric scintillation probability index.

[0015] As a further aspect of the present invention, obtaining the comprehensive observational unreliability specifically includes:

[0016] The longitudinal acceleration change frequency and the heading angle change rate amplitude are input into a pre-constructed urban canyon motion characteristic model. The urban canyon motion characteristic model outputs a motion characteristic risk score. The roadside building density is compared with a preset building density threshold and normalized to a building environment risk score. The motion characteristic risk score and the building environment risk score are weighted and summed to obtain the urban canyon model risk coefficient.

[0017] The ionospheric scintillation probability index is compared with a preset ionospheric scintillation threshold. When the ionospheric scintillation probability index is less than or equal to the ionospheric scintillation threshold, the ionospheric scintillation threat coefficient is set to zero. When the ionospheric scintillation probability index is greater than or equal to the ionospheric scintillation threshold, the ionospheric scintillation probability index is normalized, and the normalized ionospheric scintillation probability index is used as the ionospheric scintillation threat coefficient.

[0018] The risk coefficient of the urban canyon model and the ionospheric scintillation threat coefficient are multiplied together, and the result of the multiplication is used as the overall observational unreliability.

[0019] As a further embodiment of the present invention, the steps for constructing the urban canyon motion feature model are as follows:

[0020] Several sets of vehicle calibration data are collected. Each set of calibration data includes the longitudinal acceleration change frequency, heading angle change rate amplitude, and corresponding scene type label obtained at the same sampling time. The scene type label is used to identify the scene in which the vehicle is located at that sampling time.

[0021] The longitudinal acceleration change frequency and the heading angle change rate amplitude are used as input features, and the scene type label is used as the output label. The support vector machine algorithm is used for training to construct an urban canyon motion feature model. The urban canyon motion feature model includes a classification decision function for mapping the input features to motion feature risk scores.

[0022] When calling the urban canyon motion feature model, the longitudinal acceleration change frequency and the heading angle change rate amplitude are input into the classification decision function, which calculates and outputs the motion feature risk score.

[0023] As a further aspect of the present invention, the generation of a unified control parameter vector comprising pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command specifically includes:

[0024] The overall observation unreliability is compared with a preset first unreliability threshold and a second unreliability threshold, wherein the first unreliability threshold is less than the second unreliability threshold. When the overall observation unreliability is less than or equal to the first unreliability threshold, it is determined that the current mode is the standard fusion mode, the pseudorange observation noise variance scaling factor is set to 1.0, the Doppler velocity observation gain weight is set to 1.0, and the filter coupling mode command is set to normal coupling mode.

[0025] When the first unreliability threshold < the comprehensive observation unreliability ≤ the second unreliability threshold, it is determined that the current state is in the urban canyon transition mode. The pseudorange observation noise variance scaling factor is dynamically calculated based on the ratio of the difference between the comprehensive observation unreliability and the first and second unreliability thresholds. The Doppler velocity observation gain weight is set to a preset enhancement weight value, and the filter coupling mode command is set to the adaptive coupling mode.

[0026] When the overall observation unreliability exceeds the second unreliability threshold, it is determined that the current state is in ionospheric scintillation suppression mode. The pseudorange observation noise variance scaling factor is set to the preset maximum scaling limit value, the Doppler velocity observation gain weight is set to the preset maximum gain limit value, and the filter coupling mode command is set to covariance freeze mode.

[0027] The pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command are vectorized and encapsulated to generate the unified control parameter vector.

[0028] As a further embodiment of the present invention, the output of continuous positioning results specifically includes:

[0029] Obtain the original pseudorange observation noise variance matrix of the integrated navigation filter, and multiply the pseudorange observation noise variance scaling factor with the original pseudorange observation noise variance matrix to obtain the adjusted pseudorange observation noise variance matrix.

[0030] Obtain the original Doppler velocity observation gain matrix of the integrated navigation filter, and multiply the Doppler velocity observation gain weight with the original Doppler velocity observation gain matrix to obtain the adjusted Doppler velocity observation gain matrix.

[0031] According to the filter coupling mode instruction, the corresponding filter coupling mode switching is executed. When the filter coupling mode instruction is normal coupling mode, the state prediction and measurement update are executed sequentially according to the standard Kalman filter process. When the filter coupling mode instruction is adaptive coupling mode, the adjusted pseudorange observation noise variance matrix is ​​used as the fusion weight of pseudorange observation for state update in the measurement update stage.

[0032] When the filter coupling mode instruction is covariance freeze mode, the recursive operation of the covariance update matrix of the pseudorange residual in the measurement update is suspended, and the state prediction is performed using the Doppler velocity observation and the error model of the inertial navigation system corresponding to the adjusted Doppler velocity observation gain matrix.

[0033] The accelerometer zero bias in the inertial navigation system is refined by using the last batch of pseudorange data that has passed the preset quality inspection before covariance freezing. The continuous positioning result is output based on the refined accelerometer zero bias and the state prediction result.

[0034] As a further embodiment of the present invention, the method further includes the steps of performance evaluation and strategy correction of hierarchical decision scheduling, specifically including:

[0035] During the execution of the observation information fusion strategy, the innovation sequence output by the integrated navigation filter is acquired in real time, the Gaussian white noise hypothesis is tested on the innovation sequence, and the autocorrelation coefficient and chi-square test statistic of the innovation sequence are calculated.

[0036] When the autocorrelation coefficient exceeds the preset correlation threshold, it is determined that the efficiency of the current hierarchical decision scheduling does not meet the Gaussian white noise assumption, triggering the scene recognition deviation detection mechanism, obtaining the urban canyon mode risk coefficient and the ionospheric scintillation threat coefficient corresponding to the comprehensive observation unreliability at the current moment, comparing the urban canyon mode risk coefficient and the ionospheric scintillation threat coefficient with the mean in the historical sliding window respectively, and identifying whether there is a deviation in the current scene recognition.

[0037] When a deviation in scene recognition is identified, a correction factor is calculated based on the statistical characteristics of the information sequence, and the first and second unreliability thresholds used in generating the unified control parameter vector are dynamically adjusted using the correction factor.

[0038] The beneficial effects of this invention are:

[0039] This invention constructs an adaptive integrated navigation and positioning system for complex interference environments through full-process optimization including multi-dimensional scene perception, comprehensive interference quantification, hierarchical decision scheduling, and dynamic filtering control. It can comprehensively identify and adapt to GPS observation distortion caused by urban canyons and ionospheric scintillation, dynamically adjust the observation fusion weights and filtering coupling relationship, effectively isolate the impact of distorted observation data on the filtering system, and rely on the continuous positioning characteristics of inertial navigation to ensure uninterrupted positioning output. At the same time, it finely calibrates the core error parameters of inertial navigation to avoid filtering divergence and positioning result drift, significantly improving the anti-interference capability, positioning accuracy, and continuous working stability of the integrated navigation system in various complex driving scenarios, meeting the high-precision continuous positioning requirements of all-weather and all-scenario environments. Attached Figure Description

[0040] Figure 1 A flowchart illustrating a GPS high-precision continuous positioning method based on inertial navigation provided in an embodiment of the present invention;

[0041] Figure 2This is a flowchart illustrating the process of collecting multi-dimensional scene data during vehicle operation, as provided in an embodiment of the present invention.

[0042] Figure 3 A flowchart for obtaining the overall observational unreliability provided in this embodiment of the invention;

[0043] Figure 4 A flowchart for generating a unified control parameter vector provided in an embodiment of the present invention;

[0044] Figure 5 A flowchart illustrating the output of continuous positioning results provided in an embodiment of the present invention;

[0045] Figure 6 This is a flowchart for evaluating the effectiveness and correcting the strategy of hierarchical decision scheduling, provided as an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0047] Figure 1 A flowchart of a GPS high-precision continuous positioning method based on inertial navigation provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes:

[0048] S100, collect multi-dimensional scene data during vehicle driving process, the multi-dimensional scene data includes vehicle motion feature data output by inertial navigation system, environmental semantic information and ionospheric scintillation probability index;

[0049] Using an inertial navigation system as the core carrier, the system collects basic motion parameters such as longitudinal acceleration, lateral acceleration, and rate of change of heading angle in real time. By calculating the frequency of change of longitudinal acceleration and the amplitude of the rate of change of heading angle, and then extracting time-domain statistical features from the two types of calculation results, motion feature data that can fully characterize the dynamic characteristics of vehicle driving is generated, accurately capturing the inherent laws of vehicle acceleration, deceleration, steering and other actions.

[0050] Simultaneously, relying on the vehicle's high-precision map, the environment of the vehicle's current location is matched, and information such as road grade, roadside building density, and space ratio under overpasses is extracted. The distribution of roads and buildings in the physical space is transformed into quantifiable environmental semantic data, which intuitively presents the occlusion and multipath interference conditions in the GPS signal propagation path.

[0051] Simultaneously, by receiving the total electron content data of the ionosphere broadcast by the ionospheric grid model, and combining it with the current UTC time and vehicle latitude and longitude coordinates, the spatiotemporal parameters and ionospheric state parameters are input into a dedicated forecasting model, and the ionospheric scintillation probability index is output, thus completing the quantitative perception of the risk of interference from the space electromagnetic environment.

[0052] S200, calculate the urban canyon pattern risk coefficient based on the vehicle motion feature data and the environmental semantic information, and calculate the ionospheric scintillation threat coefficient based on the ionospheric scintillation probability index. Multiply and fuse the urban canyon pattern risk coefficient and the ionospheric scintillation threat coefficient to obtain the comprehensive observation unreliability.

[0053] Based on vehicle motion characteristic data and environmental semantic information, the risk coefficient of the urban canyon model is calculated. The longitudinal acceleration change frequency and heading angle change rate amplitude are input into the trained urban canyon motion characteristic model, and the motion characteristic risk score matching the urban canyon scene is output. At the same time, the roadside building density is normalized according to the rules to become the building environment risk score. Then, the two types of scores are integrated by weighted summation to form the urban canyon model risk coefficient that comprehensively reflects the degree of ground environment interference to GPS signals, and fully covers the dual interference assessment of vehicle dynamic driving and static building environment.

[0054] The ionospheric scintillation threat coefficient is calculated simultaneously based on the ionospheric scintillation probability index. After comparing the index with a preset threshold, normalization is performed, and invalid values ​​in the absence of interference are eliminated to obtain a threat coefficient that accurately reflects the impact of space ionospheric disturbances on GPS observations, thus achieving independent quantification of upper-altitude electromagnetic environment interference.

[0055] By multiplying and fusing the risk coefficient of the urban canyon model with the ionospheric scintillation threat coefficient, and coupling ground and space environmental interference through multiplication operations, a comprehensive observation unreliability that uniformly characterizes the reliability of GPS observation information is generated.

[0056] S300, based on the comprehensive observation unreliability, performs hierarchical decision scheduling to generate a unified control parameter vector containing pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command;

[0057] The quantitative index is compared with the preset two-level thresholds. Based on the comparison results, three operating modes adapted to different interference intensities are divided. Then, a unique pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command are matched for each mode. Under normal low interference conditions, the standard fusion mode is activated to maintain the basic parameter configuration and maintain stable fusion positioning efficiency. Under the condition of gradually increasing ground environment interference, the urban canyon transition mode is activated. The pseudorange observation noise variance scaling factor is dynamically generated according to the proportion of unreliability in the threshold range. The weight of Doppler velocity observation is increased simultaneously and the adaptive coupling mode is switched. Under the condition of strong interference in the space ionosphere, the ionospheric scintillation suppression mode is activated. The two types of observation control parameters are adjusted to the upper limit and the covariance freezing mode is switched. After completing the precise configuration of all parameters, the three types of core control parameters are integrated and encapsulated into a unified control parameter vector.

[0058] S400, based on the unified control parameter vector, dynamically adjust the observation information fusion strategy in the integrated navigation filter, apply the adjusted pseudorange observation noise variance scaling factor and Doppler velocity observation gain weight to the filter measurement update stage, and switch the coupling relationship between the filter state prediction and measurement update according to the filter coupling mode command, and output continuous positioning results.

[0059] By combining the pseudorange observation noise variance scaling factor with the original pseudorange observation noise variance matrix of the integrated navigation filter, the pseudorange observation fusion weights are dynamically adjusted. Simultaneously, the Doppler velocity observation gain weights are applied to the original Doppler velocity observation gain matrix, enhancing the Doppler velocity observation correction capability. Then, following the filter coupling mode instructions, the coupling relationship between filter state prediction and measurement update is switched. In normal coupling mode, the standard Kalman filtering process is followed to ensure coherent execution of state prediction and measurement update. In adaptive coupling mode, the adjusted pseudorange observation noise variance matrix is ​​used as the core basis for state update. In covariance freezing mode, the covariance recursion calculation related to pseudorange residuals is stopped. State prediction is performed based on the inertial navigation system error model and the adjusted Doppler velocity observations. Simultaneously, the pseudorange data that has passed quality inspection before covariance freezing is used to perform a refined estimate of the inertial navigation accelerometer zero bias. Finally, the refined calibrated inertial navigation parameters and state prediction results are combined to output continuous positioning results.

[0060] like Figure 2 As shown, the collection of multi-dimensional scene data during vehicle operation specifically includes:

[0061] S110: The longitudinal acceleration, lateral acceleration, and heading angle change rate of the vehicle are collected in real time through the inertial navigation system. The longitudinal acceleration change frequency is calculated based on the longitudinal acceleration, and the heading angle change rate amplitude is calculated based on the heading angle change rate. The longitudinal acceleration change frequency and the heading angle change rate amplitude are then subjected to time-domain statistical feature extraction to generate vehicle motion feature data.

[0062] The frequency of longitudinal acceleration change is used to characterize the frequency of vehicle acceleration and deceleration. It is calculated using the zero-crossing rate within a sliding window, and the formula is:

[0063] ;

[0064] in:

[0065] The frequency of longitudinal acceleration change calculated at time k is expressed in Hz.

[0066] The sampling frequency of the inertial navigation system is expressed in Hz.

[0067] The number of sampling points within the sliding time window, with a window duration of 0.5s to 2s, is used to adapt to the short-term changes in the vehicle's motion state.

[0068] This represents the mean-free longitudinal acceleration value at the i-th sampling time, in m / s². , This is the original longitudinal acceleration output by the inertial navigation system;

[0069] This is the sequence number of the current sampling time.

[0070] The rate of change of heading angle is used to characterize the severity of the vehicle's steering action. It is calculated using the root mean square magnitude within a sliding window, and the formula is:

[0071] ;

[0072] in:

[0073] The magnitude of the rate of change of heading angle calculated at time k, in rad / s;

[0074] The number of sampling points within the sliding time window is consistent with the window length used to calculate the frequency of longitudinal acceleration change.

[0075] The yaw rate is the rate of change of the heading angle (yaw rate) output by the inertial navigation system at the i-th sampling time, in rad / s.

[0076] This is the sequence number of the current sampling time.

[0077] Frequency sequence of longitudinal acceleration changes , Heading angle change rate amplitude sequence Extracting time-domain statistical features to generate vehicle motion feature data, the core calculation formula is as follows:

[0078] Mean characteristics: ;

[0079] Variance characteristics: ;

[0080] Peak factor characteristics: ;

[0081] The final generated vehicle motion feature data is a feature vector:

[0082] ;

[0083] in:

[0084] The extracted vehicle motion feature data is a 6-dimensional feature vector, which serves as the input for the subsequent urban canyon motion feature model.

[0085] , These are the mean values ​​of the frequency of longitudinal acceleration change and the amplitude of the rate of change of heading angle within the sliding window, respectively.

[0086] , These are the variances of the longitudinal acceleration change frequency and the heading angle change rate amplitude within the sliding window, respectively, reflecting the degree of dispersion of the sequence;

[0087] , These are the peak factors of the longitudinal acceleration change frequency and the heading angle change rate amplitude within the sliding window, respectively, reflecting the impact characteristics of the sequence.

[0088] S120: The road grade, roadside building density, and underpass space ratio information are obtained by matching the vehicle's current location with the high-precision vehicle map, and the road grade, roadside building density, and underpass space ratio information are used as environmental semantic information.

[0089] S130: By receiving the total electron content data of the ionosphere broadcast by the ionospheric grid model, the current UTC time and vehicle latitude and longitude coordinates are obtained. The current UTC time, the vehicle latitude and longitude coordinates and the total electron content data of the ionosphere are input into the ionospheric scintillation probability prediction model, and the ionospheric scintillation probability prediction model outputs the ionospheric scintillation probability index.

[0090] The ionospheric scintillation probability prediction model, based on current spatiotemporal parameters and ionospheric state parameters, quantitatively predicts the probability of ionospheric scintillation events occurring at the current vehicle location and time, and outputs an ionospheric scintillation probability index in the range of 0 to 1, providing a quantitative basis for subsequent assessment of the degree of interference with GPS observations and calculation of the ionospheric scintillation threat coefficient.

[0091] The logic chain of the model's input and output is divided into three layers, achieving a precise mapping from spatiotemporal-state parameters to flicker probability:

[0092] Spatiotemporal constraint layer: The model first converts UTC time to the local time of the vehicle's location, and then combines this with the geomagnetic latitude corresponding to the latitude and longitude to determine the basic prior probability of ionospheric scintillation at that spatiotemporal location. Ionospheric scintillation has significant spatiotemporal distribution characteristics: it is prevalent within 20° north and south of the magnetic equator, and the peak period is from 18:00 to 24:00 local time. This layer provides the basic spatiotemporal constraints for probabilistic forecasting.

[0093] State Correction Layer: Based on the input total ionospheric electron content (TEC) data, the model calculates the TEC level and spatiotemporal rate of change of TEC at the current location, and corrects the basic prior probability. TEC is a direct quantitative indicator of ionospheric activity level. Anomalous enhancement of TEC and dramatic gradient changes are precursors to irregular ionospheric structures. The greater the deviation of TEC from the climatological mean, the higher the correction magnitude for scintillation probability.

[0094] Probabilistic Quantization Output Layer: The model nonlinearly fuses the spatiotemporal prior probabilities with the TEC correction term, ultimately outputting an ionospheric scintillation probability index in the range of 0 to 1. The closer the index is to 1, the higher the risk of ionospheric scintillation occurring at the current time and location, leading to distortion of GPS pseudorange observations.

[0095] like Figure 3 As shown, the obtained comprehensive observational unreliability specifically includes:

[0096] S210, the longitudinal acceleration change frequency and the heading angle change rate amplitude are input into the pre-constructed urban canyon motion characteristic model, the urban canyon motion characteristic model outputs the motion characteristic risk score, the roadside building density is compared with the preset building density threshold and normalized to the building environment risk score, the motion characteristic risk score and the building environment risk score are weighted and summed to obtain the urban canyon mode risk coefficient;

[0097] A piecewise linear normalization method is used to map roadside building density to a risk score of 0-1, as shown in the formula:

[0098] ;

[0099] in:

[0100] The normalized building environment risk score ranges from [0,1]. A higher score indicates a higher risk of GPS signal obstruction and multipath effects caused by roadside buildings.

[0101] The roadside building density is matched to the vehicle's current location, with a value range of [0,1]. It is obtained from a high-precision map and represents the ratio of the building area to the total area of ​​the area within a preset range on both sides of the road.

[0102] The preset lower limit threshold for building density is usually set to 0.2. If the density is below this threshold, it is considered an open scene with no risk of being obstructed by urban canyons.

[0103] The preset upper limit threshold for building density is usually set to 0.8. If the density is higher than this threshold, it is considered a dense urban canyon scene, where the risk of occlusion is at its highest.

[0104] Weighted summation of risk coefficients for the urban canyon model:

[0105] ;

[0106] Constraints: , ,

[0107] in:

[0108] The risk coefficient for the urban canyon mode is [0,1]. The higher the coefficient, the higher the risk of GPS observation distortion in the urban canyon scenario.

[0109] The value is the motion feature risk score output by the urban canyon motion feature model, with a value range of [0,1]. The higher the score, the more the vehicle motion features conform to the driving characteristics of frequent acceleration, deceleration, and turning in the urban canyon.

[0110] This represents the normalized building environment risk score, with a value range of [0,1].

[0111] The weighting coefficient for the motion characteristic risk score is usually taken as 0.4 to 0.6, reflecting the degree to which vehicle motion characteristics contribute to the risk of urban canyons;

[0112] The weighting coefficient for the building environment risk score is usually taken as 0.4 to 0.6, reflecting the degree of contribution of the roadside building environment to the risk of urban canyons.

[0113] S220, compare the ionospheric scintillation probability index with a preset ionospheric scintillation threshold. When the ionospheric scintillation probability index ≤ the ionospheric scintillation threshold, set the ionospheric scintillation threat coefficient to zero. When the ionospheric scintillation probability index > the ionospheric scintillation threshold, normalize the ionospheric scintillation probability index and use the normalized ionospheric scintillation probability index as the ionospheric scintillation threat coefficient.

[0114] ;

[0115] in:

[0116] The normalized ionospheric scintillation threat coefficient has a value range of [0,1]. The higher the coefficient, the greater the interference threat of ionospheric scintillation to GPS observations.

[0117] The ionospheric scintillation probability index is the output of the ionospheric scintillation probability prediction model, with a value range of [0,1], representing the probability of ionospheric scintillation occurring at the current spatiotemporal location;

[0118] The preset ionospheric scintillation threshold is usually set to 0.3. If the threshold is lower than this, it is determined that there is no ionospheric scintillation threat and the threat coefficient is set to 0.

[0119] S230, perform a multiplication operation using the urban canyon model risk coefficient and the ionospheric scintillation threat coefficient, and use the result of the multiplication operation as the comprehensive observation unreliability.

[0120] In this step, the construction steps of the urban canyon motion feature model are as follows:

[0121] Several sets of vehicle calibration data are collected. Each set of calibration data includes the longitudinal acceleration change frequency, heading angle change rate amplitude, and corresponding scene type label obtained at the same sampling time. The scene type label is used to identify the scene in which the vehicle is located at that sampling time.

[0122] The longitudinal acceleration change frequency and the heading angle change rate amplitude are used as input features, and the scene type label is used as the output label. The support vector machine algorithm is used for training to construct an urban canyon motion feature model. The urban canyon motion feature model includes a classification decision function for mapping the input features to motion feature risk scores.

[0123] The urban canyon motion feature model is constructed using a binary classification SVM. Platt scaling maps the classification decision values ​​to motion feature risk scores ranging from 0 to 1. The complete classification decision function consists of two steps:

[0124] 1. Calculation of raw decision values ​​for SVM

[0125] ;

[0126] in:

[0127] The input feature vector corresponds to the original SVM decision value, which takes the real number field. Positive and negative values ​​represent the classification result, and the absolute value represents the classification confidence.

[0128] The input feature vector is the extracted vehicle motion feature data. ;

[0129] The number of support vectors obtained during training. Support vectors are samples in the training set that lie on the margin boundary of the classification hyperplane and are the core samples that determine the classification hyperplane.

[0130] The Lagrange multiplier corresponding to the i-th support vector is , which is the optimal parameter obtained by SVM training. The Lagrange multiplier corresponding to non-support vectors is 0, so only the support vectors need to be summed.

[0131] The scene type label corresponding to the i-th support vector is +1 (urban canyon scene) or -1 (non-urban canyon scene).

[0132] The bias term for the classification hyperplane is the optimal parameter obtained during SVM training, used to adjust the position of the classification hyperplane;

[0133] The kernel function is used to map low-dimensional features to a high-dimensional space to achieve non-linear classification. In this scenario, the radial basis function (RBF, Gaussian kernel) is used, and the formula is:

[0134] ;

[0135] in, This is the kernel function bandwidth parameter, a preset hyperparameter of SVM that controls the scope of the kernel function. This is the Euclidean distance between the input feature vector and the i-th support vector.

[0136] 2. Motion feature risk score mapping (Platt scaling):

[0137] ;

[0138] in:

[0139] The final output is the motion feature risk score, which ranges from [0,1]. A higher score indicates that the vehicle motion features are more consistent with the urban canyon scene.

[0140] , The fitting parameters scaled for Platt are obtained through cross-validation on the training set and are used to nonlinearly map the original SVM decision values ​​to probability values ​​in the [0,1] interval.

[0141] When calling the urban canyon motion feature model, the longitudinal acceleration change frequency and the heading angle change rate amplitude are input into the classification decision function, which calculates and outputs the motion feature risk score.

[0142] like Figure 4 As shown, the generation of a unified control parameter vector, which includes pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command, specifically includes:

[0143] S310, compare the overall observation unreliability with a preset first unreliability threshold and a second unreliability threshold, wherein the first unreliability threshold is less than the second unreliability threshold. When the overall observation unreliability is less than or equal to the first unreliability threshold, determine that the current state is in standard fusion mode, set the pseudorange observation noise variance scaling factor to 1.0, set the Doppler velocity observation gain weight to 1.0, and set the filter coupling mode command to normal coupling mode.

[0144] S320, when the first unreliability threshold < comprehensive observation unreliability ≤ second unreliability threshold, it is determined that the current state is in the urban canyon transition mode. The pseudorange observation noise variance scaling factor is dynamically calculated according to the ratio of the difference between the comprehensive observation unreliability and the first and second unreliability thresholds. The Doppler velocity observation gain weight is set to the preset enhancement weight value, and the filter coupling mode command is set to the adaptive coupling mode.

[0145] In the urban canyon transition mode, a linear interpolation method is used to dynamically calculate the scaling factor, and the formula is as follows:

[0146] ;

[0147] in:

[0148] The scaling factor for the pseudorange observation noise variance, calculated dynamically, has a range of values. The larger the coefficient, the higher the amplification of the pseudorange observation noise variance, and the lower the fusion weight of pseudorange observations in the filter;

[0149] The overall observational unreliability at the current moment, with a value range of [0,1], is obtained by multiplying the risk coefficient of the urban canyon model and the ionospheric scintillation threat coefficient.

[0150] The first unreliability threshold is a preset value, which is the boundary between the standard fusion mode and the urban canyon transition mode, and is usually set to 0.2.

[0151] The second unbelievability threshold is a preset value, which is the boundary between the urban canyon transition mode and the ionospheric scintillation suppression mode, and is usually set to 0.7.

[0152] This is the preset maximum scaling limit for pseudorange observation noise variance, usually set to 50, which corresponds to the minimum limit for pseudorange observation weights in high-uncertainty scenarios.

[0153] S330, when the overall observation unreliability is greater than the second unreliability threshold, it is determined that the current state is in ionospheric scintillation suppression mode, the pseudorange observation noise variance scaling factor is set to the preset maximum scaling upper limit value, the Doppler velocity observation gain weight is set to the preset maximum gain upper limit value, and the filter coupling mode command is set to covariance freezing mode.

[0154] S340, the pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode instruction are vectorized and encapsulated to generate the unified control parameter vector.

[0155] like Figure 5As shown, the output continuous positioning results specifically include:

[0156] S410, obtain the original pseudorange observation noise variance matrix of the integrated navigation filter, and perform a multiplication operation between the pseudorange observation noise variance scaling factor and the original pseudorange observation noise variance matrix to obtain the adjusted pseudorange observation noise variance matrix.

[0157] ;

[0158] in:

[0159] This is the adjusted pseudorange observation noise variance matrix, with dimension . , where n is the number of GPS satellites visible at the current moment, used for calculating the pseudorange observation weights in the Kalman filter measurement update process;

[0160] This is the pseudorange observation noise variance scaling factor, generated by the hierarchical decision scheduling module, with a value ≥ 1.0;

[0161] This is the original pseudorange observation noise variance matrix, with dimension 1. , is a diagonal matrix, where the diagonal elements are the original noise variance of the corresponding satellite pseudorange observations, calculated from parameters such as GPS receiver ranging accuracy and satellite elevation angle. The off-diagonal elements are 0, representing that the pseudorange observation noise of different satellites is independent of each other.

[0162] S420: Obtain the original Doppler velocity observation gain matrix of the integrated navigation filter, multiply the Doppler velocity observation gain weight with the original Doppler velocity observation gain matrix to obtain the adjusted Doppler velocity observation gain matrix.

[0163] ;

[0164] in:

[0165] The adjusted Doppler velocity observation gain matrix has a dimension of [missing information]. m is the state vector dimension of the combined navigation filter, and n is the number of GPS satellites visible at the current moment, which is used to map the Doppler velocity observation residuals into the state vector correction amount;

[0166] The weights for the Doppler velocity observation gain are generated by the hierarchical decision scheduling module and have a value ≥1.0. The larger the weight, the greater the contribution of the Doppler velocity observation to the filter state correction.

[0167] This is the original Doppler velocity observation gain matrix, with dimension 1. The value is calculated by the standard Kalman filter process based on the state prediction covariance matrix, the Doppler velocity observation matrix, and the original observation noise variance matrix, reflecting the original weights of the Doppler velocity observations for state correction.

[0168] S430, according to the filter coupling mode instruction, the corresponding filter coupling mode switching is executed. When the filter coupling mode instruction is normal coupling mode, state prediction and measurement update are executed sequentially according to the standard Kalman filter process. When the filter coupling mode instruction is adaptive coupling mode, the adjusted pseudorange observation noise variance matrix is ​​used as the fusion weight of pseudorange observation for state update in the measurement update stage.

[0169] S440, when the filter coupling mode instruction is covariance freeze mode, the recursive operation of the covariance update matrix of the pseudorange residual in the measurement update is suspended, and the state prediction is performed using the Doppler velocity observation and the error model of the inertial navigation system corresponding to the adjusted Doppler velocity observation gain matrix.

[0170] The covariance freezing mode is triggered when the overall observation unreliability exceeds the second threshold. At this point, GPS pseudorange observations are severely interfered with by urban canyons, multipath effects, or ionospheric scintillation. The pseudorange residuals are no longer zero-mean Gaussian white noise but contain a large number of systematic errors. If the covariance matrix is ​​recursively updated based on the distorted pseudorange residuals, the state covariance matrix estimation will be severely distorted, destroying the optimality of the Kalman filter and even causing filter divergence. Furthermore, the erroneous residuals will contaminate the estimation results of error parameters such as accelerometer and gyroscope zero bias in the inertial navigation system, leading to a rapid decline in the accuracy of inertial navigation position estimation.

[0171] S450: The accelerometer zero bias in the inertial navigation system is finely estimated using the last batch of pseudorange data that passed the preset quality inspection before covariance freezing. Based on the finely estimated accelerometer zero bias and the state prediction result, the continuous positioning result is output.

[0172] The GPS high-precision continuous positioning method based on inertial navigation also includes S500, a step for evaluating the effectiveness of hierarchical decision scheduling and correcting strategies, specifically including:

[0173] S510, during the execution of the observation information fusion strategy, the innovation sequence output by the combined navigation filter is acquired in real time, the Gaussian white noise hypothesis test is performed on the innovation sequence, and the autocorrelation coefficient and chi-square test statistic of the innovation sequence are calculated.

[0174] The innovation sequence is the residual sequence between the observed predicted value and the actual observed value in Kalman filtering. Zero-mean Gaussian white noise is a necessary and sufficient condition for Kalman filtering to achieve optimal estimation. By checking whether the innovation sequence meets the Gaussian white noise assumption, we can directly judge whether the current observation information fusion strategy (pseudorange noise scaling, Doppler gain adjustment, coupling mode switching) is reasonable and whether the efficiency of hierarchical decision scheduling meets the standard.

[0175] When the test fails, it indicates that the filter has deviated from the optimal state, triggering the subsequent scene recognition deviation detection and threshold dynamic adjustment process to achieve closed-loop adaptive optimization of the positioning system. The non-Gaussian white noise characteristic of the innovation sequence is an early signal of filter divergence. Real-time testing can trigger correction before the positioning result fails, avoiding large jumps in positioning.

[0176] S520, when the autocorrelation coefficient exceeds the preset correlation threshold, it is determined that the efficiency of the current hierarchical decision scheduling does not meet the Gaussian white noise assumption, triggering the scene recognition deviation detection mechanism, obtaining the urban canyon mode risk coefficient and the ionospheric scintillation threat coefficient corresponding to the comprehensive observation unreliability at the current moment, comparing the urban canyon mode risk coefficient and the ionospheric scintillation threat coefficient with the mean in the historical sliding window respectively, and identifying whether there is a deviation in the current scene recognition;

[0177] The changes in vehicle driving scenarios and ionospheric states are continuous and smooth, and the corresponding risk coefficients will not show irregular sudden changes. By comparing the current risk coefficient with the historical sliding window average, if the current risk coefficient shows a significant and irregular jump compared with the historical average, it indicates an anomaly in the scene recognition process (such as high-precision map matching errors, inertial navigation feature extraction anomalies, or ionospheric data anomalies), resulting in distorted risk coefficient calculations. By comparing the deviations of the risk coefficient of the urban canyon mode and the ionospheric scintillation threat coefficient with their respective historical averages, the source of the deviation can be accurately located as either an error in urban canyon scene recognition or an anomaly in ionospheric scintillation threat assessment, providing a clear direction for subsequent corrections.

[0178] S530, when a deviation in scene recognition is detected, a correction factor is calculated based on the statistical characteristics of the information sequence, and the first unreliability threshold and the second unreliability threshold used in the process of generating the unified control parameter vector are dynamically adjusted using the correction factor.

[0179] The theoretical covariance matrix of the new sequence: ;

[0180] Actual covariance matrix of the innovation sequence within the sliding window: ;

[0181] Covariance bias coefficient: ;

[0182] Autocorrelation deviation coefficient: ;

[0183] Chi-square test deviation coefficient:

[0184] Calculate the overall deviation: ;

[0185] Calculate the correction factor:

[0186] Threshold dynamic adjustment: ;

[0187] in:

[0188] The calculated correction factor has a range of values. It is used to dynamically adjust the first and second unreliability thresholds to correct the deviation between scene recognition and hierarchical decision-making;

[0189] The overall deviation of the innovation sequence reflects the degree to which the innovation sequence deviates from the Gaussian white noise assumption. When η≥1, it indicates that there is a significant deviation and threshold correction is required.

[0190] η_S is the covariance bias coefficient, which reflects the degree of deviation between the actual covariance and the theoretical covariance of the innovation sequence. η_S>1 indicates that the actual observation noise is greater than the filter setting value, and the observation unreliability is underestimated.

[0191] η_ρ is the autocorrelation bias coefficient, which reflects the degree of deviation between the temporal correlation of the innovation sequence and the white noise assumption. η_ρ>1 indicates that the innovation sequence has a significant correlation and the filter model does not match the actual system.

[0192] η_χ is the chi-square deviation coefficient, which reflects the degree of deviation between the Gaussian distribution characteristics of the innovation sequence and the theoretical assumptions. η_χ>1 indicates that there is a significant systematic error in the innovation sequence.

[0193] Let be the information sequence (observation residual vector) at the i-th sampling time, with a dimension of n×1;

[0194] The Kalman filter observation matrix maps the state vector to the observation vector;

[0195] The state prediction covariance matrix is ​​obtained from the Kalman filter time update stage;

[0196] The observation noise variance matrix includes the noise variance of pseudorange and Doppler velocity observations;

[0197] The number of sampling points for the sliding window of the information sequence is usually 50 to 200, balancing statistical stability and real-time performance;

[0198] The maximum absolute autocorrelation coefficient of the innovation sequence within the sliding window;

[0199] The preset autocorrelation coefficient threshold is usually set to 0.2, and the autocorrelation coefficient of the white noise sequence should be close to 0.

[0200] Let be the chi-square test statistic at the i-th sampling time, which follows a chi-square distribution with n degrees of freedom;

[0201] This is the mean of the chi-square test statistic within the sliding window; the theoretical value is equal to the dimension n of the observation vector.

[0202] The dimension of the observation vector is equal to the total number of pseudoranges and Doppler observations of visible GPS satellites at the current moment;

[0203] To correct the lower limit of the factor, it is usually set to 0.5 to avoid the threshold being lowered excessively;

[0204] To correct the upper limit of the factor, it is usually set to 2.0 to avoid the threshold being excessively increased;

[0205] , These are the first unreliability thresholds before and after adjustment, respectively.

[0206] , These are the second unreliability thresholds before and after adjustment, respectively.

[0207] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0208] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0209] 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, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A GPS high-precision continuous positioning method based on inertial navigation, characterized in that, The method includes: Collect multi-dimensional scene data during vehicle operation. The multi-dimensional scene data includes vehicle motion feature data output by the inertial navigation system, environmental semantic information, and ionospheric scintillation probability index. The risk coefficient of the urban canyon pattern is calculated based on the vehicle motion feature data and the environmental semantic information, and the ionospheric scintillation threat coefficient is calculated based on the ionospheric scintillation probability index. The risk coefficient of the urban canyon pattern and the ionospheric scintillation threat coefficient are multiplied and fused to obtain the comprehensive observation unreliability. Based on the comprehensive observation unreliability, a hierarchical decision scheduling is performed to generate a unified control parameter vector containing pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command. Based on the unified control parameter vector, the observation information fusion strategy in the integrated navigation filter is dynamically adjusted. The adjusted pseudorange observation noise variance scaling factor and Doppler velocity observation gain weight are applied to the filter measurement update stage. The coupling relationship between the filter state prediction and measurement update is switched according to the filter coupling mode command, and continuous positioning results are output.

2. The method according to claim 1, characterized in that, The collection of multi-dimensional scene data during vehicle operation specifically includes: The vehicle's longitudinal acceleration, lateral acceleration, and heading angle change rate are collected in real time by an inertial navigation system. The longitudinal acceleration change frequency is calculated based on the longitudinal acceleration, and the heading angle change rate amplitude is calculated based on the heading angle change rate. The longitudinal acceleration change frequency and the heading angle change rate amplitude are then subjected to time-domain statistical feature extraction to generate vehicle motion feature data. By matching the vehicle's current location with a high-precision vehicle map, information on road grade, roadside building density, and the proportion of space under viaducts is obtained, and the information on road grade, roadside building density, and the proportion of space under viaducts is used as environmental semantic information. By receiving the total electron content data of the ionosphere broadcast by the ionospheric grid model, the current UTC time and vehicle latitude and longitude coordinates are obtained. The current UTC time, the vehicle latitude and longitude coordinates and the total electron content data of the ionosphere are input into the ionospheric scintillation probability prediction model, and the ionospheric scintillation probability prediction model outputs the ionospheric scintillation probability index.

3. The method according to claim 2, characterized in that, The obtained comprehensive observational unreliability specifically includes: The longitudinal acceleration change frequency and the heading angle change rate amplitude are input into a pre-constructed urban canyon motion characteristic model. The urban canyon motion characteristic model outputs a motion characteristic risk score. The roadside building density is compared with a preset building density threshold and normalized to a building environment risk score. The motion characteristic risk score and the building environment risk score are weighted and summed to obtain the urban canyon model risk coefficient. The ionospheric scintillation probability index is compared with a preset ionospheric scintillation threshold. When the ionospheric scintillation probability index is less than or equal to the ionospheric scintillation threshold, the ionospheric scintillation threat coefficient is set to zero. When the ionospheric scintillation probability index is greater than or equal to the ionospheric scintillation threshold, the ionospheric scintillation probability index is normalized, and the normalized ionospheric scintillation probability index is used as the ionospheric scintillation threat coefficient. The risk coefficient of the urban canyon model and the ionospheric scintillation threat coefficient are multiplied together, and the result of the multiplication is used as the overall observational unreliability.

4. The method according to claim 3, characterized in that, The steps for constructing the urban canyon motion characteristic model are as follows: Several sets of vehicle calibration data are collected. Each set of calibration data includes the longitudinal acceleration change frequency, heading angle change rate amplitude, and corresponding scene type label obtained at the same sampling time. The scene type label is used to identify the scene in which the vehicle is located at that sampling time. The longitudinal acceleration change frequency and the heading angle change rate amplitude are used as input features, and the scene type label is used as the output label. The support vector machine algorithm is used for training to construct an urban canyon motion feature model. The urban canyon motion feature model includes a classification decision function for mapping the input features to motion feature risk scores. When calling the urban canyon motion feature model, the longitudinal acceleration change frequency and the heading angle change rate amplitude are input into the classification decision function, which calculates and outputs the motion feature risk score.

5. The method according to claim 4, characterized in that, The generation of a unified control parameter vector, which includes pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command, specifically includes: The overall observation unreliability is compared with a preset first unreliability threshold and a second unreliability threshold, wherein the first unreliability threshold is less than the second unreliability threshold. When the overall observation unreliability is less than or equal to the first unreliability threshold, it is determined that the current mode is the standard fusion mode, the pseudorange observation noise variance scaling factor is set to 1.0, the Doppler velocity observation gain weight is set to 1.0, and the filter coupling mode command is set to normal coupling mode. When the first unreliability threshold < the comprehensive observation unreliability ≤ the second unreliability threshold, it is determined that the current state is in the urban canyon transition mode. The pseudorange observation noise variance scaling factor is dynamically calculated based on the ratio of the difference between the comprehensive observation unreliability and the first and second unreliability thresholds. The Doppler velocity observation gain weight is set to a preset enhancement weight value, and the filter coupling mode command is set to the adaptive coupling mode. When the overall observation unreliability exceeds the second unreliability threshold, it is determined that the current state is in ionospheric scintillation suppression mode. The pseudorange observation noise variance scaling factor is set to the preset maximum scaling limit value, the Doppler velocity observation gain weight is set to the preset maximum gain limit value, and the filter coupling mode command is set to covariance freeze mode. The pseudorange observation noise variance scaling factor, Doppler velocity observation gain weight, and filter coupling mode command are vectorized and encapsulated to generate the unified control parameter vector.

6. The method according to claim 5, characterized in that, The output continuous positioning results specifically include: Obtain the original pseudorange observation noise variance matrix of the integrated navigation filter, and multiply the pseudorange observation noise variance scaling factor with the original pseudorange observation noise variance matrix to obtain the adjusted pseudorange observation noise variance matrix. Obtain the original Doppler velocity observation gain matrix of the integrated navigation filter, and multiply the Doppler velocity observation gain weight with the original Doppler velocity observation gain matrix to obtain the adjusted Doppler velocity observation gain matrix. According to the filter coupling mode instruction, the corresponding filter coupling mode switching is executed. When the filter coupling mode instruction is normal coupling mode, the state prediction and measurement update are executed sequentially according to the standard Kalman filter process. When the filter coupling mode instruction is adaptive coupling mode, the adjusted pseudorange observation noise variance matrix is ​​used as the fusion weight of pseudorange observation for state update in the measurement update stage. When the filter coupling mode instruction is covariance freeze mode, the recursive operation of the covariance update matrix of the pseudorange residual in the measurement update is suspended, and the state prediction is performed using the Doppler velocity observation and the error model of the inertial navigation system corresponding to the adjusted Doppler velocity observation gain matrix. The accelerometer zero bias in the inertial navigation system is refined by using the last batch of pseudorange data that has passed the preset quality inspection before covariance freezing. The continuous positioning result is output based on the refined accelerometer zero bias and the state prediction result.

7. The method according to claim 6, characterized in that, The method further includes steps for performance evaluation and strategy correction of hierarchical decision scheduling, specifically including: During the execution of the observation information fusion strategy, the innovation sequence output by the integrated navigation filter is acquired in real time, the Gaussian white noise hypothesis is tested on the innovation sequence, and the autocorrelation coefficient and chi-square test statistic of the innovation sequence are calculated. When the autocorrelation coefficient exceeds the preset correlation threshold, it is determined that the efficiency of the current hierarchical decision scheduling does not meet the Gaussian white noise assumption, triggering the scene recognition deviation detection mechanism, obtaining the urban canyon mode risk coefficient and the ionospheric scintillation threat coefficient corresponding to the comprehensive observation unreliability at the current moment, comparing the urban canyon mode risk coefficient and the ionospheric scintillation threat coefficient with the mean in the historical sliding window respectively, and identifying whether there is a deviation in the current scene recognition. When a deviation in scene recognition is identified, a correction factor is calculated based on the statistical characteristics of the information sequence, and the first and second unreliability thresholds used in generating the unified control parameter vector are dynamically adjusted using the correction factor.

Images