A Confidence-Guided Approach to Detecting and Eliminating State Domain Abrupt Faults in GNSS / INS

Abstract

This invention discloses a method for detecting and eliminating abrupt changes in the state domain of GNSS / INS based on confidence guidance. The method comprises the following steps: Step 1: Constructing a GNSS / INS compact combination filtering model; Step 2: Detecting and eliminating abrupt changes in the state domain of GNSS / INS compact combination navigation based on confidence guidance; Step 3: Using the filtering result corresponding to the final fault-free satellite group as the output result of the current epoch of GNSS / INS compact combination navigation. Compared with existing technologies, this invention constructs detection statistics based on state estimation differences and detects position, velocity, and attitude components separately. A truncation mechanism is introduced during the construction of the generalized inverse matrix of the covariance matrix to improve the numerical stability of the detection statistics. In the fault elimination stage, a normalized chi-square score is introduced for confidence ranking, avoiding the problem of traditional random search methods relying entirely on random sampling to construct the initial satellite group. Through confidence-ordered expansion and confidence backtracking confirmation mechanisms, false eliminations can be reduced and accuracy improved under multiple fault conditions.

Description

Technical Field

[0001] This invention relates to the field of satellite navigation and positioning integrity monitoring, and in particular to a method for detecting and eliminating GNSS / INS state domain abrupt change faults based on confidence guidance. Background Technology

[0002] GNSS / INS integrated navigation fully leverages the complementary advantages of the long-term stability of GNSS (Global Navigation Satellite System) and the short-term continuous autonomy of INS (Inertial Navigation System), ensuring the continuity of the navigation system. However, in actual observations, GNSS observations are easily affected by multipath effects, occlusion, and other factors, leading to abrupt errors in the integrated navigation solution results. When abnormal observations are not identified and eliminated in a timely manner, fault information may enter the filtering update process, further affecting subsequent state predictions and navigation solution results, thereby reducing the reliability of the integrated navigation system's positioning results. Therefore, it is necessary to study a fault detection and exclusion (FDE) method suitable for GNSS / INS integrated navigation systems in complex environments. This method should not only ensure timely fault detection but also achieve accurate fault location and effective elimination, thereby improving the integrity monitoring capability of the integrated navigation system in complex environments. This is of great value for enhancing the reliability and safe operation of navigation systems in applications such as transportation and aerospace.

[0003] Existing GNSS / INS integrated navigation fault detection and elimination methods mainly include Multiple Hypothesis Solution Separation (MHSS) and Random Search Method (RSM). MHSS-based fault detection and elimination typically requires constructing main filters, sub-filters, and secondary sub-filters for different fault hypotheses, and performing filtering and consistency checks on multiple measurement subsets separately. While this method can achieve fault detection and elimination, the number of hypothesis combinations to be constructed and tested increases rapidly with the number of visible satellites and potential faults, leading to a significant computational burden in multi-fault scenarios. Traditional random search method fault detection and elimination constructs an initial fault-free satellite group by randomly selecting some satellites, and then further eliminates detected and faulty satellites from this group. Although this method can reduce the number of candidate combinations to some extent, the initial satellite group is generated randomly. If it contains a faulty satellite, initialization will fail, requiring multiple re-initializations and reducing fault elimination efficiency.

[0004] This invention provides a confidence-guided method for detecting and eliminating sudden faults in the state domain of GNSS / INS integrated navigation. First, it utilizes state domain information to construct a fractional fault detection statistic, enabling fault detection at the position, velocity, and attitude states respectively, thereby improving the sensitivity of sudden fault detection. Then, in the fault elimination phase, all visible satellites are ranked according to their confidence based on the normalized chi-square scores of satellite observations. A confidence-ordered expansion and confidence backtracking confirmation mechanism is combined to determine the set of faulty satellites, reducing computational complexity and improving elimination efficiency and accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide a confidence-guided method for detecting and eliminating GNSS / INS state domain mutation faults, in order to solve the problems of high computational complexity and low initialization efficiency of existing fault elimination methods in multi-fault scenarios.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for detecting and eliminating state domain abrupt changes in GNSS / INS tightly integrated navigation is proposed, and the method steps are as follows. Step 1: Construct a GNSS / INS tightly integrated filter model. Based on the integrated navigation state estimation results of the previous epoch, use IMU measurement information to perform INS mechanical orchestration, and recursively obtain the INS predicted navigation state of the current epoch's position, velocity, and attitude. Based on the INS predicted navigation state, subtract the actual GNSS observations from the predicted observations to construct the error state equation and the GNSS / INS tightly integrated measurement equation. Kalman filtering is used to perform one-step prediction of the state and measurement update for the current epoch, resulting in an estimate of the prior state for the current epoch. Prior state covariance matrix Posterior state estimation Posterior state covariance matrix It is used to characterize the current epoch integrated navigation state and its uncertainties, and serves as the basic input for subsequent state domain abrupt fault detection and fault elimination; This invention constructs a GNSS / INS tightly integrated filtering model. The model uses the navigation state obtained through recursive INS mechanical orchestration as prior information and constructs corresponding predicted observations based on the prior position and velocity. Subsequently, the difference between the actual GNSS observations and the predicted observations is calculated to establish a tightly integrated measurement equation, and the integrated navigation state is updated in real time using Kalman filtering. During the filtering process, the model can jointly estimate state variables such as position error, velocity error, attitude error, gyroscope bias error, and accelerometer bias error, and outputs the prior state estimate, prior state covariance, posterior state estimate, and posterior state covariance at each epoch. The filtering results are not only used to realize GNSS / INS tightly integrated navigation solutions, but also provide necessary state variables and their covariance information for subsequent state domain abrupt fault detection, satellite confidence ranking, and fault backtracking and elimination.

[0007] Step 2: Detect and eliminate abrupt changes in the state domain of GNSS / INS tightly integrated navigation based on confidence guidance. The method and steps are as follows: Step 2.1: Obtain the state estimation difference and its covariance; To improve the sensitivity of abrupt fault detection, this invention does not directly use measurement domain information as the sole detection object. Instead, it constructs a detection statistic based on the state estimation difference between the Kalman filter prior state estimate and the posterior state estimate, defining the state estimation difference of the current epoch as: ; Under the valid null hypothesis Down, Mathematical expectation and covariance matrix They are respectively: ; ; in, Indicates the dimension of the state domain; Indicates length is The zero-moment vector; Indicates Kalman gain; Represents the measurement prediction covariance matrix; Step 2.2, for the covariance matrix Perform truncation; To reflect the integrity of the position, velocity, and attitude components in the integrated navigation system separately, the state estimation difference is... Divided into positional components velocity components and attitude components And construct the corresponding covariance matrix respectively. , and Singular value decomposition is performed on its covariance matrix to construct a truncated generalized inverse matrix. , ; Taking the positional components as an example: their covariance matrix Perform singular value decomposition: ; Where U and V represent unitary matrices, respectively; The singular value diagonal matrix is ​​represented by the following form: ; in, , This represents the dimension of the state domain.

[0008] According to relevant mathematical theories, matrices Moore-Penrose generalized inverse matrix The construction method is as follows: when singular values When, take its reciprocal; when singular values When, the corresponding position is 0, where, To avoid the problem of abnormally amplifying the detection statistic by directly taking the reciprocal of the smallest singular values ​​in the approximate singular matrix, the truncation threshold is defined as:

[0009] in, Represents the maximum singular value. This represents the relative cutoff coefficient, which can be selected. .

[0010] When constructing the generalized inverse matrix, if a singular value is greater than or equal to the truncation threshold, its reciprocal is taken; if a singular value is less than the truncation threshold or equal to 0, the corresponding position is set to 0. The processing method is as follows: ; Assumption If all singular values ​​of the matrix are greater than or equal to the truncation threshold, then the above processing method can be used to obtain... truncated generalized inverse matrix As shown below: ; Therefore, construct truncated generalized inverse matrix for: ; Step 2.3: Construct the covariance of the detection statistics for different state components, with the following formula: ; ; ; in, For positional components; For velocity components; For attitude components; Step 2.4: Calculate the chi-square detection threshold using the given false alarm rate, as shown in the following formula; ; in, Indicates the first The detection threshold for each epoch; The state component to be detected can be: Position components, velocity component or Attitude components; Indicates the false alarm rate; Indicates the confidence level. Let represent the chi-square cumulative distribution function with 3 degrees of freedom; This indicates that the condition is met. The minimum value; Step 2.5, using steps 2.3 and 2.4 to... epochs Perform a fault detection. If no fault is detected, the process ends; if a fault is detected, proceed to step 2.6. Since this invention detects three three-dimensional state components—position, velocity, and attitude—the detection dimension for each component is fixed at 3, and the detection threshold does not fluctuate with the number of visible satellites. If the detection statistic of any component exceeds its detection threshold, a sudden fault is determined to exist in the current epoch; if the detection statistics of all three components do not exceed the threshold, no fault is determined to exist in the current epoch.

[0011] Step 2.6: Perform confidence sorting; Step 2.7: Select an initial candidate satellite group and perform fault detection on it; calculate the normalized chi-square score of all visible satellites and construct a provisional fault-free satellite group. ; Step 2.8, will The remaining satellites, excluding those already in the array, are then subjected to confidence-ordered expansion in sequence. The temporarily fault-free satellite group The remainder besides The satellites are sorted from low to high according to the normalized chi-square scores obtained in step 2.6 to form a sequence of satellites to be expanded; Step 2.9, Construct a small-scale candidate set And perform confidence backtracking confirmation; When a sudden fault is detected during the confidence-ordered expansion process in step 2.8, the suspicious satellite is not directly identified as a faulty satellite, but is confirmed by confidence backtracking. Step 3: Repeat steps 2.8 and 2.9 until completion. After confidence-ordered expansion and confidence backtracking confirmation of all remaining satellites, the final output is the posterior state estimate, covariance matrix, and set of satellites judged as faulty for the current fault-free satellite group. The filtering result corresponding to the final fault-free satellite group is used as the output result of the current epoch GNSS / INS compact navigation.

[0012] Preferably, in step 1, the method for constructing the GNSS / INS compact combination filter model is as follows: Step 1.1: Obtain the raw GNSS observation information, IMU measurement information, and the combined navigation state estimation results of the previous epoch for the current epoch; the raw GNSS observations are obtained from the GNSS global navigation satellite system, and the IMU measurement information is obtained from the inertial measurement unit (IMU) of the INS inertial navigation system; Step 1.2: Based on the combined navigation state estimation results of the previous epoch, INS mechanical orchestration is performed using IMU measurement information to recursively obtain the INS predicted navigation state of the current epoch. The INS predicted navigation state includes the position, velocity and attitude information of the vehicle. Step 1.3: Based on the INS predicted navigation state, the difference between the actual GNSS observations and the predicted observations is calculated to construct an error state model of GNSS / INS tight combination filtering; The state equation formula for the error state model is as follows: ; in, express The state vector of each epoch. Represents the state transition matrix. Represents white noise in the system process; Step 1.4: Calculate the GNSS prediction observations corresponding to the current epoch based on the position, velocity and other information obtained from INS prediction; Taking pseudorange observation as an example: the predicted pseudorange is calculated based on the INS predicted position and the satellite position, and then the difference is made with the actual GNSS pseudorange observation to construct a compact combination measurement vector. The formula is as follows: ; in, This represents the actual GNSS observations; This represents the GNSS predicted observations calculated based on the INS predicted navigation status.

[0013] If a differential positioning model is used, the difference between the double-difference pseudorange observation and the virtual double-difference pseudorange observation predicted by INS can be used to construct a double-difference measurement vector; if carrier phase or Doppler observation is used, the corresponding measurement equations can also be constructed in the same way.

[0014] The GNSS / INS compact combination measurement equation for the error state model is as follows: ; in, express Measurement vectors for each epoch; Represents the measurement matrix; Indicates measurement noise; Step 1.5: After establishing the above state equation and measurement equation, Kalman filtering is used to complete the one-step prediction of the state and the update of the measurement for the current epoch, thus obtaining the epoch prior state estimate for the current epoch. Prior state covariance matrix Posterior state estimation Posterior state covariance matrix Filtering results.

[0015] The posterior state estimation results are used to correct the INS predicted navigation state, while the aforementioned prior and posterior state information can serve as the basic input for subsequent state domain abrupt change fault detection and troubleshooting.

[0016] Preferably, in step 1.1, the raw GNSS observation information mainly includes pseudorange observations, and carrier phase, Doppler, and other observations can be further introduced according to actual application requirements; the IMU measurement information includes angular velocity, specific force, or their integral form output by gyroscopes and accelerometers.

[0017] Preferably, the state vector of the current epoch is based on a 15-dimensional error state. It can be represented as: ; in, , indicating position error; , indicating speed error; , indicating attitude error; This represents the zero bias error of the gyroscope; , representing the zero bias error of the accelerometer; e, n, and u represent the directions in the northeast-northeast navigation coordinate system, respectively; , , These represent the directions in the carrier coordinate system, and the above error states are used to describe the deviation of the navigation state obtained by INS recursion from the actual navigation state.

[0018] Preferably, in step 1.5, the prior state estimate of the current epoch filter is assumed to be... The prior covariance matrix is The measurement vector is The measured predicted value is The measurement prediction covariance matrix is Then the Kalman gain and the posterior state estimate can be expressed as: ; ; ; This yields the posterior state estimate for the current epoch. and posterior covariance matrix This provides a foundation for subsequent state domain mutation fault detection.

[0019] Preferably, in step 2.2, for each three-dimensional state component, singular value decomposition and truncated generalized inverse are used to construct the detection statistic. If the covariance matrix conditions are good, the truncated generalized inverse is consistent with the standard Moore-Penrose generalized inverse. If the covariance matrix is ​​approximately singular, the truncation operation can set the inverse of singular values ​​less than the threshold to zero, thereby preventing the detection statistic from being abnormally amplified.

[0020] Preferably, the confidence sorting method in step 2.6 is as follows: Once a sudden fault is detected in the current epoch, the Kalman filter innovation is calculated. and the new information covariance matrix They are respectively: ; ; in, Indicates the first The Kalman filter innovation vector for each epoch is used to reflect the difference between the actual measurement and the predicted measurement. Represents the measurement noise covariance matrix; Then, the normalized chi-square score is defined for each component of the Kalman filter innovation as follows: ; in, This represents the i-th component of the measurement innovation vector; This represents the i-th diagonal element of the measurement prediction covariance matrix. The smaller the value, the better the consistency between the corresponding measurement and the current filter prediction, and the less prone it is to failure; while The larger the value, the worse the consistency between the corresponding measurement and the current filter prediction, and the more prone it is to failure; Finally, the normalized chi-square scores corresponding to all visible satellites are sorted in ascending order to obtain the satellite confidence ranking results.

[0021] Preferably, in step 2.7, the fault detection method is as follows: Let the number of all visible satellites at the current epoch be . Based on the satellite confidence ranking results obtained in step 2.6, the satellites with lower normalized chi-square scores are selected. The initial candidate satellite group consists of 10 satellites. , Then, to Perform a Kalman filter update, and use the methods in steps 2.3 and 2.4 to perform fault detection on the updated state results. If no fault is detected, the initial candidate satellite group is recorded as a provisionally fault-free satellite group. If a fault is detected, the previous satellites will be reselected based on the satellite confidence ranking results obtained in step 2.6. These satellites form a new initial candidate satellite group, and abrupt fault detection is performed again; Repeat the above process until no faults are detected in the selected initial candidate satellite group, and then designate it as the provisionally fault-free satellite group. .

[0022] This step replaces the purely random initialization in the traditional random search method with confidence sorting, which can reduce the probability of faulty satellites being included in the initial candidate satellite group, thereby reducing the invalid computation caused by repeated initialization.

[0023] As a preferred option, in step 2.9, the confidence-ordered expansion method is as follows: According to the sorting results, one satellite is selected from the sequence of satellites to be expanded and temporarily added to the current provisional fault-free satellite group. After each satellite is added, a Kalman filter update is performed, and the abrupt fault detection methods in steps 2.3 and 2.4 are used to check the updated state. If no fault is detected, the satellite is retained. If a fault is detected, the satellite is marked as a suspicious satellite and the confidence backtracking confirmation step is initiated.

[0024] As a preferred option, the specific method for confidence backtracking confirmation in step 2.9 is as follows: The current provisionally fault-free satellite group Construct a small candidate set based on Candidate set The suspicious satellites marked in step 2.8, and The higher score in the normalized card square pattern Composed of 10 satellites ( (2 or 3 can be selected). Subsequently, the candidate set Each satellite in the process is sequentially eliminated based on hypotheses. After each satellite is eliminated, the remaining satellites are used to perform a new Kalman filter update, and the abrupt fault detection methods in steps 2.3 and 2.4 are used to detect the updated state results. If no fault is detected under a certain elimination scheme, and the normalized chi-square score corresponding to the elimination scheme is the lowest, then the elimination scheme is taken as the result of this fault elimination, and the eliminated satellite is confirmed as a faulty satellite.

[0025] The normalized chi-square score is calculated as follows: for the candidate set After eliminating a particular satellite, the normalized chi-square scores of the remaining satellites are sorted from lowest to highest, and then calculated. The average of the normalized chi-square scores.

[0026] Preferably, in step 3, after completing the fault-free satellite group... After the confidence-ordered expansion and confidence backtracking confirmation of all remaining satellites, the final fault-free satellite group is re-filtered and updated to obtain the corresponding posterior state estimate and posterior state covariance matrix. The filtering result is then used as the GNSS / INS compact navigation output for the current epoch. If a fault alarm is triggered again in subsequent epochs, the above steps are repeated to achieve sudden fault detection and elimination in continuous epochs.

[0027] Compared with the prior art, the advantages of the present invention are as follows: (1) Based on the state estimation difference, a detection statistic is constructed, and the position, velocity and attitude components are detected separately, so that the detection dimension is fixed at 3, which can reduce the impact of detection threshold fluctuation on the sensitivity of sudden fault detection. (2) In the covariance matrix The introduction of a truncation mechanism during the construction of the generalized inverse matrix can suppress statistical anomalies caused by the amplification of the inverse of the minimum singular value in the approximate singular matrix, thereby improving the numerical stability of the detection statistics. (3) Introducing normalized chi-square scores for confidence ranking during the troubleshooting stage can avoid the problem that the traditional random search method relies entirely on random sampling to construct the initial satellite group. (4) Through the confidence-ordered expansion and confidence backtracking confirmation mechanism, false elimination can be reduced and accuracy can be improved under multiple fault conditions; Fifth, the method does not require the construction of a large number of sub-filters and secondary sub-filters as the multi-hypothesis solution separation method, and the overall computational complexity is lower, making it more suitable for real-time GNSS / INS tightly integrated navigation integrity monitoring applications. Attached Figure Description

[0028] Figure 1 A general design flowchart provided for embodiments of the present invention; Figure 2 A flowchart for detecting abrupt changes in the state domain of tightly coupled GNSS / INS navigation with a truncation mechanism, provided in an embodiment of the present invention; Figure 3 A flowchart for detecting and eliminating sudden changes in the state domain of GNSS / INS tightly coupled navigation based on confidence guidance, provided for embodiments of the present invention. Detailed Implementation

[0029] The present invention will be further described below: a GNSS / INS state domain abrupt change fault detection and elimination method based on confidence guidance, such as... Figure 1 As shown, the method steps are as follows: Step 1: Establish a GNSS / INS compact combination filter model and obtain the current epoch state estimation results; Acquire the raw GNSS observation information, inertial measurement unit (IMU) measurement information, and the combined navigation state estimation results from the previous epoch for the current epoch.

[0030] Based on the combined navigation state estimation results of the previous epoch, INS mechanical orchestration is performed using IMU measurement information to recursively obtain the INS predicted navigation state of the current epoch, including position, velocity, and attitude information.

[0031] Subsequently, GNSS predicted observations are calculated based on the INS predicted navigation state, and the difference between the actual GNSS observations and the predicted observations is used to construct a GNSS / INS tightly integrated measurement model. Based on this, Kalman filtering is employed to complete the one-step state prediction and measurement update for the current epoch, obtaining the prior state estimate, prior state covariance, posterior state estimate, and posterior state covariance for the current epoch. These filtering results are used to characterize the current epoch integrated navigation state and its uncertainties, and serve as the basic input for subsequent state domain abrupt change fault detection and troubleshooting.

[0032] Let the prior state estimate of the current epoch filter be... The prior covariance matrix is The measurement vector is The measured predicted value is The measurement prediction covariance matrix is Then the Kalman gain and the posterior state estimate can be expressed as: ; ; ; This yields the posterior state estimate for the current epoch. and posterior covariance matrix This provides a foundation for subsequent state domain mutation fault detection.

[0033] Step 2: This invention proposes a confidence-guided method for detecting and eliminating GNSS / INS state domain abrupt changes in faults. The method steps are as follows: (1) Truncate the covariance matrix and perform a split-quantization chi-square test in the state domain; such as Figure 2 As shown, After completing the filtering update, the difference between the prior state estimate and the posterior state estimate is calculated. And calculate its covariance matrix. .Will The states are divided into three three-dimensional components: position, velocity, and attitude, and the corresponding covariance sub-matrices are extracted for each.

[0034] For each three-dimensional state component, singular value decomposition and truncated generalized inverse are used to construct the detection statistic. If the covariance matrix is ​​well-conditional, the truncated generalized inverse is consistent with the standard Moore-Penrose generalized inverse; if the covariance matrix is ​​approximately singular, the truncation operation can set the inverse of singular values ​​less than a threshold to zero, thereby preventing the detection statistic from being abnormally amplified.

[0035] In this embodiment, the false alarm rate can be set to Since the three detection components—position, velocity, and attitude—all have a dimension of 3, the chi-square thresholds corresponding to these three components remain consistent and do not change with the number of visible satellites. If , or If any statistic exceeds the threshold, the current epoch enters the fault troubleshooting process; otherwise, the fault-free GNSS / INS tightly coupled navigation result for the current epoch is output.

[0036] (2) Calculate the normalized chi-square scores of all visible satellites and construct a provisional fault-free satellite group. ; When a fault is detected in the current epoch, the measurement information is used. Covariance Matrix Calculate the normalized chi-square score .according to The satellites are sorted from lowest to highest score, and the top-scoring satellites are selected. The initial candidate satellite group consists of 10 satellites. If there is a faulty satellite in the initial candidate satellite group, then... The candidate groups are reduced and reconstructed to improve the success rate of constructing the initial fault-free satellite group. Once no faulty satellites remain in the initial candidate group, it is designated as a provisional fault-free satellite group. .

[0037] (3) The remaining satellites, excluding those already in the array, are then subjected to confidence-ordered expansion in sequence. like Figure 3 As shown, the remaining satellites are then added sequentially in ascending order of their normalized chi-square scores. After each satellite is added, the Kalman filter is updated again, and abrupt fault detection is performed again on the position, velocity, and attitude components. When a satellite is added, if... If no fault is detected, it means that the satellite is in harmony with the current... The other satellites are quite consistent, so they should be retained. In the middle; if If a fault is detected, the confidence backtracking confirmation step must be initiated.

[0038] (4) Construct a small-scale candidate set And perform confidence backtracking confirmation; When a satellite triggers an alarm, a candidate set is constructed. ,in For the suspicious satellites marked in the previous step, to for The higher score in the normalized card square pattern One satellite. For the candidate set. Each satellite in the dataset undergoes a hypothesis elimination process to obtain... There are 10 candidate exclusion schemes. For each exclusion scheme, the filtering is updated again and fault detection is performed. If an exclusion scheme can make the position, velocity, and attitude detection statistics all below the detection threshold, then the exclusion scheme is considered valid. If there are multiple valid schemes, the scheme with the lowest average normalized chi-square score is selected as the final exclusion scheme.

[0039] By using the aforementioned confidence-based local backtracking confirmation, the problem of erroneous removal caused by directly removing newly added satellites when triggering alarms, which is common in traditional random search methods, is avoided. This method is particularly suitable for scenarios with multiple faulty satellites.

[0040] Step 3: Output the final result; After completion After confirming the confidence order expansion and confidence backtracking of all remaining satellites, the final fault-free satellite group and the set of satellites judged as faulty are determined for the current epoch. The final fault-free satellite group is then filtered and updated again to obtain the corresponding posterior state estimate and posterior state covariance matrix. This filtering result is used as the GNSS / INS compact navigation output for the current epoch. If a fault alarm is triggered again in a subsequent epoch, the above steps are repeated to achieve abrupt fault detection and elimination across consecutive epochs.

[0041] "Analysis and Comparison of Computational Complexity with Traditional Methods": (1) Traditional random search methods require starting from the initialization phase. Randomly selected from 100 satellites These are selected as the initial fault-free satellite group. In the case of a single fault, if the selected... If the faulty satellite is included in the initialization subset, it means that the current initialization subset does not meet the fault-free condition. Therefore, the initialization should be judged as a failure, and random initialization should be performed again. Thus, the probability of initialization failure can be expressed as: ; Similarly, in the case of a double fault, if the extracted If any one of the two faulty satellites is among the initialization subsets, then the current initialization subset still does not meet the fault-free condition. Therefore, the initialization should be judged as a failure and re-initialized. The probability of initialization failure can be expressed as: ; Therefore, we can conclude that: ; For traditional RSM, as the number of faulty satellites increases, the randomly selected initial subset is more likely to include faulty satellites, which increases the likelihood of repeated initialization and leads to more invalid iterations and additional computations.

[0042] In contrast, this invention can sort satellites using normalized chi-square scores and prioritize the selection of satellites with lower scores to form the initial candidate group, thereby reducing the possibility of faulty satellites in the initial candidate group and reducing the invalid calculations caused by repeated initialization.

[0043] (2) The multiple hypothesis separation method (MHSS) requires the construction of a main filter, multiple sub-filters, and secondary sub-filters, with approximately [number] detections per epoch. Magnitude.

[0044] The number of single-epoch detections in this invention mainly consists of initial candidate group detection, ordered expansion detection of remaining satellites, and a small number of local backtracking confirmation detections, which can be expressed as: ; in, This indicates the initial number of candidate satellites when the fault detection is passed for the first time. This indicates the number of times a local backtracking confirmation is triggered during the confidence-ordered expansion process. This represents the number of highly suspicious satellites selected from the current fault-free satellite group within the backtracking candidate set. Because... The constants are typically small, thus the present invention has lower overall complexity compared to MHSS. The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed.

[0045] The foregoing has provided a detailed description of the GNSS / INS navigation state domain abrupt change fault detection and elimination method based on confidence guidance provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Modifications and improvements to the present invention are possible without exceeding the concept and scope specified in the appended claims. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for detecting and eliminating GNSS / INS state domain abrupt changes based on confidence guidance, characterized in that: The steps are as follows: Step 1: Construct a GNSS / INS tightly integrated filter model. Based on the integrated navigation state estimation results of the previous epoch, use IMU measurement information to perform INS mechanical orchestration, and recursively obtain the INS predicted navigation state of the current epoch's position, velocity, and attitude. Based on the INS predicted navigation state, subtract the actual GNSS observations from the predicted observations to construct the error state equation and the GNSS / INS tightly integrated measurement equation. Kalman filtering is used to perform one-step prediction of the state and measurement update for the current epoch, resulting in an estimate of the prior state for the current epoch. Prior state covariance matrix Posterior state estimation Posterior state covariance matrix It is used to characterize the current epoch integrated navigation state and its uncertainties, and serves as the basic input for subsequent state domain abrupt fault detection and fault elimination; Step 2: Detect and eliminate abrupt changes in the state domain of GNSS / INS tightly integrated navigation based on confidence guidance. The method and steps are as follows: Step 2.1: Obtain the state estimation difference and its covariance; A detection statistic is constructed based on the state estimation difference between the prior and posterior state estimates obtained through Kalman filtering. The state estimation difference for the current epoch is defined as follows: ； Under the valid null hypothesis Down, Mathematical expectation and covariance matrix They are respectively: ； ； in, Indicates the dimension of the state domain; Indicates length is The zero-moment vector; Indicates Kalman gain; Represents the measurement prediction covariance matrix; Step 2.2, for the covariance matrix Perform truncation; The difference in state estimation Divided into positional components velocity components and attitude components And construct the corresponding covariance matrix respectively. , and Singular value decomposition is performed on its covariance matrix to construct a truncated generalized inverse matrix. , ; Step 2.3: Construct the covariance of the detection statistics for different state components; Step 2.4: Calculate the chi-square detection threshold using the given false alarm rate; Step 2.5, using steps 2.3 and 2.4 to... epochs Perform a fault detection. If no fault is detected, the process ends; if a fault is detected, proceed to step 2.

6. Step 2.6: Perform confidence sorting; Step 2.7: Select an initial candidate satellite group and perform fault detection on it; calculate the normalized chi-square score of all visible satellites and construct a provisional fault-free satellite group. ; Step 2.8, will The remaining satellites, excluding those already in the system, are then subjected to confidence-ordered expansion in sequence. The temporarily fault-free satellite group The remainder besides The satellites are sorted from low to high according to the normalized chi-square scores obtained in step 2.6 to form a sequence of satellites to be expanded; Step 2.9, Construct a small-scale candidate set And perform confidence backtracking confirmation; When a mutation fault is detected during the confidence-ordered expansion process in step 2.8, a confidence backtracking confirmation is performed. Step 3: Repeat steps 2.8 and 2.9 until completion. After confidence-ordered expansion and confidence backtracking confirmation of all remaining satellites, the final output is the posterior state estimate, covariance matrix, and set of satellites judged as faulty for the current fault-free satellite group. The filtering result corresponding to the final fault-free satellite group is used as the output result of the current epoch GNSS / INS compact navigation.

2. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 1, characterized in that: In step 1, the method for constructing the GNSS / INS compact combination filter model is as follows: Step 1.1: Obtain the raw GNSS observation information, IMU measurement information, and combined navigation state estimation results of the previous epoch for the current epoch; Step 1.2: Based on the combined navigation state estimation results of the previous epoch, INS mechanical orchestration is performed using IMU measurement information to recursively obtain the INS predicted navigation state of the current epoch. The INS predicted navigation state includes the position, velocity and attitude information of the vehicle. Step 1.3: Based on the INS predicted navigation state, the difference between the actual GNSS observations and the predicted observations is calculated to construct an error state model of GNSS / INS tight combination filtering; The state equation formula for the error state model is as follows: ； in, express The state vector of each epoch. Represents the state transition matrix. Represents white noise in the system process; Step 1.4: Calculate the GNSS prediction observations corresponding to the current epoch based on the position, velocity and other information obtained from INS prediction; The GNSS / INS compact combination measurement equation formula for the error state model is as follows: ； in, express Measurement vectors for each epoch; Represents the measurement matrix; Indicates measurement noise; Step 1.5: After establishing the above state equation and measurement equation, Kalman filtering is used to complete the one-step prediction of the state and the update of the measurement for the current epoch, thus obtaining the epoch prior state estimate for the current epoch. Prior state covariance matrix Posterior state estimation Posterior state covariance matrix Filtering results.

3. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 2, characterized in that: In step 1.1, the raw GNSS observation information mainly includes pseudorange observations, and carrier phase, Doppler, and other observations can be further introduced according to actual application requirements; the IMU measurement information includes angular velocity, specific force, or their integral form output by gyroscopes and accelerometers.

4. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 2, characterized in that: In step 1.3, the state vector of the current epoch It can be represented as: ； in, , indicating position error; , indicating speed error; , indicating attitude error; This represents the zero bias error of the gyroscope; , representing the zero bias error of the accelerometer; e, n, and u represent the directions in the northeast-northeast navigation coordinate system, respectively; , , These represent the directions in the carrier coordinate system, and the above error states are used to describe the deviation of the navigation state obtained by INS recursion from the actual navigation state.

5. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 2, characterized in that: In step 1.5, let the prior state estimate of the current epoch filter be... The prior covariance matrix is The measurement vector is The measured predicted value is The measurement prediction covariance matrix is Then the Kalman gain and the posterior state estimate can be expressed as: ； ； ； This yields the posterior state estimate for the current epoch. and posterior covariance matrix This provides a foundation for subsequent state domain mutation fault detection.

6. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 1, characterized in that: In step 2.2, for each three-dimensional state component, singular value decomposition and truncated generalized inverse are used to construct the detection statistic. If the covariance matrix conditions are good, the truncated generalized inverse is consistent with the standard Moore-Penrose generalized inverse. If the covariance matrix is ​​approximately singular, the truncation operation can set the reciprocal of singular values ​​smaller than the threshold to zero, thereby preventing the detection statistic from being abnormally amplified.

7. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 1, characterized in that: In step 2.6, the confidence sorting method is as follows: Once a sudden fault is detected in the current epoch, the Kalman filter innovation is calculated. and the new information covariance matrix They are respectively: ； ； in, Indicates the first The Kalman filter innovation vector for each epoch is used to reflect the difference between the actual measurement and the predicted measurement. Represents the measurement noise covariance matrix; Then, the normalized chi-square score is defined for each component of the Kalman filter innovation as follows: ； in, This represents the i-th component of the measurement innovation vector; This represents the i-th diagonal element of the measurement prediction covariance matrix; The smaller the value, the better the consistency between the corresponding measurement and the current filter prediction, and the less prone it is to failure; while The larger the value, the worse the consistency between the corresponding measurement and the current filter prediction, and the more prone it is to failure; Finally, the normalized chi-square scores corresponding to all visible satellites are sorted in ascending order to obtain the satellite confidence ranking results.

8. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 7, characterized in that: In step 2.7, the fault detection method is as follows: Let the number of all visible satellites at the current epoch be . Based on the satellite confidence ranking results obtained in step 2.6, the satellites with lower normalized chi-square scores are selected. The initial candidate satellite group consists of 10 satellites. , Then, to Perform a Kalman filter update, and use the methods in steps 2.3 and 2.4 to perform fault detection on the updated state results. If no fault is detected, the initial candidate satellite group is recorded as a provisionally fault-free satellite group. ; If a fault is detected, the previous satellites will be reselected based on the satellite confidence ranking results obtained in step 2.

6. These satellites form a new initial candidate satellite group, and abrupt fault detection is performed again; Repeat the above process until no faults are detected in the selected initial candidate satellite group, and then designate it as the provisionally fault-free satellite group. .

9. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 8, characterized in that: In step 2.9, the confidence-ordered expansion method is as follows: According to the sorting results, one satellite is selected from the satellite sequence to be expanded and temporarily added to the current provisional fault-free satellite group. After each satellite is added, a Kalman filter update is performed, and the abrupt fault detection methods in steps 2.3 and 2.4 are used to check the updated state. If no fault is detected, the satellite is retained. middle; If a fault is detected, the satellite is marked as a suspicious satellite and the confidence backtracking confirmation step is initiated.

10. The method for detecting and eliminating sudden changes in the state domain of GNSS / INS navigation based on confidence guidance according to claim 9, characterized in that: In step 2.9, the specific method for confidence backtracking confirmation is as follows: The current provisionally fault-free satellite group Construct a small candidate set based on Candidate set The suspicious satellites marked in step 2.8, and The higher score in the normalized card square pattern Composed of 10 satellites ( (2 or 3 can be selected). Subsequently, the candidate set Each satellite in the process is sequentially eliminated based on hypotheses. After each satellite is eliminated, the remaining satellites are used to perform a new Kalman filter update, and the abrupt fault detection methods in steps 2.3 and 2.4 are used to detect the updated state results. If no fault is detected under a certain elimination scheme, and the normalized chi-square score corresponding to the elimination scheme is the lowest, then the elimination scheme is taken as the result of this fault elimination, and the eliminated satellite is confirmed as a faulty satellite.

Images