A method for autonomous integrity monitoring for a satellite navigation receiver

Abstract

The application discloses a kind of autonomous integrity monitoring methods for satellite navigation receiver, convolutional neural network is used in the present application and the slope weighted least square RAIM combined algorithm of optimization.The combined algorithm includes: first, the characteristic slope of satellite is calculated to select satellite, after eliminating the satellite with the largest characteristic slope, when the number of satellites is greater than or equal to 5, the optimized slope weighted least square RAIM algorithm is used, and the corresponding weight matrix is constructed based on the RAIM algorithm based on least square residual;When the number of satellites is less than 5, data regression learning prediction is carried out using convolutional neural network, to determine whether there is a faulty satellite.The present application has the function of more accurate monitoring whether satellite signal is faulty, to achieve fault monitoring and troubleshooting, so as to realize the autonomous integrity monitoring of satellite navigation receiver.

Description

Technical Field

[0001] This invention belongs to the field of autonomous integrity monitoring technology for satellite navigation system receivers. It is applicable to satellite receivers autonomously verifying the consistency of positioning results through redundant observations in various environments, realizing fault detection, and ensuring navigation and positioning accuracy. It mainly relates to an autonomous integrity monitoring method for satellite navigation receivers. Background Technology

[0002] With increasingly stringent safety requirements for autonomous driving, navigation systems on unmanned vehicles are exhibiting a trend towards greater informatization, intelligence, and integration. Accuracy and reliability have become crucial characteristics of autonomous driving navigation systems. Precise positioning and attitude determination, as well as fault detection and isolation, have become important disciplines and technologies in the field of autonomous driving, receiving increasing attention and experiencing rapid development in recent years.

[0003] Global Navigation Satellite Systems (GNSS) can provide three-dimensional position and velocity information to all parts of the world, 24 / 7. Most vehicles are equipped with GNSS. However, due to the extremely high safety requirements of autonomous driving, whether GNSS is used as an auxiliary or primary navigation system, in addition to navigation accuracy, it must also meet integrity requirements.

[0004] System integrity refers to the ability to issue alarms to users when the system is unavailable, in addition to providing positioning, navigation, and timing functions. Receiver Autonomous Integrity Monitoring (RAIM) verifies the consistency of positioning results based on redundant observations from the user receiver, enabling fault detection and ensuring navigation and positioning accuracy.

[0005] RAIM needs to address two issues: whether satellites are faulty and which satellite has the fault. Currently, there are three main methods for receiver autonomous integrity monitoring: range comparison, verification, and least squares residual. It can be proven that these three methods are mathematically equivalent. Due to limitations in receiver memory capacity and computing speed, high demands are placed on the complexity, computational load, and reliability of the RAIM algorithm. Summary of the Invention

[0006] This invention addresses the shortcomings of existing least-squares residual RAIM algorithms, which do not consider the impact of satellite geometry on positioning and require more than five satellites for calculation. Furthermore, these algorithms suffer from low fault detection rates when satellite pseudorange deviations are small, leading to missed alarms. This invention provides an autonomous integrity monitoring method for satellite navigation receivers.

[0007] This method employs a convolutional neural network combined with an optimized slope-weighted least squares (RAIM) algorithm. The optimized RAIM algorithm can monitor and eliminate faulty satellites when there are at least five satellites, while the convolutional neural network can more accurately detect faults when there are fewer than five satellites. This enables the satellite receiver to autonomously monitor its integrity in various environments. The method of this invention has a lower false alarm rate and a lower missed alarm rate, and higher computational efficiency.

[0008] To achieve the above objectives, the technical solution adopted in this invention is as follows: a combined algorithm of convolutional neural network and optimized slope-weighted least squares (RAIM) is used. The combined algorithm includes: first, selecting satellites by calculating their characteristic slopes, and then eliminating satellites with the largest characteristic slopes. When the number of satellites is greater than or equal to 5, the optimized slope-weighted least squares (RAIM) algorithm is used to construct a corresponding weighting matrix based on the least squares residual RAIM algorithm. When the number of satellites is less than 5, a convolutional neural network is used for data regression learning and prediction to determine whether there are any faulty satellites.

[0009] As an improvement to the present invention, a method for autonomous integrity monitoring of satellite navigation receivers specifically includes the following steps:

[0010] S1. Establishing a pseudorange observation model: The satellite pseudorange observation model can be represented as follows:

[0011] Y = GX + ε

[0012] Where Y is an n-dimensional vector representing the difference between the observed pseudorange and the approximate calculated pseudorange, where n is the number of observable satellites; G is an n×4-dimensional coefficient matrix; X is a 4-dimensional vector of undetermined parameters, including three user position correction parameters and one receiver clock offset correction parameter; ε is an n-dimensional observation pseudorange noise vector, which is denoted as ε+b if there is a bias.

[0013] S2. Satellite Selection Based on Characteristic Slope: Considering the influence of satellite characteristic slope, satellites with larger characteristic slopes contribute more to positioning accuracy. However, when the same fault deviation exists, the test statistic is smaller, indicating a higher risk of missed alarms. While missing alarms must be strictly avoided, false alarms can be tolerated to some extent. Preliminary satellite selection is achieved by calculating the characteristic slope of each satellite, filtering out the satellite with the highest value, and eliminating that satellite.

[0014] S3. When the number of remaining satellites is greater than or equal to 5, the fault is monitored using the characteristic slope-weighted RAIM algorithm: A weight matrix is ​​constructed, and the characteristic slope of the satellites is used for weighting. Satellites with larger slopes are assigned the smallest possible weights to reduce their positioning errors, lower the probability of missed alarms, and improve the reliability of the algorithm. Observational information is assigned to weighting factors to obtain a new pseudorange residual vector. Detection limit σ rDue to σ r It depends only on the given false alarm probability, therefore σ r It can be given in advance, and will be calculated in real time during navigation. With σ r In comparison, if This indicates that a fault has been detected, an alert is issued to the user, and the faulty satellite is removed by the Balda data detection method.

[0015] S4. When the number of remaining satellites is less than 5, a convolutional neural network is used to monitor for faults: A dataset is established using data from 4 satellites per group and corresponding receiver data, and training and testing sets are created. After data normalization, the data is flattened, and a network structure is constructed. Parameters such as the number of training iterations, batch size, and learning rate are set. Model training and prediction are performed, and the network is saved for subsequent monitoring when the success rate reaches 100% after adjusting the parameters.

[0016] As an improvement of the present invention, the formula for calculating the satellite feature slope in step S2 is as follows:

[0017]

[0018] In the formula, SLOPE i Let K be the characteristic slope of the i-th satellite, and K = (G T G) -1 G T F = IG(G T G) -1 G T

[0019] As another improvement of the present invention, in step S3, a weight diagonal matrix is ​​constructed:

[0020]

[0021] Based on the least squares principle, the least squares solution for the user state can be calculated:

[0022]

[0023] The pseudorange residual vector is:

[0024]

[0025] make It is called a mapping matrix.

[0026] The vector γ contains satellite ranging error information and can be used as a basis for determining whether there are faulty satellites. The posterior unit weight mean square error of the synthesized pseudorange residual vector is:

[0027]

[0028] In the formula, n represents the current number of satellites.

[0029] As another improvement of the present invention, in step S4, when the number of satellites is less than 5 and the traditional RAIM algorithm based on least squares residuals cannot be used, a convolutional neural network is used for satellite integrity monitoring, which has high accuracy and is easy to use.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) Satellite selection is based on satellite characteristic slope. The selection rule is to eliminate satellites that contribute little to accuracy, i.e., satellites with large characteristic slopes. Satellite characteristic slope affects the positioning accuracy of satellites. When there are enough visible satellites, satellites can be screened by characteristic slope to improve the spatial distribution of satellites. Moreover, this method can avoid a large amount of computation and improve computational efficiency.

[0032] (2) The weight matrix is ​​constructed using the satellite feature slope. The larger the satellite feature slope, the greater the positioning error caused by the satellite and the greater the probability of missing alarms. Therefore, the characteristics of the satellite feature slope are used for weighting. Satellites with larger slopes are assigned smaller weights to reduce their positioning error, reduce the probability of missing alarms, improve the reliability of the algorithm, and ensure the safety of unmanned vehicles in operation.

[0033] (3) A convolutional neural network is used to learn and predict faulty satellites. When the number of satellites is equal to 4, the traditional least squares RAIM algorithm cannot monitor faulty satellites. This method uses the data from the semi-physical experiment to train the convolutional neural network. Through this neural network, not only can faults be detected under the premise of limited data, but it is also more intelligent, improves the accuracy of fault detection, and significantly reduces the false alarm rate. Attached Figure Description

[0034] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0035] Figure 2 The parameter settings for the convolutional neural network in the test examples of this invention;

[0036] Figure 3 This is a training progress graph of the convolutional neural network in the test examples of this invention;

[0037] Figure 4 This refers to the prediction results of the convolutional neural network training set in the test examples of this invention;

[0038] Figure 5 These are the prediction results of the convolutional neural network test set in the test examples of this invention;

[0039] Figure 6This is a graph showing the relationship between single-satellite fault detection rate and pseudorange deviation in the test examples of this invention.

[0040] Figure 7 This is a graph showing the relationship between multi-satellite fault detection rate and pseudorange deviation in the test examples of this invention. Detailed Implementation

[0041] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0042] Example 1: An autonomous integrity monitoring method for satellite navigation receivers, such as... Figure 1 As shown, the specific steps include the following:

[0043] S1. Establishing a pseudorange observation model: The satellite pseudorange observation model can be represented as follows:

[0044] Y = GX + ε

[0045] Where Y is an n-dimensional vector representing the difference between the observed pseudorange and the approximate calculated pseudorange, where n is the number of observable satellites; G is an n×4-dimensional coefficient matrix; X is a 4-dimensional vector of undetermined parameters, including three user position correction parameters and one receiver clock offset correction parameter; ε is an n-dimensional observation pseudorange noise vector, which is denoted as ε+b if there is a bias.

[0046] S2. Satellite Selection Based on Characteristic Slope: Considering the influence of satellite characteristic slope, satellites with larger characteristic slopes contribute more to positioning accuracy. However, when the same fault deviation exists, the test statistic is smaller, indicating a higher risk of missed alarms. While strictly avoiding missed alarms, false alarms can be tolerated to some extent.

[0047] The formula for calculating satellite characteristic slope is as follows:

[0048]

[0049] In the formula, SLOPE i Let K be the characteristic slope of the i-th satellite, and K = (G T G) -1 G T F = IG(G T G) -1 G T

[0050] By calculating the characteristic slope of each satellite, the maximum value is filtered out and the satellite is removed, thus achieving rapid satellite selection.

[0051] S3. When the number of satellites is greater than or equal to 5, fault monitoring is performed using the feature slope-weighted RAIM algorithm: A weight matrix is ​​constructed, and the characteristic slope of the satellites is used for weighting. Satellites with larger slopes are assigned smaller weights to reduce their positioning errors, lower the probability of missed alarms, and improve the reliability of the algorithm. A weight diagonal matrix is ​​constructed as follows:

[0052]

[0053] The observed information is assigned a weighting factor, denoted as Y = GX + ε, W

[0054] Based on the least squares principle, the least squares solution for the user state can be calculated:

[0055]

[0056] The pseudorange residual vector is:

[0057]

[0058] Let Q = IG(G) T WG) -1 G T W is called the mapping matrix.

[0059] The vector γ contains satellite ranging error information and can be used as a basis for determining whether there are faulty satellites. The posterior unit weight mean square error of the synthesized pseudorange residual vector is:

[0060]

[0061] In the formula, n is the current number of satellites, and γ is denoted as γ. T Wγ is SSE.

[0062] Sure Detection limit σ r Due to σ r It depends only on the given false alarm probability, therefore σ r This can be given in advance. If the components of the pseudorange error vector ε are mutually independent normally distributed random errors, then the mean is 0 and the variance is... In the absence of pseudorange faults, the system should be in normal detection mode; any detection alarms that occur are false alarms. Therefore, given the false alarm probability P... FA The following probability equation should hold:

[0063]

[0064] The above formula determines... If the detection limit T is , then Detection limit:

[0065]

[0066] During navigation calculation, the calculation will be performed in real time. With σ r In comparison, if This indicates a fault has been detected, an alert is issued to the user, and the faulty satellite is removed using the Balda data detection method. The basic idea is to construct a statistic based on the least squares residual vector. This statistic follows a certain distribution, and given a confidence level, the presence of a gross error in a residual can be statistically determined by testing the statistic. From the relationship between residuals and observation errors, the statistic can be defined as:

[0067]

[0068] Calculate the statistics d for each satellite. i The satellite corresponding to the maximum value is the faulty satellite, which will trigger an alarm and be removed.

[0069] S4. When the number of satellites is less than 5, a convolutional neural network is used to monitor faults: Each group consists of data from 4 satellites and corresponding receiver data to create a dataset, which is then divided into training and testing sets. The input is 4*4+3=19 data points. The output is 2 when a faulty satellite is detected, and 1 when no fault is detected. Data classification and prediction are performed. After data normalization, the data is flattened, and the network structure is constructed. Parameters such as the number of training iterations, batch size, and learning rate are set. Model training and prediction are performed. When the success rate reaches 100%, the network is saved for subsequent monitoring.

[0070] Test case

[0071] An autonomous integrity monitoring method for satellite navigation receivers is presented. In this test case, satellite reception data was collected by the satellite receiver in an open area, and data processing was implemented using MATLAB software. First, the raw data is verified by performing the following operations to prove that there are no faulty satellites. Then, a random error is added to the pseudorange of one or more satellites, and the following operations are performed again to check whether the experimental results can trigger an alarm and remove a faulty satellite when one is detected.

[0072] S1. Establishing a pseudorange observation model: The satellite pseudorange observation model can be represented as follows:

[0073] Y = GX + ε

[0074] Where Y is an n-dimensional vector representing the difference between the observed pseudorange and the approximate calculated pseudorange, where n is the number of observable satellites; G is an n×4-dimensional coefficient matrix; X is a 4-dimensional vector of undetermined parameters, including three user position correction parameters and one receiver clock offset correction parameter; and ε is an n-dimensional observation pseudorange noise vector.

[0075] S2. Satellite selection based on satellite characteristic slope: using the formula for calculating satellite characteristic slope:

[0076]

[0077] In the formula, SLOPE i Let K be the characteristic slope of the i-th satellite, and K = (G T G) -1 G T F = IG(G T G) -1 G T

[0078] By calculating the characteristic slope of each satellite, the maximum value is filtered out and the satellite is removed, thus achieving rapid satellite selection.

[0079] S3. When the number of remaining satellites is greater than or equal to 5, the fault is monitored by the characteristic slope-weighted RAIM algorithm: Construct a weighted diagonal matrix:

[0080]

[0081] Sure Detection limit σ r Due to σ r It depends only on the given false alarm probability, therefore σ r The parameters can be given in advance, and subsequent tests are performed in groups of 5 satellites, i.e., n=5. If the components of the pseudorange error vector ε are mutually independent normally distributed random errors, then the mean is 0 and the variance is... Other In the absence of pseudorange faults, the system should be in normal detection mode; any detection alarms that occur are false alarms. Therefore, given the false alarm probability P... FA =0.001, the following probability equation should hold:

[0082]

[0083] The above formula determines... The detection limit T is determined using the function T in MATLAB: T 2 =chi2inv(1-0.001,1);

[0084] but Detection limit:

[0085]

[0086] Taking the first 5 satellites, and applying the least squares principle, the least squares solution for the user state can be calculated:

[0087]

[0088] The pseudorange residual vector is:

[0089]

[0090] The vector γ contains satellite ranging error information and can be used as a basis for determining whether there are faulty satellites. The posterior unit weight mean square error of the synthesized pseudorange residual vector is:

[0091]

[0092] In the formula, n is the current number of satellites, and γ is denoted as γ. T Wγ is SSE.

[0093] After solving, the calculation will be performed in real time. With σ r In comparison, if This indicates a fault has been detected, an alert is issued to the user, and the faulty satellite is removed using the Balda data detection method. The statistics are:

[0094]

[0095] Calculate the statistics d for each satellite. i The satellite corresponding to the maximum value is the faulty satellite, which is flagged and removed from the list. A new satellite's data is then added, and the S3 feature slope weighted RAIM algorithm monitoring method is repeated until all satellites in the group have participated in the inspection. If... If no fault is detected, the first of the selected 5 satellites is removed, and data from the next satellite is added. The S3 feature slope weighted RAIM algorithm monitoring method is repeated until all satellites in the group have participated in the inspection.

[0096] S4. When the number of satellites is less than 5, a convolutional neural network is used to monitor faults: Each group consists of data from 4 satellites and corresponding receiver data to create a dataset, which is then divided into a training set and a test set. The input is 4*4+3=19 data points. The output is 2 when there is a faulty satellite and 1 when there is no fault. Data classification and prediction are performed. After data normalization, the data is flattened, and the network structure is constructed as follows: Figure 2 As shown, the model includes a 2x1 kernel for generating convolutions, a normalization layer, a ReLU activation layer, a pooling layer, a fully connected layer, a loss function layer, and a classification layer. Parameters are set to 500 training iterations, a batch size of 128, and a learning rate of 0.001. Model training and prediction are performed as follows. Figure 3 When the parameters are adjusted to achieve a 100% success rate, such as... Figure 4 Figure 5 Save the network data for future monitoring.

[0097] After verifying that the original data showed no satellite faults, artificial biases of varying magnitudes were added to the pseudoranges of individual satellites. Monitoring was then performed using the method described above, and the fault detection rate was calculated for each set of biases. Figure 6It can be observed that even lower pseudorange bias can result in a higher fault detection rate, demonstrating stronger detection performance compared to the traditional least-squares bias RAIM algorithm. By introducing artificial biases of varying magnitudes into the pseudoranges of multiple satellites and monitoring them using the method described above, the fault detection rate is calculated for each set of biases. Figure 7 It can be found that the method of the present invention also has high detection performance for multi-star faults.

[0098] In summary, the method of this invention is applicable to satellite navigation receivers to achieve autonomous integrity monitoring in various environments. It can monitor single-satellite and multi-satellite faults, and can also perform integrity monitoring even when signal obstruction is severe. This method is more accurate, with lower false alarm and false alarm rates. It solves problems such as the inability of satellite receivers to monitor and alarm when the pseudorange error of the faulty satellite is small, the inability to monitor only 4 satellites, and the occurrence of multiple faulty satellites.

[0099] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A method for autonomous integrity monitoring of satellite navigation receivers, characterized in that, Specifically, the steps include the following: S1. Establish a satellite pseudorange observation model. S2. Satellite Feature Slope Selection: Considering the influence of satellite feature slope, calculate the feature slope of each satellite, filter for the maximum value and remove the satellite, achieving preliminary and rapid satellite selection. S3. When the number of satellites is greater than or equal to 5, faults are monitored using the characteristic slope-weighted RAIM algorithm: a weight matrix is ​​constructed, and the characteristics of the satellite characteristic slopes are used for weighting to improve the reliability of the algorithm. Observation information is assigned to weighting factors to obtain a new pseudorange residual vector and determine... Detection limit σ r Due to σ r It depends only on the given false alarm probability, therefore σ r It can be given in advance, and will be calculated in real time during navigation. With σ r In comparison, if This indicates that a fault has been detected, an alert is issued to the user, and the faulty satellite is removed using the Balda data detection method. S4. When the number of satellites is less than 5, the convolutional neural network is used to monitor the fault: a dataset is established by each group of 4 satellite data and corresponding receiver data, and the training set and test set are divided. After the data is normalized, the data is flattened and the network structure is constructed. At the same time, the training times, batch size and learning rate parameters are set to train the model and make model predictions. When the parameters are adjusted to achieve a success rate of 100%, the network is saved for subsequent monitoring.

2. The autonomous integrity monitoring method for satellite navigation receivers as described in claim 1, characterized in that: The satellite pseudorange observation model for step S1 is represented as follows: Y = GX + ε Where Y is an n-dimensional vector representing the difference between the observed pseudorange and the approximate calculated pseudorange, where n is the number of observable satellites; G is an n×4-dimensional coefficient matrix; X is a 4-dimensional vector of undetermined parameters, including three user position correction parameters and one receiver clock offset correction parameter; ε is an n-dimensional observation pseudorange noise vector, which is denoted as ε+b if there is a bias.

3. The autonomous integrity monitoring method for satellite navigation receivers as described in claim 2, characterized in that: The formula for calculating the satellite feature slope in step S2 is as follows: In the formula, SLOPE i Let K be the characteristic slope of the i-th satellite, and K = (G T G) -1 G T F = IG(G T G) -1 G T .

4. The autonomous integrity monitoring method for satellite navigation receivers as described in claim 2, characterized in that: In step S3, a weight diagonal matrix is ​​constructed:

5. The autonomous integrity monitoring method for satellite navigation receivers as described in claim 4, characterized in that: In step S3, the feature slope-weighted RAIM algorithm, based on the least squares principle, can calculate the least squares solution for the user state: The pseudorange residual vector is: Let S = IG(G) T WG) -1 G T W is called the mapping matrix. The vector γ contains satellite ranging error information and can be used as a basis for determining whether there is a faulty satellite. The post-hoc unit weighted mean square error of the combined pseudorange residual vector is: In the formula, n is the current number of satellites. Sure Detection limit σ r Due to σ r It depends only on the given false alarm probability, therefore σ r It can be given in advance that if the components of the pseudorange error vector ε are mutually independent normally distributed random errors, then the mean is 0 and the variance is... In the absence of pseudorange faults, the system should be in normal detection mode. If a detection alarm occurs, it is a false alarm. Therefore, a false alarm probability P is given. FA , let γ T For Wγ to be SSE, the following probability equation should hold: The above formula determines... If the detection limit T is , then Detection limit: During navigation calculation, the calculation will be performed in real time. With σ r In comparison, if This indicates that a fault has been detected and an alert has been issued to the user.

6. The autonomous integrity monitoring method for satellite navigation receivers as described in claim 2, characterized in that: In step S3, the Balda data detection method is used to remove faulty satellites: based on the relationship between residuals and observation errors, the statistic can be set as: Calculate the statistics d for each satellite. i The satellite corresponding to the maximum value is the faulty satellite, which will trigger an alarm and be removed.

7. The autonomous integrity monitoring method for satellite navigation receivers as described in claim 2, characterized in that: In step S4, a convolutional neural network is established. When the number of satellites is less than 5, the traditional RAIM algorithm based on least squares residuals cannot be used. In this case, the convolutional neural network is used to monitor the integrity of the satellites.

Images