A multi-sensor graph optimization integrated navigation and fault diagnosis method

Abstract

The present invention discloses a multi-sensor graph optimization combined navigation and fault diagnosis method, which relates to the solution and robustness improvement of a GNSS / INS / airspeed meter / altimeter multi-sensor combined navigation system, and belongs to the technical field of calculation and extrapolation. The method obtains system sensor measurements, processes the measurements into INS factors and AVP factors, constructs a system navigation factor graph, and adds a sliding window to ensure the real-time performance of the algorithm. Next, the navigation system is optimized by factor graph optimization to realize the navigation function. On this basis, a fault diagnosis system is constructed by the residual of factor graph optimization, and the existing combined navigation data is classified and trained using a recurrent convolutional neural network to obtain a normal model and a fault model of the system, thereby identifying and diagnosing the faults of the navigation system.

Description

Technical Field

[0001] The present invention relates to the technical field of solution and fault diagnosis of a multi-sensor combined navigation system, and in particular to a GNSS / INS / airspeed indicator / altimeter multi-sensor graph optimized combined navigation fault diagnosis algorithm. Background Art

[0002] Navigation has always been an important research topic, with rich applications and development potential in various fields. High-performance navigation with high precision and high fault tolerance has broad application prospects and significant technical advantages, such as aircraft cluster formation, aerial refueling, rendezvous and docking, etc., which has become a current research hotspot and an important direction for future development; at the same time, unmanned driving, robot formation, mobile location services, smart earth, etc. are coming to us quickly. How to effectively improve navigation accuracy and high fault tolerance and give full play to the application efficiency of high-performance navigation has become one of the core issues in current navigation research. The Global Navigation Satellite System (GNSS) has all-weather positioning, attitude measurement, speed measurement and other capabilities. Further fusion with inertial sensors can complement each other's advantages and improve the overall navigation performance. However, the traditional GNSS / INS combined navigation technology is far from meeting the growing demand for high-performance navigation.

[0003] At present, the research on GNSS / INS integrated navigation at home and abroad is mainly based on the Bayesian filtering method. High-performance navigation often faces complex unknown dynamic environments. The traditional integrated navigation design has poor compatibility and is difficult to achieve plug-and-play. Factor graph optimization is an emerging branch of optimization theory that has the ability to handle large-scale constraints. It is not only superior to the filtering method in terms of accuracy, but also has a highly adaptable architecture. At present, graph optimization theory has become a hot topic in simultaneous localization and mapping (SLAM) research, but it is rarely applied to GNSS navigation and GNSS / INS integrated navigation. However, the traditional graph optimization algorithm, because it models and optimizes all the historical information of the entire system, the corresponding matrix size continues to grow over time, the amount of calculation continues to increase, and the real-time performance of the algorithm is difficult to guarantee. It is necessary to find new methods to ensure the real-time performance of the algorithm while ensuring the performance of the algorithm.

[0004] For high-performance relative navigation algorithms, fault tolerance and diagnosis is another difficult problem that needs to be solved urgently. Due to the GNSS non-line-of-sight (NLOS) problem, multipath effect, susceptibility to occlusion, etc. in complex environments, the navigation system model has obvious nonlinearity, non-Gaussianity and uncertainty. Standard graph optimization frameworks, Kalman filters, etc. are based on Gaussian error models, and navigation performance will be significantly deteriorated in complex environments. At present, research on improving navigation fault tolerance, such as GNSS integrity monitoring, mostly requires strict prior knowledge such as system models and noise statistical characteristics. However, it is difficult to obtain prior information in complex environments, and it is difficult to meet the requirements of good navigation and positioning robustness.

[0005] Regarding the fault diagnosis of the integrated navigation system, the traditional method is mainly a model-based fault diagnosis method, which establishes a model suitable for the data characteristics of the fault diagnosis object, and then efficiently completes the fault diagnosis. Its shortcomings are that it is difficult to establish an accurate mathematical model for systems with complex structures and numerous units, and it is easily disturbed by environmental factors during actual operation. Therefore, other fault diagnosis methods are needed as a supplement to obtain ideal fault diagnosis results. The data-based fault diagnosis technology does not need to establish an accurate model of the diagnosis object, and can well deal with the difficulty of modeling the fault diagnosis of complex nonlinear systems, which can be a good supplement to the model-based method. Summary of the invention

[0006] The purpose of the invention is to propose a GNSS / INS / airspeed meter / altimeter map optimization combined navigation fault diagnosis algorithm in view of the shortcomings of the above-mentioned background technology. Under the condition of ensuring the optimization accuracy of the traditional combined navigation system, the airspeed meter and altimeter information are introduced, the operating efficiency of the system is improved by increasing the sliding window, and the accuracy and fault tolerance of the algorithm are improved by more accurate noise modeling and increasing the switch constraints based on residuals. On top of this, a new idea for combined navigation fault diagnosis is provided by combining the method of combining factor graph optimization residuals with MRNN, thereby solving the fault tolerance and diagnosis problems of the combined navigation system.

[0007] The present invention adopts the following technical solutions to achieve the above-mentioned invention objectives:

[0008] A multi-sensor map optimization integrated navigation and fault diagnosis method, characterized in that it comprises the following steps:

[0009] Step 1, initialize the navigation system, obtain the initial measurement values ​​of GNSS, INS, airspeed indicator and altimeter sensors, and obtain the initial values ​​of system measurement;

[0010] Step 2: Obtain the measurement values ​​of each sensor at the current moment, and construct observation values ​​based on each measurement value as the input of the navigation system;

[0011] Step 3: Use the graph optimization method to construct a factor graph based on the observed values. Specifically, calculate the residual of the observed values ​​to obtain the measurement factors: INS factor and AVP factor, compare the residual with the given threshold, judge and add the switch constraint, and complete the factor graph at the current moment.

[0012] Step 4: Update the factor graph based on the measurement factor at the time, and add a sliding window to the factor graph so that when a new measurement factor enters, the amount of calculation is kept within a certain range;

[0013] Step 5, nonlinearly optimize the INS factor and AVP factor in the window in step 4, calculate the state transition probability corresponding to each factor, the state transition posterior probability of the entire factor graph, and the state when the posterior probability is maximum, convert the graph optimization problem into a nonlinear least squares problem, and obtain the optimal navigation solution in turn; return to step 2 to read the measurement value at the next moment, i.e., the moment, and repeat steps 2 to 5 until the combined navigation is completed;

[0014] Step 6: Based on the residual optimized by the factor graph in step 3, a fault diagnosis system is constructed to identify and diagnose the faults of the combined navigation results of the navigation system. Furthermore, the system sensor measurements include GNSS measurements, INS measurements, airspeed meter measurements, and barometric altimeter measurements. The observations are constructed through measurements and processed into INS factors and AVP factors. The measurement residuals are modeled through maximum a posteriori estimation, and the Gaussian mixture model is used to model the observation residuals of the system.

[0015] Furthermore, the sensor measurements in step 2 include INS measurements, GNSS measurements, airspeed meter measurements, and barometric altimeter measurements. The observations are constructed by the measurements. The observations include attitude A, velocity v, and position p information, expressed as z = [A, v, p]; the observation value z at the current k moment is constructed as and It is expressed as:

[0016]

[0017] in, The 3D attitude calculated by pre-integration for INS measurement, is the three-dimensional velocity measured by INS, The three-dimensional position information measured by INS; considering the accuracy advantages of different sensors, The three-dimensional attitude measured by GNSS; is the three-dimensional speed measured by the airspeed indicator; include Three components, among which is the longitude information measured by GNSS, is the latitude information measured by GNSS, This is the altitude information measured by the altimeter.

[0018] Furthermore, step 3 uses the residuals to construct the INS factors at time k and AVP factor Specifically:

[0019]

[0020] Among them, ||·|| represents the Mahalanobis distance, and It represents the attitude, velocity and position information at time k estimated from time k-1 according to dynamic modeling.

[0021] Furthermore, the switch constraint is designed such that if the residual corresponding to a certain measurement factor of the system at the current moment is greater than a threshold, an additional constraint factor is added to the measurement factor at that moment, thereby limiting and improving the impact of the measurement factor on the entire system factor graph.

[0022] Furthermore, in step 4, a sliding window is added to the factor graph, and the sliding window is moved in real time. The window size is N, which limits the number of INS factors involved in the optimization to N. For INS factors outside the window, the Schur's complement algorithm is used to convert the INS factor information before the window into prior information according to the operation frequency, that is, to calculate the prior probability and reduce the number of nodes in the factor graph, thereby reducing the number of factors, that is, the amount of calculation, while ensuring that the sliding window does not significantly affect the accuracy of the algorithm. The dynamic sliding window strategy decides to use the Schur's complement algorithm to marginalize the factor nodes in the factor graph a period of time ago according to the operation frequency, convert the information of these factors into prior information, reduce the number of nodes in the factor graph, and thus reduce the amount of calculation for each factor graph optimization.

[0023] Furthermore, the Gaussian mixture model is used to model the system residual as the coupling of multiple Gaussian noises, and the weight of each Gaussian distribution is determined by the parameters of the sensor itself. n and the parameter μ of a single Gaussian distribution n ,Σ n , establish a Gaussian mixture model.

[0024] Further, step 5 is specifically as follows: Step 5.1, calculate the state transition probability expression at time k corresponding to the INS factor and the AVP factor, expressed as in, are the corresponding probability normalization parameters respectively; Step 5.2, calculate the state transition posterior probability P(X k |Z k ), expressed as: Among them, K ins ,K avp are the number of INS and AVP factors respectively, P0 represents the prior probability calculated based on the factors outside the previous sliding window after the sliding window is added; Step 5.3, find the state when the posterior probability is maximum Expressed as in, and represent the X that maximizes and minimizes the expression f respectively. k , and finally get

[0025]

[0026] Furthermore, in step 6, based on the fault diagnosis system, the training set is constructed using the residual optimized by the factor graph, and the memory convolutional neural network MRNN is constructed. The parameters of the MRNN are obtained through training, and then the new factor graph optimized residual is input into the MRNN to determine the working status of the system at this time.

[0027] Furthermore, the memory convolutional neural network MRNN includes MRNN blocks connected in sequence, and the output of the previous MRNN block is linearly rectified and input into the next MRNN block; each MRNN block simultaneously extracts the spatial information provided by the convolution layer and the temporal information provided by the memory layer to diagnose the working status of the integrated navigation.

[0028] Furthermore, the residuals of factor graph optimization are expressed as By selecting a certain number of residuals at regular intervals, the number of selected residuals is equal to the number of classifications used in MRNN, and the scores of each group of residuals corresponding to each classification category are calculated respectively, and the classification category with the highest final score is used as the working state of the system.

[0029] Furthermore, the working status is mainly divided into normal operation, GNSS positioning sudden change, GNSS positioning gradual change, airspeed indicator speed measurement sudden change, airspeed indicator speed measurement slow change, altimeter positioning sudden change, altimeter positioning slow change and INS bias gradual change.

[0030] The memory convolutional neural network MRNN is constructed based on the training set as follows: for any k moment, select the residual r of c consecutive moments k+1 ,r k+2 ,…,r k+c-1 ,r k+c , MRNN fault diagnosis is expressed as f MRNN :

[0031] f MRNN ([r k+1 ,r k+2 ,…,r k+c-1 ,r k+c ])=[s k+1 ,s k+2 ,…,s k+c-1 ,s k+c ]

[0032] where s k+1 ,s k+2 ,…,s k+c-1 ,s k+c Represents r k+1 ,r k+2 ,…,r k+c-1 ,r k+cThe corresponding working states include normal operation, sudden change of GNSS positioning, gradual change of GNSS positioning, sudden change of airspeed indicator speed measurement, slow change of airspeed indicator speed measurement, sudden change of altimeter positioning, slow change of altimeter positioning and gradual change of INS bias.

[0033] The present invention adopts the above technical solution and has the following beneficial effects: the present invention improves on the basis of the traditional graph optimization algorithm and proposes a GNSS / INS graph optimization fault diagnosis method. The method first uses the Gaussian mixture model to enhance the robustness of the navigation system to non-Gaussian errors in the measurement; then, the sliding window method is used to reduce the number of factors of the system and thus reduce the amount of calculation of the system, thereby enhancing the calculation efficiency of the navigation system and meeting the real-time requirements of realistic navigation; finally, the residual is used to further increase the switch constraint, thereby further enhancing the fault tolerance performance of the system to fault errors. As a fault diagnosis system, the intelligent method of MRNN training can effectively use the residual information of each sensor optimized by the graph to match and obtain the working state of the combined navigation system. And by combining factor graph optimization with MRNN, it can solve the problem that the traditional fault diagnosis based on the filtering algorithm cannot diagnose the INS fault, and at the same time, the fault tolerance and fault diagnosis of navigation are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a combined navigation fault diagnosis system according to an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of a factor graph optimization model according to an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of an MRNN block according to an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of the MRNN structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The technical solution of the invention is described in detail below with reference to the accompanying drawings.

[0039] The present invention discloses a GNSS / INS / airspeed meter / altimeter map optimization combined navigation fault diagnosis algorithm, which obtains system sensor measurements, processes the measurements into INS factors and AVP factors, constructs a system navigation factor graph, and then optimizes the navigation system factor graph to achieve the navigation function. Next, a fault diagnosis system is constructed through the residual of factor graph optimization, and the existing combined navigation data is classified and trained using a memory convolutional neural network to obtain a model, thereby identifying and diagnosing navigation system faults.

[0040] The robust graph optimization integrated navigation algorithm of the present invention is described below by taking into account the integrated navigation system consisting of GNSS, INS, airspeed indicator and altimeter carried by a conventional aircraft. The flowchart of the entire method is as follows: Figure 1 As shown, it specifically includes the following 6 steps.

[0041] Step 1: Initialize the entire navigation system, initialize the prior information at the given initial time, obtain the initial position, velocity and attitude information [A0, v0, p0], and obtain the initial readings of the GNSS, INS, airspeed indicator and altimeter sensors.

[0042] Step 2: Collect the observation values ​​of GNSS, INS, airspeed indicator and altimeter at the current moment as system input, and obtain the attitude of the navigation system expressed in the current carrier coordinate system, the speed and longitude, latitude and altitude position information in the north-east-earth coordinate system.

[0043] Next, the state quantity and observation quantity of the system are clarified. Among them, the observation quantity is collected to update the estimation of the state quantity, and the optimal estimation is obtained by optimizing the estimation, which is regarded as the true value of the system state at this time.

[0044] The state quantity x of the system k Select the navigation parameters as follows:

[0045]

[0046] Among them, A,v,p, They are the attitude, velocity, position, accelerometer bias and gyroscope bias of the navigation object respectively.

[0047] Without loss of generality, we discuss the k moment. Construct the observation quantity of the system at the k moment and

[0048]

[0049] in, The 3D attitude calculated by pre-integration for INS measurement, is the three-dimensional velocity measured by INS, The 3D position information measured by INS. Next, considering the accuracy advantages of different sensors, we select The three-dimensional attitude measured by GNSS; is the three-dimensional speed measured by the airspeed indicator; and include Three components, among which is the longitude information measured by GNSS, is the latitude information measured by GNSS, This is the altitude information measured by the altimeter.

[0050] Step 3: Construct the factor graph of the navigation system: Use the graph optimization method to model the system observations using the factor graph. Figure 2 As shown. The residual of the observation value z is calculated, the measurement factor at time k is constructed, and the residual is compared with the set threshold, and the switch constraint is judged and added to complete the factor graph at the current moment.

[0051] The residual is the difference between the estimated value and the measured value at each moment, and the residual is used to construct the INS factor at moment k and AVP factor

[0052]

[0053] Among them, ||·|| represents the Mahalanobis distance, and It represents the attitude, velocity and position information at time k estimated from time k-1 according to the dynamic modeling. For residual modeling, considering that the actual situation is often non-Gaussian error, a Gaussian mixture model is used.

[0054] Gaussian mixture modeling regards the system residual as the coupling of multiple Gaussian noises, and determines the weight π of each Gaussian distribution through the parameters of the sensor itself n and the parameter μ of a single Gaussian distribution n ,Σ n , establish a Gaussian mixture model.

[0055] The Gaussian mixture model is specifically:

[0056]

[0057] Where p(x) is the probability function of the Gaussian mixture model; N is the total number of components; N(x|μ n ,Σ n ) is the n-component in the mixture model, each component conforms to the Gaussian distribution, μ n ,Σ n are the mean vector and covariance matrix of the Gaussian distribution, respectively. In this example, μ n ∈R 9 ,Σ n ∈R 9×9 ; π n Represents the weight of each component, satisfying The Gaussian mixture model represents the model distribution that the residual random variable obeys, which will affect the calculation of the covariance matrix required in the Mahalanobis distance. The variables of the factor graph appear in the form of a posteriori probability during the final optimization, so this distribution will change the expression of probability. In this example, a combination of three Gaussian distributions is used to simulate the distribution of the fitting residual, that is, N=3.

[0058] Adding switch constraints to the measurement factors, for various types of interference, the residuals often change greatly. At this time, the expression constructed by simply using the Mahalanobis distance will cause system failure. At this time, it is necessary to introduce switch constraints, such as Figure 3 The switchable constraint is obtained by giving a threshold ε of each measurement residual. ins ,ε avp , and the residual at the current k moment By comparison, when the residual is greater than the threshold, a new switching factor is introduced based on the original factor:

[0059]

[0060] Where, without loss of generality, the superscript Q∈{ins,avp} describes a specific measurement factor; represents the weight function, satisfying At this time, the expression of the measurement factor corresponding to time k under the switch constraint is You can write:

[0061]

[0062] Step 4, in order to reduce the complexity of the operation and improve the real-time performance of the algorithm, add a sliding window to the factor graph. When a new measurement comes in, in order to ensure that the amount of calculation does not increase over time, it is necessary to discard the factors with too large time intervals, so as to maintain the amount of calculation within a certain range. In this example, the INS factors are restricted. The number of INS factors involved in the optimization is limited to less than 1000. For factors that are earlier than these 1000 factors, the marginalization method is used, and the Schur complement algorithm is used to calculate the prior probability P0 of the corresponding sliding window, thereby reducing the number of factors, that is, the amount of calculation, while ensuring that the sliding window does not significantly affect the accuracy of the algorithm.

[0063] Step 5: Perform nonlinear optimization on the factor graph after adding the sliding window. First, calculate the state transition probability expression at time k corresponding to the INS factor and the AVP factor:

[0064]

[0065] in, are the corresponding probability normalization parameters respectively. Next, the state transition posterior probability P(X k |Z k ):

[0066]

[0067] Among them, K ins ,Kavp are the number of INS and AVP factors, respectively. P0 represents the prior probability calculated based on the factors outside the sliding window after the sliding window is added. In this example, considering that the frequency of INS is much higher than that of other sensors, there are 2000≥K ins ≥K avp Next, we find the state when the posterior probability is the largest You can get:

[0068]

[0069] in, and represent the X that maximizes and minimizes the expression f respectively. k Finally, we can get:

[0070]

[0071] Therefore, the graph optimization problem is transformed into a nonlinear least squares problem, and the optimal navigation solution can be obtained by using the Gauss-Newton method. Wait for the measurement information at time k+1 to be read, and then return to step 2. Repeat steps 2 to 5 to complete the normal navigation operation of the integrated navigation system.

[0072] Step 6: Based on the normal operation of the integrated navigation system, diagnose whether the working state of the navigation system is normal and whether there is any fault. This system uses MRNN to judge the working state of the system. The working states are mainly divided into normal operation, GNSS positioning sudden change, GNSS positioning gradual change, airspeed meter speed measurement sudden change, airspeed meter speed measurement slow change, altimeter positioning sudden change, altimeter positioning slow change and INS bias gradual change.

[0073] First, the test set is constructed by residuals. For any k moments, the residuals r of c consecutive moments are selected. k+1 ,r k+2 ,…,r k+c-1 ,r k+c , as the training set and test set of MRNN.

[0074] MRNN fault diagnosis can be expressed as f MRNN :

[0075] f MRNN ([r k+1 ,r k+2 ,…,r k+c-1 ,r k+c ])=[s k+1 ,s k+2 ,…,s k+c-1 ,s k+c ]

[0076] where sk+1 ,s k+2 ,…,s k+c-1 ,s k+c Represents r k+1 ,r k+2 ,…,r k+c-1 ,r k+c The corresponding working states include normal operation, sudden change of GNSS positioning, gradual change of GNSS positioning, sudden change of airspeed indicator speed measurement, slow change of airspeed indicator speed measurement, sudden change of altimeter positioning, slow change of altimeter positioning and gradual change of INS bias.

[0077] Initialize the MRNN parameters and use the training set as input to train the MRNN network. When the score is unstable, that is, the state classification is unstable, update the MRNN network parameters through back propagation. When the score tends to be stable, that is, the state classification is stable, the training can be considered complete, and the MRNN network parameters are determined to obtain a trained MRNN network.

[0078] The test set is input into the trained MRNN network, and the working status of the system is output to determine whether the system is in a normal working state or other fault state. The working status includes:

[0079] Normal operation, sudden change of GNSS positioning, gradual change of GNSS positioning, sudden change of airspeed indicator speed measurement, slow change of airspeed indicator speed measurement, sudden change of altimeter positioning, slow change of altimeter positioning and gradual change of INS bias.

[0080] The basic structure of MRNN is similar to other neural networks and can be divided into individual modules. Figure 3 As shown in the figure, each module contains the traditional RNN convolution layer and an additional memory layer M. The convolution layer reduces the size of information but greatly enhances the semantics; while the memory layer does not change the size of information but slowly enhances the semantics. The overall neural network structure is implemented as follows: Figure 4 As shown. Figure 4 The structure can be divided into different levels of neural networks, and each level extracts the spatial information provided by the convolution layer and the temporal information provided by the memory layer at the same time. The information between multiple levels enters the subsequent operation at the same time, which is conducive to improving the operation speed and reducing the time required for convergence. By training the training set, the obtained neural network can be used to judge the working status of the system during actual operation, that is, fault diagnosis.

Claims

1. A multi-sensor map optimization combined navigation and fault diagnosis method, characterized in that: The following steps are involved: Step 1, initialize the navigation system, obtain the initial measurement values ​​of GNSS, INS, airspeed indicator and altimeter sensors, and obtain the initial values ​​of system measurement; Step 2, obtain the measurement values ​​of each sensor at the current k moment, and construct the observation value z based on each measurement value as the input of the navigation system; Step 3: Use the graph optimization method to construct a factor graph based on the observed values. Specifically, the residual of the observed value z is calculated to obtain the measurement factors: INS factor and AVP factor, and the residual is compared with the given threshold, and the switch constraint is judged and added to complete the factor graph at the current moment. Step 4: Update the factor graph based on the measurement factor at time k, and add a sliding window to the factor graph so that when a new measurement factor enters, the amount of calculation is kept within a certain range; Step 5, nonlinearly optimize the INS factor and AVP factor in the window in step 4, calculate the state transition probability at time k corresponding to each factor, the state transition posterior probability of the entire factor graph, and the state when the posterior probability is maximum, convert the graph optimization problem into a nonlinear least squares problem, and obtain the optimal navigation solution in turn; return to step 2 to read the measurement value at the next time, that is, time k+1, and repeat steps 2 to 5 until the combined navigation is completed; Step 6: Based on the residual optimized by the factor graph in step 3, a fault diagnosis system is constructed to identify and diagnose the faults of the combined navigation results of the navigation system. Specifically, based on the fault diagnosis system, a training set is constructed using the residual optimized by the factor graph, and a memory convolutional neural network MRNN is constructed. The parameters of the MRNN are obtained through training, and then the new factor graph optimized residual is input into the MRNN to determine the working state of the system at this time. The memory convolutional neural network MRNN includes MRNN blocks connected in sequence, and the output of the previous MRNN block is linearly rectified and input into the next MRNN block; each MRNN block simultaneously extracts the space information provided by the convolution layer and the time information provided by the memory layer to diagnose the working status of the integrated navigation.

2. The multi-sensor image optimization combined navigation and fault diagnosis method according to claim 1, characterized in that: In step 2, the sensor measurements include INS measurements, GNSS measurements, airspeed meter measurements, and barometric altimeter measurements. The observations are constructed through the measurements. The observations include attitude A, velocity v, and position p information, expressed as z = [A, v, p]. The observation value z at the current k moment is constructed as the optimal combination under and It is expressed as: in, The 3D attitude calculated by pre-integration for INS measurement, is the three-dimensional velocity measured by INS, The three-dimensional position information measured by INS; considering the accuracy advantages of different sensors, The three-dimensional attitude measured by GNSS; is the three-dimensional speed measured by the airspeed indicator; include Three components, among which is the longitude information measured by GNSS, is the latitude information measured by GNSS, This is the altitude information measured by the altimeter.

3. The multi-sensor image optimization combined navigation and fault diagnosis method according to claim 2 is characterized in that: Step 3: Use the residuals to construct the INS factors at time k and AVP factor Specifically: Among them, ||·|| represents the Mahalanobis distance, and It represents the attitude, velocity and position information at time k estimated from time k-1 according to dynamic modeling.

4. The multi-sensor image optimization combined navigation and fault diagnosis method according to claim 3 is characterized in that: The switch constraint in step 3 is that if the residual corresponding to a certain measurement factor of the system at the current moment is greater than the threshold, an additional constraint factor is added to the measurement factor at that moment, thereby limiting and improving the influence of the measurement factor on the entire system factor graph.

5. The multi-sensor image optimization combined navigation and fault diagnosis method according to claim 4 is characterized in that: In step 4, a sliding window is added to the factor graph and moved in real time. The window size is N, which limits the number of INS factors involved in the optimization to N. For INS factors outside the window, the Schur complement algorithm is used to convert the INS factor information before the window into prior information according to the operation frequency, that is, to calculate the prior probability P0, and reduce the number of nodes in the factor graph. This ensures that the sliding window will not significantly affect the accuracy of the algorithm while reducing the number of factors, that is, the amount of calculation.

6. The multi-sensor image optimization combined navigation and fault diagnosis method according to claim 5, characterized in that: Step 5 is as follows: Step 5.1, calculate the state transition probability expression at time k corresponding to the INS factor and the AVP factor, expressed as in, are the corresponding probability normalization parameters respectively; Step 5.2, calculate the state transition posterior probability P(X k |Z k ), expressed as: Among them, K ins ,K avp are the number of INS and AVP factors, respectively. P0 represents the prior probability calculated based on the factors outside the previous sliding window after the sliding window is added. Step 5.3, find the state when the posterior probability is maximum Expressed as in, and represent the X that maximizes and minimizes the expression f respectively. k , and finally get 7. The multi-sensor image optimization integrated navigation and fault diagnosis method according to claim 1, characterized in that: The residuals of factor graph optimization are expressed as By selecting a certain number of residuals at regular intervals, the number of selected residuals is equal to the number of classifications used in MRNN, and the scores of each group of residuals corresponding to each classification category are calculated respectively, and the classification category with the highest final score is used as the working state of the system.

8. The multi-sensor image optimization integrated navigation and fault diagnosis method according to claim 7, characterized in that: The memory convolutional neural network MRNN is constructed based on the training set as follows: for any k moment, select the residual r of c consecutive moments k+1 ,r k+2 ,…,r k+c-1 ,r k+c , MRNN fault diagnosis is expressed as f MRNN : f MRNN ([r k+1 ,r k+2 ,…,r k+c-1 ,r k+c ])=[s k+1 ,s k+2 ,…,s k+c-1 ,s k+c ] where s k+1 ,s k+2 ,…,s k+c-1 ,s k+c Represents r k+1 ,r k+2 ,…,r k+c-1 ,r k+c The corresponding working states include normal operation, sudden change of GNSS positioning, gradual change of GNSS positioning, sudden change of airspeed indicator speed measurement, slow change of airspeed indicator speed measurement, sudden change of altimeter positioning, slow change of altimeter positioning and gradual change of INS bias.

Images