Adaptive maintenance of pseudorange availability and navigation error correction in denial environment

Abstract

The application discloses a method for adaptive maintenance of pseudo-position availability and correction of navigation error in denial environment, a pseudo-position sensor is constructed through a memory network, a training mode, a prediction mode and a verification mode are designed, the prediction accuracy and state of the memory network are evaluated on line under the condition that the motion environment of a carrier and the measurement condition dynamically change, and the parameters of the memory network are dynamically adjusted to complete the maintenance of the pseudo-position availability in the shortest training time; in the denial environment, the system mode is flexibly selected, the inertial navigation system is corrected in time by using the memory network, and the availability of the network is maintained. The application can maintain the prediction accuracy of the pseudo-position on line under complex application conditions and shorten the training time according to the performance requirement, improves the training efficiency and the availability ratio of the navigation system, and is suitable for practical application.

Description

Technical Field

[0001] This invention relates to positioning and navigation methods, and more particularly to a method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in denied environments. Background Technology

[0002] Integrated navigation refers to combining an inertial navigation system and auxiliary sensors in an appropriate manner. By leveraging the complementary performance of inertial navigation and auxiliary sensors, higher system accuracy can be achieved than with a single system. However, when the auxiliary sensor signal is interrupted, causing a significant drop in positioning accuracy, the system switches to a pure inertial navigation system. But the inertial navigation system is affected by the accumulation of errors, causing the positioning accuracy to gradually decrease over time.

[0003] Research on neural network-based integrated navigation technology has leveraged its intelligent characteristics to provide significant assistance to inertial navigation systems. However, traditional static neural networks, when applied to integrated navigation, cannot analyze the temporal characteristics of inertial navigation system errors. Furthermore, when processing large amounts of data, they suffer from slow convergence and a tendency to get trapped in local minima. While dynamic memory networks offer the potential to overcome these issues, they lack effective maintenance of their usability in response to changes in environment and measurement characteristics. This leads to over-reliance on sample size and parameter rigidity, hindering the utilization of their dynamic characteristics and their auxiliary role in inertial navigation. In some cases, they may even generate incorrect predictions, causing navigation error divergence. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide an adaptive maintenance and navigation error correction method for pseudo-positioning availability under denied environments that can overcome the problem of decreased positioning accuracy when auxiliary sensors fail, thereby curbing error divergence in inertial navigation systems and improving positioning accuracy.

[0005] Technical solution: The present invention provides an adaptive maintenance method for pseudo-positioning availability and navigation error correction in a denied environment, comprising the following steps:

[0006] S1, obtain the angle increment and velocity increment output by the inertial navigation system;

[0007] S2, acquire the aircraft's longitude and latitude position increments when the auxiliary sensors are effective;

[0008] S3, the inertial navigation system enters the training mode, using the normalized angle increment and velocity increment output by the inertial navigation system as input, and the normalized position increment output by the auxiliary sensor as output to train the memory network, and compares the training error with the expected error, and uses the fireworks algorithm to adaptively adjust the parameters of the memory network.

[0009] S4, when the auxiliary sensor fails, the inertial navigation system enters prediction mode and uses the trained memory network to predict the position increment of the integrated navigation.

[0010] S5, calculates the position of the aircraft to be located when the auxiliary sensor fails;

[0011] S6. After the auxiliary sensor recovers, the inertial navigation system enters the verification mode, compares the difference between the memory network and the integrated navigation with the set expected error size, and determines whether it is necessary to continue training the memory network; if it is necessary to continue training the memory network, it returns to step S3.

[0012] Furthermore, in step S1, the specific implementation steps for obtaining the angle increment and velocity increment output by the inertial navigation system are as follows:

[0013] S11, acquire the measurement data output by the gyroscope and accelerometer required for aircraft positioning;

[0014] S12 calculates the output data from the gyroscope and accelerometer, converts it to obtain the carrier angular velocity and carrier acceleration, and stores the carrier angular velocity information ω output by the inertial navigation system during the aircraft's flight. IMU With carrier acceleration information a IMU ;

[0015] S13, based on the obtained angular velocity and acceleration information of the carrier output by the inertial navigation system, the inertial navigation system t i Time and t i+Δ The angle and velocity information output at any given time are subtracted to obtain the aircraft's angle increment and velocity increment output by the inertial navigation system. The calculation formula is as follows:

[0016]

[0017] Where, ΔΩ IMU For the angle increment of the aircraft system, Δv IMU For the speed increment of the aircraft system, For t i+Δ The carrier's angular velocity is obtained by measuring and converting data using a gyroscope. For t i The carrier's angular velocity is obtained by measuring and converting data using a gyroscope. For t i+Δ The acceleration of the carrier is measured and converted by a constant accelerometer. For t i The acceleration of the carrier is measured and converted by a constant accelerometer. For t i Time to t i+Δ The carrier's X-axis angle increment at time t. For t iTime to t i+Δ The Y-axis angle increment of the carrier at time t, For t i Time to t i+Δ The Z-axis angle increment of the carrier at time t, For t i Time to t i+Δ The increment of the carrier's X-axis velocity at time t. For t i Time to t i+Δ The increment of the Y-axis velocity of the carrier at any given time.

[0018] Furthermore, in step S2, the specific steps for obtaining the aircraft's longitude and latitude position increments when the auxiliary sensors are effective are as follows:

[0019] S21, acquire the longitude L and latitude λ of the aircraft measured by the auxiliary sensors;

[0020] S22, based on the longitude and latitude positions of the aircraft obtained in step S21, adjust the auxiliary sensor t i Time and t i+Δ The position information output at any given time is differentially analyzed to obtain the spacecraft's longitude and latitude position increments:

[0021]

[0022] in, For t i+Δ The longitude position of the spacecraft at any given time. For t i The longitude position of the spacecraft at any given time. For t i+Δ The latitude and position of the spacecraft at any given time. For t i The latitude and position of the spacecraft at any given time.

[0023] Furthermore, in step S3, the detailed implementation steps of adaptively adjusting the memory network parameters using the fireworks algorithm are as follows:

[0024] S31, if the auxiliary sensor is not malfunctioning, take the angle increment and velocity increment output by the inertial navigation system obtained in step S1 as the training input sequence, perform normalization processing, and obtain standardized input data X for training. Tr ∈{x t |t i <t<t i+Δ}, its expression is:

[0025]

[0026] in, Let be the increment of the carrier's X-axis velocity at time t. Let t be the increment of the Y-axis velocity of the carrier. Let t be the carrier's X-axis angle increment. Let t be the Y-axis angle increment of the carrier. The Z-axis angular increment of the carrier at time t; T represents the transpose matrix;

[0027] S32, if the auxiliary sensor is not malfunctioning, take the aircraft's longitude and latitude position increments obtained in step S2 as the training output sequence, perform normalization processing, and obtain standardized output data Y for training. Tr ∈{y t |t i <t<t i+Δ}, its expression is:

[0028]

[0029] in, This represents the longitude increment at time t. This represents the latitude increment at time t;

[0030] S33, the inertial navigation system enters training mode, and the standardized input data X is transferred. Tr ∈{x t |t i <t<t i+Δ The standardized output data Y is used as the training input. Tr ∈{y t |t i <t<t i+Δ As the training output, it establishes the training data framework for the memory network;

[0031] S34, train the memory network using the gradient-based Adam method, updating the value θ using gradient optimization. t for:

[0032]

[0033] Where η is the learning rate. The gradient is the average value at the first time step. Let ε be the variance of the gradient at the second time step, and ε be a random number in the range of 0 to 1.

[0034] S35, define the sequence input layer and determine the size of the input sequence at each time step;

[0035] S36, Create a memory layer and determine the number of hidden units in the layer;

[0036] S37, Create a fully connected layer with a specified output size;

[0037] S38, Create a regression output layer and perform data fitting;

[0038] S39, Adjust the parameters of the memory network. When the training error exceeds a set threshold, the parameters of the memory network are optimized by selecting the number of optimization operators and the optimization radius. The number of hidden layers and the learning rate in the memory network parameters are adaptively determined and adjusted. The expression is as follows:

[0039]

[0040]

[0041] Where f(N,η) is the training architecture of the memory network, and the optimization parameters are the number of hidden layers N and the learning rate η, S e To find the optimal number of operators, A e To find the optimal radius, y max and y min Let b be the range of optimization, and b be a constant that adjusts the number of optimization operators. To adjust the constant of the optimization radius, α is the minimum mechanical quantity to prevent division by zero, and q is the initial number of optimization operators;

[0042] S310 sets the maximum number of learning epochs after the error stabilizes. When the memory network has trained to the maximum number of learning epochs after the error stabilizes, and the mean absolute error (MAE) is less than the set threshold, the memory network stops training. The formula for calculating the mean absolute error is:

[0043]

[0044] Let the current memory network be at layer l, time t, and the error propagated from the input of layer l+1 at time t be denoted as . The error propagated back from layer l at time t+1 is denoted as... n is the number of columns in the training output array, and n is the number of each output state variable.

[0045] Furthermore, in step S4, when the auxiliary sensor fails, the system enters prediction mode, and the specific steps for predicting the integrated navigation position increment using the trained memory network are as follows:

[0046] S41, When the auxiliary sensor fails, the system switches from training mode to prediction mode, and the angle increment and velocity increment X output by the inertial navigation system when the auxiliary sensor fails are recorded. Tr ∈{x t |t j <t<t j+Δ The data is fed into a trained memory network for learning, and the network outputs a prediction array Y. Tr ∈{y t |t j <t<t j+Δ}, and Y Tr De-standardization yields the memory network's predicted position increment array Y. pr ={y pr1 ,y pr2 ,...,y prs},in:

[0047]

[0048] in, To train the output array Y Tr ∈{y t |t j <t<t j+Δ The mean of} To train the output array Y Tr ∈{y t |t j <t<t j+Δ The standard deviation of}; m is the number of rows in the training output array; s is the number of columns in the prediction input and output arrays;

[0049] S42, Output the array of pseudo-observation information of position increments predicted by the memory network:

[0050] Y pr ={y pr1 ,y pr2 ,...,y prs}

[0051] Furthermore, in step S5, the specific steps for calculating the position of the aircraft to be located when the auxiliary sensor fails are as follows:

[0052] S51, obtain the aircraft's position at the moment the auxiliary sensor signal is lost. The location increment pseudo-observation information Y predicted by the memory network pr ={y pr1 ,y pr2 ,...,y prs}Accumulate to The false position observation information calculated after the auxiliary sensor signal is lost. The expression is as follows:

[0053]

[0054] Among them, t s The time when the auxiliary sensor signal is lost is 's', and the prediction duration is 's'.

[0055] Furthermore, in step S6, the system enters verification mode. If it is necessary to continue training the memory network, the specific steps of returning to step S3 are as follows:

[0056] S61, monitor training progress and calculate training residual Δ c Its expression is:

[0057]

[0058] Among them, Y prva ={y pr1 ,y pr2 ,...,y pru} represents the pseudo-observation information of the location increment predicted by the memory network in verification mode, Y Te ={y Te1 ,y Te2 ,...,y Teu} represents the incremental navigation result after recovery by the auxiliary sensor, and u represents the verification duration. In the verification mode, the prediction frequency of the memory network remains unchanged.

[0059] S62, training residual Δ c Compared with the expected prediction error limit, when the training residual Δ c If the error falls below the expected prediction limit, training is stopped; otherwise, proceed to step S3 and re-enter training mode.

[0060] Compared with the prior art, the significant advantages of this invention are as follows:

[0061] 1. In the event of sensor failure, the optimal memory network model obtained through training is used to predict the position increment information of the auxiliary sensor by inputting the angle increment and velocity increment of the vehicle system output by the inertial navigation system. The position increment can be integrated to obtain the position pseudo-observation information, which provides error correction for the solution results of the inertial navigation system and prevents the positioning accuracy of the system from declining rapidly.

[0062] 2. When the training error fails to meet the expected error requirements, this invention adaptively adjusts the parameters of the memory network and optimizes the learning time of the memory network. Training stops when the training error is less than a set threshold, which can effectively shorten the training time and improve the training efficiency and availability of the navigation system.

[0063] 3. During the period of auxiliary sensor failure, this invention uses a memory network to predict the position increment and suppress the divergence of the inertial navigation system. After the auxiliary sensor signal is recovered, the integrated navigation continues and the difference between the memory network prediction and the integrated navigation and the set expected error size are analyzed to determine whether further training is needed and to update the memory network in a timely manner. This achieves adaptive maintenance of availability and dynamic maintenance of the credibility of pseudo-positioning solutions, thereby effectively correcting the inertial navigation system during the period of auxiliary sensor signal loss and improving the system's positioning accuracy.

[0064] 4. This invention designs three system modes for three scenarios: auxiliary sensor effectiveness, auxiliary sensor failure, and auxiliary sensor recovery. These are training mode, prediction mode, and verification mode. This invention can rationally select the system mode based on the state of the auxiliary sensor signal and flexibly switch between the three modes: training mode is used when the auxiliary sensor is effective, prediction mode is used when the auxiliary sensor fails, and verification mode is used after the auxiliary sensor signal recovers. It determines whether the system needs to re-enter training mode, which improves the system's adaptability to changes in application environment and measurement conditions. While reducing dependence on the number of samples, it maintains optimal auxiliary prediction accuracy.

[0065] 5. In the event of auxiliary sensor signal failure, this invention can prevent the system from entering pure inertial mode by relying on only a short training period compared with traditional methods, thus achieving better positioning accuracy and making it suitable for practical applications in complex denial environments. Attached Figure Description

[0066] Figure 1 This is a schematic diagram illustrating the principle and flow of the method of the present invention;

[0067] Figure 2(a) shows a comparison curve between the longitude position increment predicted by the method of the present invention and the ideal longitude position increment with regularity;

[0068] Figure 2(b) shows a comparison curve between the latitude position increment predicted by the method of the present invention and the ideal latitude position increment with regularity;

[0069] Figure 3(a) shows the X-axis angle increment curve of the aircraft system for training the memory network and predicting the input quantity of the input quantity of the aircraft system according to the method of the present invention.

[0070] Figure 3(b) shows the Y-axis angle increment curve of the aircraft system for training the memory network and predicting the input quantity of the predicted input quantity using the method of the present invention.

[0071] Figure 3(c) shows the Z-axis angle increment curve of the aircraft system for training the memory network and predicting the input quantity of the input quantity of the method of the present invention.

[0072] Figure 3(d) shows the X-axis velocity increment curve of the aircraft system for training the memory network and predicting the input quantity of the input quantity of the method of the present invention.

[0073] Figure 3(e) shows the Y-axis velocity increment curve of the aircraft system for training the memory network and predicting the input quantity of the input quantity of the method of the present invention.

[0074] Figure 4(a) shows the comparison curve between the predicted output of the memory network considering the failure stage of the present invention and the longitude increment of the ideal track position;

[0075] Figure 4(b) shows the comparison curve between the predicted output of the memory network considering the failure stage of the present invention and the latitude increment of the ideal track position;

[0076] Figure 4(c) shows the comparison curve between the predicted output of the two-memory network considering the failure stage and the longitude increment of the ideal track position in the method of the present invention.

[0077] Figure 4(d) shows the comparison curve between the predicted output of the two-memory network considering the failure stage and the latitude increment of the ideal track position in the method of the present invention.

[0078] Figure 5 This is a comparison chart of navigation curves for the pseudo-positioning navigation error correction method used in the denied environment according to the method of the present invention;

[0079] Figure 6 This is a comparison diagram of the positional error of each method in the embodiment of the present invention considering the failure stage;

[0080] Figure 7 This is a comparison diagram of the positional errors of each method in the embodiment of the present invention considering the failure stage.

[0081] Figure 8 In the method of this invention, the number of training iterations of the memory network is adaptively adjusted according to the training situation.

[0082] Figure 9 In the method of this invention, the training prediction ratio of the memory network is adaptively adjusted according to the training situation.

[0083] Figure 10 These are curves showing the changes of the three system modes in the method of this invention over time.

[0084] Figure 11 The figures show the position error comparison curves and system mode change diagrams for each method in the embodiments of the present invention. Detailed Implementation

[0085] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0086] This invention utilizes a memory network to correct the inertial navigation system in the event of auxiliary sensor failure, and adaptively adjusts the parameters of the memory network. The invention designs three modes for the memory network: training mode, prediction mode, and verification mode. In application, the operating mode is adaptively adjusted according to changes in the environment and measurement characteristics, realizing an adaptive maintenance system for pseudo-positioning availability and a navigation error correction system. This invention enables intelligent navigation of the aircraft under small sample conditions through the memory network, reducing the positioning accuracy deviation caused by auxiliary sensor failure, reducing inertial navigation system error divergence, and improving positioning accuracy, thereby enhancing the aircraft's adaptability to environments where auxiliary sensors fail.

[0087] like Figure 1 As shown, the method for correcting false positioning navigation errors in a denied environment includes the following steps:

[0088] Step 1: Obtain the output data of the inertial navigation system required for integrated navigation and positioning;

[0089] Step 11: Obtain the measurement data output by the gyroscope and accelerometer required for aircraft positioning;

[0090] Step 12: Calculate the output data of the gyroscope and accelerometer, convert them to obtain the carrier angular velocity and carrier acceleration, and store the carrier angular velocity ω output by the inertial navigation system during the flight of the aircraft. IMU With the acceleration a of the carrier IMU ;

[0091] Step 13, based on the carrier angular velocity ω output by the obtained inertial navigation system... IMU With the acceleration a of the carrier IMU For inertial navigation systems t i Time and t i+Δ The angle and velocity information output at any given time are subtracted to obtain the aircraft's angle and velocity increments output by the inertial navigation system. The calculation formula is as follows:

[0092]

[0093] Where, ΔΩ IMU For the angle increment of the aircraft system, Δv IMU For the speed increment of the aircraft system, For t i+Δ The carrier's angular velocity is obtained by measuring and converting data using a gyroscope. For t i The carrier's angular velocity is obtained by measuring and converting data using a gyroscope. For t i+Δ The acceleration of the carrier is measured and converted by a constant accelerometer. For t i The acceleration of the carrier is measured and converted by a constant accelerometer. For t i Time to t i+Δ The carrier's X-axis angle increment at time t. For t i Time to t i+Δ The Y-axis angle increment of the carrier at time t, For t i Time to t i+Δ The Z-axis angle increment of the carrier at time t, For t i Time to t i+Δ The increment of the carrier's X-axis velocity at time t. For t i Time to t i+Δ The increment of the Y-axis velocity of the carrier at any given time.

[0094] Step 2: Obtain the aircraft's longitude and latitude position increments when the auxiliary sensors are effective;

[0095] Step 21: Obtain the aircraft positioning information L and λ measured by the auxiliary sensors, where L is the longitude position of the aircraft and λ is the latitude position of the aircraft.

[0096] Step 22: Based on the longitude and latitude position information of the aircraft obtained in Step 21, adjust the auxiliary sensor t. i Time and t i+Δ The position information output at any time is differentially analyzed to obtain the spacecraft's longitude and latitude position increments;

[0097]

[0098] Among them, L ti+Δ For t i+Δ The longitude position of the spacecraft at any given time. For t i The longitude position of the spacecraft at any given time. For t i+Δ The latitude and position of the spacecraft at any given time. For t i The latitude and position of the spacecraft at any given time.

[0099] Step 3: The system enters training mode, uses the angle and velocity increments output by the normalized inertial navigation system as inputs, and the position increments output by the normalized auxiliary sensor as outputs to train the memory network. The training error is compared with the expected error, and the fireworks algorithm is used to adaptively adjust the parameters of the memory network.

[0100] Step 31: When the auxiliary sensor is not malfunctioning, take the angle and velocity increment data output by the inertial navigation system when the auxiliary sensor is effective, obtained in Step 13, as the training input sequence and perform normalization processing; obtain the standardized input data X for training. Tr ∈{x t |t i <t<t i+Δ}, its expression is:

[0101]

[0102] in, Let be the increment of the carrier's X-axis velocity at time t. Let t be the increment of the Y-axis velocity of the carrier. Let t be the carrier's X-axis angle increment. Let t be the Y-axis angle increment of the carrier. The Z-axis angular increment of the carrier at time t; T represents the transpose matrix;

[0103] Step 32: When the auxiliary sensor is not malfunctioning, the system is in training mode. The aircraft's longitude and latitude position increments when the auxiliary sensor was effective, obtained in step 22, are taken as the training output sequence and normalized to obtain standardized output data Y for training. Tr ∈{y t |t i <t<t i+Δ}, its expression is:

[0104]

[0105] in, This represents the longitude increment at time t. This represents the latitude increment at time t;

[0106] Step 33: The system enters training mode, establishes a memory network, and uses the standardized angle and velocity increment data X output by the inertial navigation system. Tr ∈{x t |t i <t<t i+Δ The position information Y output by the auxiliary sensor is used as the training input. Tr ∈{y t |t i <t<t i+Δ As the training output, it establishes the training data framework for the memory network;

[0107] Step 34: Train the memory network using the gradient-based Adam method, updating the gradient value θ. t for:

[0108]

[0109] Where η is the learning rate. The gradient is the average value at the first time step. Let ε be the variance of the gradient at the second time step, and ε be a random number in the range of 0 to 1.

[0110] Step 35: Define the sequence input layer and determine the size of the input sequence at each time step;

[0111] Step 36: Create a memory layer and determine the number of hidden units in the layer, i.e., the number of neurons in the layer;

[0112] Step 37: Create a fully connected layer with a specified output size;

[0113] Step 38: Create a regression output layer and perform data fitting;

[0114] Step 39: Adjust the memory network parameters. When the training error exceeds a set threshold, the parameters of the memory network are optimized by selecting the number of optimization operators and the optimization radius. The number of hidden layers and the learning rate in the memory network parameters are adaptively determined and adjusted. The expression is as follows:

[0115]

[0116]

[0117] Where f(N,η) is the training architecture of the memory network, and the optimization parameters are the number of hidden layers N and the learning rate η, S e To find the optimal number of operators, A e To find the optimal radius, y max and y min Let b be the range of optimization, and b be a constant that adjusts the number of optimization operators. To adjust the constant of the optimization radius, α is the minimum mechanical quantity to prevent division by zero, and q is the initial number of optimization operators.

[0118] Based on the optimization radius A e Number of optimization operators S within the range e The primary principle is to filter the optimization parameter results and calculate the probability of each operator being selected, expressed as:

[0119]

[0120]

[0121] Among them, R P (N,η) represents the probability of the optimization operator being selected, K is the set of all results calculated by the optimization model, e is the result number corresponding to all results calculated by the optimization model, and R(N,η) is the sum of the distances between the current optimization operator and all other optimization operators. P The system is trained using the larger (N,η) optimization results, and finally the parameters with the smallest prediction error are selected as the number of hidden layers and the learning rate of the memory network.

[0122] Step 310: Set the maximum number of learning epochs after the error stabilizes. When the memory network has trained to the maximum number of learning epochs after the error stabilizes, and the mean absolute error (MAE) is less than the set threshold, the memory network stops training. The formula for calculating the mean absolute error is:

[0123]

[0124] Here, assuming the current memory network is at layer l and time t, the error propagated from the input of layer l+1 at time t is denoted as... The error propagated back from layer l at time t+1 is denoted as... n is the number of columns in the training output array, that is, the number of each output state variable.

[0125] Step 4: When the auxiliary sensor fails, the system enters prediction mode and uses the trained memory network to predict the position increment of the integrated navigation.

[0126] Step 41: When the auxiliary sensor fails, the system switches from training mode to prediction mode, and the angle and velocity increment X output by the inertial navigation system when the auxiliary sensor fails is recorded. Tr ∈{x t |t j <t<t j+Δ The data is fed into a trained memory network for learning, and the network outputs a prediction array Y. Tr ∈{y t |t j <t<t j+Δ}, and Y Tr De-standardization yields the memory network's predicted position increment array Y. pr ={y pr1 ,y pr2 ,...,y prs},in:

[0127]

[0128] in, That is, the training output array Y Tr ∈{y t |t j <t<t j+Δ The mean of} That is, the training output array Y Tr ∈{y t |t j <t<t j+Δ The standard deviation of}, m is the number of rows in the training output array, i.e. the dimension of the output state, and s is the number of columns in the prediction input and output arrays;

[0129] Step 42: Output the array Y of location increment pseudo-observation information predicted by the memory network. pr ={y pr1 ,y pr2 ,...,y prs}

[0130] Step 5: Calculate the position of the aircraft to be located when the auxiliary sensors fail;

[0131] Step 51: Obtain the position of the aircraft at the moment when the auxiliary sensor signal is lost. The array Y of location increment pseudo-observation information predicted by the memory network pr ={y pr1 ,ypr2 ,...,y prs}Accumulate to The false position observation information calculated after the auxiliary sensor signal is lost. for:

[0132]

[0133] Among them, t s The time when the auxiliary sensor signal is lost is s, which is the prediction duration. In the prediction mode, the prediction frequency of the memory network is 1Hz.

[0134] Step 6: After the auxiliary sensor recovers, the system enters the verification mode, compares the difference between the memory network and the integrated navigation with the set expected error size, and determines whether it is necessary to continue training the memory network. If so, it returns to step 3.

[0135] Step 61: Monitor training progress and calculate training residual Δ c Its expression is:

[0136]

[0137] Among them, Y prva ={y pr1 ,y pr2 ,...,y pru} represents the pseudo-observation information of the location increment predicted by the memory network in verification mode, Y Te ={y Te1 ,y Te2 ,...,y Teu} represents the incremental navigation result after recovery by the auxiliary sensor, u represents the verification duration, and the memory network prediction frequency remains 1Hz in the verification mode;

[0138] Step 62, the training residual Δ c Compared with the expected prediction error limit, when the training residual Δ c If the error falls below the expected prediction limit, training is stopped; otherwise, proceed to step 3 and re-enter training mode.

[0139] To verify the effectiveness of the proposed method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in denied environments, digital simulation analysis was performed on this intelligent integrated navigation method. In the simulation images, method 1 of this invention was not used for integrated navigation, and method 2 of this invention was not used for traditional static neural network assistance.

[0140] Figure 2(a)-Figure 2(b) To demonstrate the accuracy of the memory network's predictions, a comparison image was created between the predicted longitude and latitude increments and the regular ideal longitude and latitude increments. Monotonically increasing and sinusoidally varying prediction values ​​were used.

[0141] The simulation time for the aircraft trajectory used in the simulation experiment was 1500 seconds. During the 1-900 and 1000-1200 seconds, the auxiliary sensors were effective, while during the 900-1000 and 1200-1500 seconds, the auxiliary sensors had no signal. Figures 3(a)-3(e) This is the input for training and prediction in the memory network. Figures 4(a)-4(b) To consider the comparison curve between the predicted output of the memory network and the ideal latitude and longitude increments during the failure phase, Figures 4(c)-4(d) The comparison curve between the predicted output of the memory network and the ideal latitude and longitude increment is taken into account in the failure stage. Figure 5 To compare the longitude and latitude positioning results with those of integrated navigation and traditional static neural network-assisted navigation after adopting this method, statistical analysis of the failure stage shows that the proposed method can reduce the root mean square error to 10. -4 With a resolution of approximately 1000 degrees, the prediction performance of this invention is superior to that of traditional static neural networks.

[0142] Depend on Figure 6 Simulation results show that the auxiliary sensor fails within 900-1000 seconds. Within 67 seconds of the auxiliary sensor failure, the navigation performance of the present invention is inferior to that of a divergent pure inertial navigation system. After 67 seconds of failure, the navigation error of the present invention is smaller than that of a pure inertial navigation system. Therefore, the memory network in the present invention can effectively suppress inertial navigation divergence under these conditions and outperform the prediction effect of traditional static neural network-assisted systems.

[0143] Depend on Figure 7 Simulation results show that the auxiliary sensor fails within 1200-1500 seconds. Within 100 seconds of the auxiliary sensor failure, the navigation performance of the present invention is inferior to that of a pure inertial navigation system with divergent properties. After 100 seconds of failure, the navigation error of the present invention is smaller than that of a pure inertial navigation system. Therefore, the memory network in the present invention can effectively suppress inertial navigation divergence under this condition and is superior to the prediction effect of traditional static neural network assistance.

[0144] Figures 8-9 An analysis image that adaptively adjusts the corresponding parameters based on the training progress of the memory network.

[0145] Tables 1 and 2 show the root mean square error of the output prediction based on the adaptive adjustment of the corresponding parameters according to the method of the present invention.

[0146] Depend on Figures 8-9As shown in Tables 1 and 2, the simulation results indicate that, with unchanged training data, too few training iterations lead to insufficient training of the memory network and large prediction errors; too many training iterations result in a more stable prediction error but also increase training time. Therefore, the number of training iterations for the memory network needs to be selected appropriately based on the actual situation. Regarding the training-to-prediction ratio of the memory network, the larger the ratio of training data to prediction data, the more comprehensive the memory network's understanding of the carrier's motion, and the more accurate the predicted output.

[0147] Depend on Figures 10-11 It can be seen that the system is in training mode when the auxiliary sensor is effective from 0 to 900 seconds; in prediction mode when the auxiliary sensor fails from 900 to 1000 seconds; in verification mode when the auxiliary sensor recovers from 1000 to 1200 seconds, as shown by the simulation images, the verification prediction error of the memory network is less than the set threshold, so no further training is needed; and in prediction mode when the auxiliary sensor fails again from 1200 to 1500 seconds, the system is back in prediction mode. Therefore, the system in this invention can flexibly switch between the three modes, and the system has high adaptability to different navigation conditions.

[0148] Table 1. Output prediction root mean square error (RMSE) of the pseudo-localization navigation error correction method used in rejected environments under different training epochs.

[0149] Training Reincarnations 50 100 200 300 400 Longitude increment prediction error (m) 4153.76 1396.39 24.34 20.90 24.97 Latitude increment prediction error (m) 44.39 5.20 0.86 0.79 1.03 Longitude prediction error (m) 50443.6 16656.2 290.88 246.90 304.96 Latitude prediction error (m) 4425.46 748.57 210.83 227.21 202.27

[0150] Table 1 (continued)

[0151] Training Reincarnations 500 600 800 1000 Longitude increment prediction error (m) 22.75 12.24 21.68 22.33 Latitude increment prediction error (m) 0.99 0.85 0.92 1.06 Longitude prediction error (m) 295.7 145.10 259.52 984.43 Latitude prediction error (m) 192.89 239.56 327.52 357.13

[0152] Table 2. Output Root Mean Square Error (RMSE) of the Adaptive Maintenance and Navigation Error Correction Methods for False Positioning Availability under Different Training-Prediction Ratios in Rejection Environments.

[0153] Training prediction ratio 1:1 5.5：4.5 6:04 7：3 Longitude increment prediction error (m) 9224.46 8719.52 4786.06 2615.12 Latitude increment prediction error (m) 5317.45 3940.22 614.99 77.79 Longitude prediction error (m) 370133.49 165720.94 6033.37 4400.85 Latitude prediction error (m) 394815.23 357277.68 34939.26 8498.66

[0154] Table 2 (continued)

[0155] Training prediction ratio 8:02 9:01 9.5：0.5 9.8：0.2 Longitude increment prediction error (m) 2035.87 78.57 78.99 72.52 Latitude increment prediction error (m) 8.46 2.21 1.88 1.84 Longitude prediction error (m) 3083.50 157.71 110.22 105.48 Latitude prediction error (m) 2677.86 556.53 56.53 55.36

[0156] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in a denied environment, characterized in that, Includes the following steps: S1, obtain the angle increment and velocity increment output by the inertial navigation system; S2, acquire the aircraft's longitude and latitude position increments when the auxiliary sensors are effective; S3, the inertial navigation system enters the training mode, using the normalized angle increment and velocity increment output by the inertial navigation system as input, and the normalized position increment output by the auxiliary sensor as output to train the memory network, and comparing the training error with the expected error, and using the fireworks algorithm to adaptively adjust the parameters of the memory network. S4, when the auxiliary sensor fails, the inertial navigation system enters prediction mode and uses the trained memory network to predict the position increment of the integrated navigation. S5, calculates the position of the aircraft to be located when the auxiliary sensor fails; S6. After the auxiliary sensor recovers, the inertial navigation system enters the verification mode, compares the difference between the memory network and the integrated navigation with the set expected error size, and determines whether it is necessary to continue training the memory network; if it is necessary to continue training the memory network, return to step S3. In step S3, the detailed implementation steps of adaptively adjusting the memory network parameters using the fireworks algorithm are as follows: S31, when the auxiliary sensor is not faulty, take the angle increment and velocity increment output by the inertial navigation system obtained in step S1 as the training input sequence, perform normalization processing, and obtain standardized input data for training. Its expression is: in, for The instantaneous X-axis velocity increment of the carrier for The incremental velocity of the carrier along the Y-axis at any given moment. for The X-axis angle increment of the carrier at any time for The Y-axis angle increment of the carrier at any time for The Z-axis angle increment of the carrier at any given time; Represents the transpose matrix; S32, if the auxiliary sensor is not malfunctioning, take the aircraft's longitude and latitude position increments obtained in step S2 as the training output sequence, perform normalization processing, and obtain standardized output data for training. Its expression is: in, for Increment of longitude position at any time for Increment of latitude and position at time; S33, the inertial navigation system enters training mode, and standardized input data is transferred. Standardized output data as training input As the training output, it establishes the training data framework for the memory network; S34, train the memory network using the gradient-based Adam method, updating values ​​using gradient optimization. for: in, For learning rate, The gradient is the average value at the first time step. Let the gradient be the variance at the second time step. A random number in the range of 0 to 1; S35, define the sequence input layer and determine the size of the input sequence at each time step; S36, Create a memory layer and determine the number of hidden units in the layer; S37, Create a fully connected layer with a specified output size; S38, Create a regression output layer and perform data fitting; S39, Adjust the parameters of the memory network. When the training error exceeds a set threshold, the parameters of the memory network are optimized by selecting the number of optimization operators and the optimization radius. The number of hidden layers and the learning rate in the memory network parameters are adaptively determined and adjusted. The expression is as follows: in, To optimize the training architecture of the memory network, the optimal parameter is the number of hidden layers. and learning rate , To find the optimal number of operators, To find the optimal radius, and To find the optimal range, To adjust the constant number of optimization operators, To adjust the constant of the optimization radius, To prevent the mechanical minimum quantity of division by zero, The initial number of optimization operators; S310 sets the maximum number of learning epochs after the error stabilizes. When the memory network has been trained to the maximum number of learning epochs after the error stabilizes, and the mean absolute error... When all values ​​are less than the set threshold, the memory network stops training. The formula for calculating the mean absolute error is: Here, we assume the current memory network is at position i. layer, Time, from time The input propagation error of the layer is denoted as... ;from time The error propagated from the layer is denoted as ; The number of columns in the training output array is denoted by , and the number of each output state variable is denoted by .

2. The method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in a denied environment according to claim 1, characterized in that, In step S1, the specific steps for obtaining the angle increment and velocity increment output by the inertial navigation system are as follows: S11, acquire the measurement data output by the gyroscope and accelerometer required for aircraft positioning; S12 calculates the output data from the gyroscope and accelerometer, converts it to obtain the carrier angular velocity and carrier acceleration, and stores the carrier angular velocity information output by the inertial navigation system during the aircraft's flight. With carrier acceleration information ; S13, based on the obtained angular velocity and acceleration information of the carrier output by the inertial navigation system, the inertial navigation system... Time and The angle and velocity information output at any given time are subtracted to obtain the aircraft's angle increment and velocity increment output by the inertial navigation system. The calculation formula is as follows: in, For the aircraft system angle increment, For the speed increment of the aircraft system, for The carrier's angular velocity is obtained by measuring and converting data using a gyroscope. for The carrier's angular velocity is obtained by measuring and converting data using a gyroscope. for The acceleration of the carrier is measured and converted by a constant accelerometer. for The acceleration of the carrier is measured and converted by a constant accelerometer. for Time's up The carrier's X-axis angle increment at time t. for Time's up The Y-axis angle increment of the carrier at time t, for Time's up The Z-axis angle increment of the carrier at time t, for Time's up The increment of the carrier's X-axis velocity at time t. for Time's up The increment of the Y-axis velocity of the carrier at any given time.

3. The method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in a denied environment according to claim 1, characterized in that, In step S2, the specific steps for obtaining the aircraft's longitude and latitude position increments when the auxiliary sensors are effective are as follows: S21, Obtain the longitude position of the aircraft measured by the auxiliary sensors. and latitude ; S22, based on the longitude and latitude positions of the aircraft obtained in step S21, adjust the auxiliary sensors. Time and The position information output at any given time is differentially analyzed to obtain the spacecraft's longitude and latitude position increments: in, for The longitude position of the spacecraft at any given time. for The longitude position of the spacecraft at any given time. for The latitude and position of the spacecraft at any given time. for The latitude and position of the spacecraft at any given time.

4. The method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in a denied environment according to claim 1, characterized in that, In step S4, when the auxiliary sensor fails, the system enters prediction mode and uses the trained memory network to predict the position increment of the integrated navigation system. The specific steps are as follows: S41, When the auxiliary sensor fails, the system switches from training mode to prediction mode, and calculates the angle and velocity increments output by the inertial navigation system when the auxiliary sensor fails. The data is fed into a trained memory network for learning, and the network outputs a prediction array. and will De-standardization yields the array of position increments predicted by the memory network. ,in: in, For training output array The mean, For training output array Standard deviation; The number of rows in the training output array; To predict the number of columns in the input and output arrays; S42, Output the array of pseudo-observation information of position increments predicted by the memory network: .

5. The method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in a denied environment according to claim 4, characterized in that, In step S5, the specific steps for calculating the position of the aircraft to be located when the auxiliary sensor fails are as follows: S51, obtain the aircraft's position at the moment the auxiliary sensor signal is lost. The location increment pseudo-observation information predicted by the memory network Accumulate to The position pseudo-observation information calculated after the auxiliary sensor signal is lost is obtained. The expression is as follows: in, To help determine when the sensor signal is lost, For the predicted duration.

6. The method for adaptive maintenance of pseudo-positioning availability and correction of navigation errors in a denied environment according to claim 5, characterized in that, In step S6, the system enters verification mode. If it is necessary to continue training the memory network, the system returns to step S3. The specific steps are as follows: S61, monitor training progress and calculate training residuals Its expression is: in, This refers to the pseudo-observation information of the location increment predicted by the memory network in verification mode. To assist in the incremental navigation results recovered by the sensors, To verify the duration, the prediction frequency of the memory network remains unchanged in the verification mode; S62 will train the residual Compared with the expected prediction error limit, when the training residual If the error falls below the expected prediction limit, training is stopped; otherwise, proceed to step S3 and re-enter training mode.

Images