Multi-source navigation method and system based on LTCs-EKF

Abstract

The invention discloses a multi-source navigation method and system based on LTCs-EKF, and is applied to the fields of inertial navigation enhancement technology, multi-source fusion navigation technology and navigation positioning. The problem that in the prior art, in a complex high-dynamic scene, a sensor loses efficacy, and consequently the fault tolerance and reliability of a system are seriously reduced is solved. The invention designs a multi-source navigation method of Liquid Time-Content Networks (LTCs)-Extended Kalman Filter (EKF), the output error of an original algorithm is compensated by adopting the liquid time constant network, the liquid time constant network is used for training and learning the characteristics of prediction state, updated state estimation, Kalman gain and the like of the Extended Kalman Filter, and the output error of the original algorithm is compensated by adopting the output error of the original algorithm. The error compensation of the original algorithm is output, and finally the compensated attitude is output, so that the fault tolerance and reliability of multi-source navigation are improved. According to the invention, in a high dynamic environment of satellite rejection, deep fusion of the sensor is realized, high-precision navigation and positioning requirements in complex and changeable scenes are met, fault tolerance performance of a multi-source navigation system is improved, and positioning and navigation precision, real-time performance and reliability of the system are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the fields of inertial navigation enhancement technology, multi-source fusion navigation technology and navigation positioning, and particularly relates to a multi-source navigation positioning technology and system. Background Art

[0002] All aspects of human production, life and social development are inseparable from modern navigation technology. The sustainability and reliability of its service capabilities are an important foundation for national security. Currently, satellite navigation systems dominate the world with their advantages of wide coverage and high precision, while other autonomous navigation methods mostly exist as supplements. After the widespread use of the US Global Positioning System (GPS), all countries have recognized the broad prospects of satellite navigation and have developed independent satellite navigation systems, including China's Beidou satellite navigation system, Russia's GLONASS system and the European Union's Galileo satellite navigation system. However, with the rapid development of electronic countermeasures technology, the problems of satellite navigation systems being susceptible to interference and deception, and weak signals under harsh conditions have become increasingly prominent. Especially in the military field, satellite navigation systems face a series of challenges: (1) Due to geographical constraints, my country has relatively few ground tracking and control stations, which are difficult to meet the growing navigation needs; (2) Under wartime conditions, ground stations, as key nodes for navigation signal resolution, are vulnerable to physical destruction or electromagnetic suppression in military confrontation scenarios, which may lead to the risk of global service interruption; (3) The development of satellite deception technology, which misleads targets from their designated locations by sending false satellite signals. To address these issues, inertial navigation, with its advantages of autonomous navigation, has become the core of modern navigation systems. However, due to its own limitations, inertial navigation cannot provide long-term, high-precision navigation services. To improve navigation systems based on inertial and satellite navigation, multi-source navigation systems have been proposed. These systems use multiple sensors to assist and complement each other, compensating for inertial navigation errors, improving system fault tolerance and reliability, and enabling high-precision navigation and positioning services in complex environments. However, how to deeply couple multi-source navigation with inertial navigation's real-time dynamic data and build a more efficient information fusion mechanism remains a key technical bottleneck restricting the improvement of multi-source navigation accuracy. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a multi-source navigation method and system based on LTCs-EKF, which realizes the deep coupling of real-time dynamic data of multi-source navigation and inertial navigation, and improves the navigation accuracy of the system.

[0004] One of the technical solutions adopted by the present invention is: a multi-source navigation system based on LTCs-EKF, comprising: a micro single-board computer module, an FPGA module, an inertial sensor module, other sensor modules, and an external reference attitude measurement module; the FPGA module, the inertial sensor module, other sensor modules, and the external reference attitude measurement module are all connected to the micro single-board computer module;

[0005] The FPGA module collects data from the inertial sensor module and other sensor modules under the control of the micro single-board computer module to obtain multi-source sensor data;

[0006] The external reference navigation measurement module is used to output reference navigation data;

[0007] The micro single-board computer module includes a solving unit, an EKF-based multi-source fusion unit, an LTCs-based error compensation unit, and an output unit. The solving unit solves the multi-source sensor data collected by the FPGA; the EKF-based multi-source fusion unit outputs a multi-source navigation estimation value based on the solving result; the LTCs-based error compensation unit uses the multi-source navigation estimation value output by the EKF-based multi-source fusion unit as training data, and uses the reference navigation data output by the external reference navigation measurement module as the true value for iterative training; the output unit compensates the output of the EKF-based multi-source fusion unit based on the output of the trained LTCs, thereby outputting a compensated multi-source navigation result.

[0008] The second technical solution adopted by the present invention is: a multi-source navigation method based on LTCs-EKF, comprising:

[0009] S1, collecting multi-source sensor data and solving the multi-source sensor data;

[0010] S2. Construct an LTCs-EKF network. Specifically, the EKF performs prediction based on the data obtained in step S1 to obtain a multi-source navigation prediction result. The LTCs input is obtained based on the multi-source navigation prediction result, and the LTCs output value is used as the error compensation value of the EKF output Euler angle.

[0011] S3, training the LTCs-EKF network of step S2 based on the solution result of step S1;

[0012] S4. Compensate the multi-source navigation estimation value of the current iteration of the EKF with the output value of the current iteration of the LTCs in the trained LTCs-EKF network to obtain compensated multi-source navigation data.

[0013] Beneficial effects of the present invention: The positioning system of the present invention uses low-cost MEMS sensors and designs a multi-source navigation hardware and software system. When a sensor fails, it can ensure system stability, improve the system's fault tolerance and reliability, and meet the high-precision positioning requirements of complex scenarios;

[0014] The present invention adopts a multi-source fusion navigation method based on LTCs-EKF and introduces a liquid time constant network to compensate for the output error of the original algorithm. The liquid time constant network expands the predicted state of the Kalman filter, the updated state estimation, the Kalman gain and other characteristics through training and learning, and outputs the error compensation of the original algorithm, and finally outputs the compensated attitude, thereby improving the navigation accuracy of the multi-source navigation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a multi-source navigation system framework based on LTCs-EKF;

[0016] Figure 2 It is a multi-source navigation algorithm framework based on LTCs-EKF network;

[0017] Figure 3 For the time synchronization process;

[0018] Figure 4 Drive workflows for sensors;

[0019] Figure 5 For data communication workflow;

[0020] Figure 6 LTCs structure of multi-source navigation algorithm based on LTCs-EKF network;

[0021] Figure 7 A multi-source navigation training framework based on LTCs-EKF network;

[0022] Figure 8 The training process of the neural network;

[0023] Figure 9 A chart comparing the navigation accuracy of the method of the present invention and the prior art. DETAILED DESCRIPTION

[0024] To facilitate those skilled in the art to understand the technical content of the present invention, the present invention is further explained below with reference to the accompanying drawings.

[0025] like Figure 1As shown, a multi-source fusion navigation system based on LTCs-EKF of the present invention includes: a micro single-board computer module, an FPGA module, an inertial sensor module, other sensor modules, an external reference attitude measurement module and a power module; the FPGA module, the inertial sensor module, the other sensor modules and the external reference attitude measurement module are all connected to the micro single-board computer module, and the power module is used to supply power to the micro single-board computer module, the FPGA module, the inertial sensor module, the other sensor modules and the external reference attitude measurement module.

[0026] Other sensor modules in this embodiment include: a satellite receiver module, a star sensor module, a magnetometer module, and a barometric altimeter module;

[0027] The micro single-board computer module is the computing core of the system and is used to solve the navigation algorithm.

[0028] FPGA module: used for sensor data acquisition, time synchronization and fault detection.

[0029] Data communication between modules is realized based on the communication interface.

[0030] External reference navigation measurement module: In order to evaluate the system accuracy, considering that it is difficult to use equipment such as turntables to accurately control the attitude under outdoor conditions, the system requires a reference navigation measurement module and uses the attitude output by the external reference navigation measurement module as a reference.

[0031] The navigation system of this embodiment achieves high-precision, high-real-time, passive and autonomous deep-fusion multi-source navigation by calculating the angular velocity and acceleration information collected by the inertial sensor, the position information received by the satellite receiver, the star image information collected by the star sensor, the magnetic field information collected by the magnetometer, and the altitude information collected by the barometric altimeter.

[0032] Because the traditional EKF-based multi-source fusion navigation algorithm is limited by the local linearization and noise distribution of EKF, it cannot be effectively modeled. Moreover, when the system fails, the effective parameters are reduced and high-precision positioning services cannot be provided. Since LTCs not only have the ability to process time series data, but also have adaptive calculations, and also introduce dynamic time constant characteristics, they can effectively adapt to the changes in sensor noise over time to improve the navigation and positioning accuracy and fault tolerance of the system. Therefore, LTCs are used to compensate for the output errors of the EKF-based multi-source navigation algorithm. The framework of the multi-source navigation algorithm based on the LTCs-EKF network is as follows: Figure 2 shown.

[0033] Therefore, the present invention also provides a navigation method based on the above navigation system, comprising:

[0034] S1, such as Figure 3 As shown, the sensor data is collected, the collected sensor data is solved, and the collected raw data is time-synchronized and corrected; specifically: the collected data signal is sent to each sensor driver according to the sampling frequency, and the driver reads the sensor information after receiving the signal and transmits the information to the navigation algorithm;

[0035] The configuration and information reading process of each sensor is as follows Figure 4 As shown, this process is a known technology and will not be elaborated in detail in the present invention.

[0036] Data communication process such as Figure 5 As shown, the data communication process ensures that the initialization, data collection, data synchronization and data output of the sensor are carried out in an orderly manner. The data communication process is a well-known technology and will not be elaborated in detail in the present invention.

[0037] like Figure 1 As shown, the collected sensor data is solved: inertial sensor data is solved to obtain attitude angle, speed, and position information; magnetometer data is solved to obtain heading angle information; satellite navigation data is solved to obtain speed and position information; celestial navigation data is solved to obtain attitude angle information; and barometric altimeter data is solved to obtain altitude information. The specific solution process is a known technology and will not be elaborated in detail in the present invention.

[0038] S2. Construct an LTCs-EKF network, and train the constructed LTCs-EKF network based on the data collected in step S1, so as to obtain accurate multi-source navigation results. Specifically, the EKF makes predictions based on the data collected in step S1, and uses the estimated multi-source navigation estimation value as the input of the LTCs to train the LTCs, and then uses the output value of the LTCs as the error compensation value of the Euler angle output by the EKF, so as to finally obtain accurate multi-source navigation results.

[0039] S21, construct the state model and observation model of EKF;

[0040] S211. The process of constructing the state model is:

[0041] The INS array state model is established, and filtering modeling is performed based on the INS error model. The state error of the multi-source navigation system is expressed as:

[0042]

[0043] Where: is the system state error, X(t) is the system state vector, t represents the time variable; A(t) is the state coefficient matrix; G(t) is the error coefficient matrix; W(t) is the white noise random error vector.

[0044] The system state vector X is:

[0045]

[0046] It includes the basic navigation parameter error of the 9-dimensional inertial navigation system and the error state of the 9-dimensional inertial instrument. is the platform error angle, δV E δV N δV U Velocity error in the northeast direction, δLδλδh Latitude, longitude, and altitude position errors, ε bx ε by ε bz is the gyroscope random constant, ε rx ε ry ε bz is a gyroscope first-order Markov process, Accelerometer first-order Markov process.

[0047] S212. The process of constructing the observation model is:

[0048] Establish the system observation equation, use the difference of the system error as the observation quantity, and establish the observation equation of each combination. The sensor observation model can be expressed as:

[0049] Z k =H k X k +v k

[0050] Where: Z k represents the observation vector; H k represents the observation matrix; X k represents the system state vector at time k; v k represents the observation noise vector.

[0051] The INS / GNSS position and velocity measurement error equations are expressed as:

[0052]

[0053] Where Z V =[V INS -V GNSS ],Z P =[P INS -P GNSS ],V INS ,V GNSS are the inertial sensor velocity observation value and the satellite navigation velocity observation value, respectively, P INS ,P GNSS are the inertial sensor position observation value and the satellite navigation position observation value respectively; H V 、HP 、V V 、V P They represent the velocity observation matrix, position observation matrix, velocity observation noise vector, and position observation noise vector in INS / GNSS respectively; H and V represent the observation matrix and observation noise vector respectively.

[0054] INS / celestial navigation system (CNS) attitude angle observation equation:

[0055] The attitude measurement of CNS is used as the attitude observation quantity in the multi-source navigation system, that is, the difference between the roll angle, pitch angle and heading angle information given by the inertial navigation system and the corresponding information given by the celestial navigation system. The observation equation is:

[0056] Z A (t)=[A INS -A CNS ]=H A (t)X+V A (t)

[0057] Where A INS ,A GNSS are the inertial sensor attitude angle observation value and the CNS attitude angle observation value respectively; H A (t), V A (t) represents the INS / celestial navigation system (CNS) attitude angle observation matrix and observation noise vector, respectively.

[0058] INS / Mag magnetometer (Magnetic) heading angle observation equation:

[0059] In the combined navigation system of the inertial navigation system and the magnetometer, there is one set of observation quantities, namely the heading angle observation quantity, which is the difference between the heading angle information given by the inertial navigation system and the corresponding information given by the magnetometer.

[0060] Z a (t) Mag =[ψ INS -ψ Mag ]=H a (t)X(t)+V a (t)

[0061] Where, ψ INS is the heading angle observation value of the inertial navigation system; ψ Mag is the magnetometer heading angle observation value; H a (t), V a(t) represents the INS / Mag magnetometer heading angle observation matrix and observation noise vector respectively.

[0062] INS / barometric altimeter altitude measurement error equation:

[0063] In the combined navigation system of the inertial navigation system and the barometric altimeter, there is one set of observation quantities, namely the altitude observation quantity, which is the difference between the altitude information given by the inertial navigation system and the corresponding information given by the barometric altimeter.

[0064] Z p (t) Baro =[h INS -h Baro ]=H p (t)X(t)+V p (t)

[0065] Where h INS is the altitude observation value of the inertial navigation system, h Baro is the barometric altimeter altitude observation value, H p (t), V p (t) represent the INS / barometric altimeter altitude observation matrix and the observation noise vector, respectively.

[0066] S22, EKF performs prediction based on the data obtained in step S1; the specific process is as follows:

[0067] initialization:

[0068] Among them, x0 represents the initial state quantity, that is, the initial state of the carrier in the system state vector X; represents the state estimate of x0, E[·] represents the expectation; P0 represents the initial error covariance matrix;

[0069] (1) Time update:

[0070] 1) Prediction status

[0071]

[0072] Where f1(·) represents the nonlinear state transfer function, is the predicted state at time k, is the posterior estimate at time k-1;

[0073] 2) Calculate the state transfer Jacobian matrix:

[0074]

[0075] Among them, F k-1 Indicates Calculate the state transfer Jacobian matrix at Represents X k-1 state estimation;

[0076] 3) Covariance prediction:

[0077]

[0078] Among them, Γ k-1 Process noise allocation matrix, Q k-1 process noise covariance matrix;

[0079] (2) Forecast update:

[0080] 1) Calculate the observation Jacobian matrix:

[0081]

[0082] Among them, H k For Calculate the observation Jacobian matrix at h(·) nonlinear observation function;

[0083] 2) Calculate the Kalman gain:

[0084]

[0085] Among them, K k Kalman gain matrix, R k Observation noise covariance matrix;

[0086] 3) Status Update:

[0087]

[0088] is the updated state estimate;

[0089] 4) Covariance update:

[0090] P k =(IK k H k )P k|k-1

[0091] Among them, P k represents the updated error covariance matrix, and I represents the identity matrix;

[0092] After the processing of step S22, the multi-source navigation estimation value can be output and input into the neural network for training to compensate the multi-source navigation result and improve the navigation accuracy and fault tolerance performance.

[0093] S23, training the LTCs neural network based on the multi-source navigation estimation value obtained in step S22;

[0094] The Kalman gain K includes accelerometer measurements and measurement and control components for satellite navigation, star sensors, altimeters, and magnetometers. Due to differences in the output rates of other sensors and inertial units or sensor failure, some sampling points may contain output data from all sensors. In such cases, the Kalman gain K lacks data. Therefore, you need to manually append a zero matrix to ensure consistent input length.

[0095] The output value of the neural network is the error compensation value of the EKF output Euler angle, so the final output is:

[0096] θ'=θ+δθ

[0097] V′=V+δV

[0098] P′=P+δP

[0099] In order to ensure both the learning rate and fitting ability of LTCs, the size of the hidden layer is set to 64 neurons. Therefore, the LTCs structure of the multi-source navigation algorithm based on the LTCs-EKF network is as follows: Figure 6 As shown in the figure, the LTCs neural network consists of three parts: input layer, hidden layer and output layer.

[0100] Input layer: pose estimate Speed ​​estimate Position estimate The attitude correction value θ, velocity correction value V, position correction value P, and Kalman gain K are input into the neural network as LTCs;

[0101] Hidden layer: set to 64 neurons;

[0102] Output layer: output angle compensation value δθ, speed compensation value δV and position compensation value δP.

[0103] For the training phase, the output of the external reference system is used as the true value, and the rest is the same as the prediction phase, such as Figure 7 As shown, the error compensation values ​​of the EKF output attitude angle, velocity, and position used to train the neural network are:

[0104] δθ=θ'-θ

[0105] δV=V′-V

[0106] δP=P′-P

[0107] The construction and training of neural networks uses the TensorFlow machine learning framework. TensorFlow is an open-source and free machine learning platform with a complete and flexible ecosystem of tool libraries and community resources. It provides a large number of application programming interfaces (APIs). TensorFlow can be used on various operating systems, such as Windows and Linux, and on various devices, including embedded devices and computers, ensuring strong compatibility. Therefore, this part was implemented using TensorFlow.

[0108] The data for neural network training, verification, and testing are collected by the multi-source navigation system of LTCs-EKF.

[0109] First, the data is divided into training set, validation set, and test set. The training process of the neural network is as follows: Figure 8 shown.

[0110] The LTCs network based on ordinary differential equations (ODE) can be expressed as:

[0111]

[0112] Among them, h t In hidden state, dh t / dt is the derivative of the hidden state, τ is the time constant, f2(·) is the activation function, x t is the neural network input, A and θ are the LTCs network parameters;

[0113] This equation can also be written to include the system time constant:

[0114]

[0115] In LTCs, neurons are not only used to calculate the hidden state h t The derivative dh t / dt, and also calculate the time constant τ of the system sys In LTCs, the time constant is not fixed, leading to the term "liquid time constant network," where "liquid" refers to the system's variable time constant. The time constant is a key parameter controlling the speed and coupling sensitivity of ODEs. Therefore, the dynamic system of feature recognition in LTCs is different at each time point, enabling the network's parameters to be adjusted during operation.

[0116] (1) Forward propagation

[0117] The forward propagation of LTCs inevitably requires solving the ODE of the hidden state. Therefore, this embodiment uses a hybrid Euler method ODE solver. The update algorithm of the forward propagation is as follows:

[0118]

[0119] Among them, ⊙ represents element multiplication, and FuseODESolver represents the program pseudocode statement calling the function.

[0120] (2) Training

[0121] To ensure the accuracy of LTCs calculation, the time series back propagation algorithm is used. The algorithm steps are as follows:

[0122]

[0123] Among them, l() is the single-step loss function, x(t) represents the multi-source navigation fusion result output by EKF; y(t) represents the difference between the external reference navigation result and the output of the EKF algorithm; Indicates the prediction error compensation value output by LTCs.

[0124] S3. The output of the trained LTCs-EKF network is used as the final multi-source navigation result.

[0125] As shown in Table 1 and Figure 9 As shown, the method of the present invention significantly improves navigation accuracy compared to the prior art.

[0126] Table 1 Comparison of navigation accuracy between the method of the present invention and the prior art

[0127] algorithm Attitude angle RMSE Speed ​​RMSE Position RMSE Traditional EKF multi-source fusion navigation algorithm 0.8° 1m / s 2m Multi-source fusion navigation method of LTCs-EKF 0.2° 0.5m / s 1m

[0128] Those skilled in the art will appreciate that the embodiments described herein are intended to aid understanding of the principles of the present invention, and it should be understood that the scope of the present invention is not limited to such specific descriptions and embodiments. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, and the like made within the spirit and principles of the present invention are intended to be included within the scope of the claims.

Claims

1. A multi-source navigation system based on LTCs-EKF, characterized by: include: A micro single-board computer module, an FPGA module, an inertial sensor module, other sensor modules, and an external reference attitude measurement module; the FPGA module, the inertial sensor module, other sensor modules, and the external reference attitude measurement module are all connected to the micro single-board computer module; The FPGA module collects data from the inertial sensor module and other sensor modules under the control of the micro single-board computer module to obtain multi-source sensor data; The external reference navigation measurement module is used to output reference navigation data; The micro single-board computer module includes a solution unit, an EKF-based multi-source fusion unit, an LTCs-based error compensation unit, and an output unit. The solution unit solves the multi-source sensor data collected by the FPGA. The EKF multi-source fusion unit outputs multi-source navigation estimation values ​​based on the solution results; The error compensation unit of the LTCs uses the multi-source navigation estimation value output by the EKF multi-source fusion unit as training data, and uses the reference navigation data output by the external reference navigation measurement module as the true value for iterative training; The output unit compensates the output of the EKF multi-source fusion unit based on the output of the trained LTCs, thereby outputting a compensated multi-source navigation result.

2. The multi-source navigation system based on LTCs-EKF according to claim 1, characterized in that: Other sensor modules include, but are not limited to, satellite receiver modules, star sensor modules, magnetometer modules, and barometric altimeter modules.

3. A multi-source navigation method based on LTCs-EKF, characterized in that: include: S1, collecting multi-source sensor data and solving the multi-source sensor data; S2. Construct an LTCs-EKF network. Specifically, the EKF performs prediction based on the data obtained in step S1 to obtain a multi-source navigation prediction result. The LTCs input is obtained based on the multi-source navigation prediction result, and the LTCs output value is used as the error compensation value of the EKF output Euler angle. S3, training the LTCs-EKF network of step S2 based on the solution result of step S1; S4. Compensate the multi-source navigation estimation value of the current iteration of the EKF with the output value of the current iteration of the LTCs in the trained LTCs-EKF network to obtain compensated multi-source navigation data.

4. The multi-source navigation method based on LTCs-EKF according to claim 3, characterized in that: The multi-source sensor data in step S1 includes: inertial sensor data, magnetometer data, satellite navigation data, celestial navigation data, and barometric altimeter data.

5. The multi-source navigation method based on LTCs-EKF according to claim 4, characterized in that: Step S3 is specifically as follows: S31, constructing a multi-source navigation system state error model based on the solution result of step S1; S32. Constructing observation equations for different combined navigations based on the multi-source navigation system state error model of step S31; S33, bringing the data obtained from the calculation in step S1 into the observation equations of different combined navigation constructed in step S32, and performing multi-source navigation prediction based on the EKF algorithm; S34, constructing a training data set according to the multi-source navigation prediction result output by the EKF algorithm and the difference between the multi-source navigation prediction result output by the EKF algorithm and the external reference navigation result; S35 . Training LTCs based on the training data set constructed in step S34 .

Images