CN114689047B - Deep learning-based integrated navigation method, device, system and storage medium - Google Patents

Deep learning-based integrated navigation method, device, system and storage medium Download PDF

Info

Publication number
CN114689047B
CN114689047B CN202210616148.2A CN202210616148A CN114689047B CN 114689047 B CN114689047 B CN 114689047B CN 202210616148 A CN202210616148 A CN 202210616148A CN 114689047 B CN114689047 B CN 114689047B
Authority
CN
China
Prior art keywords
gps
ins
deep learning
value
learning model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210616148.2A
Other languages
Chinese (zh)
Other versions
CN114689047A (en
Inventor
崔金强
刘耀华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Peng Cheng Laboratory
Original Assignee
Peng Cheng Laboratory
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Peng Cheng Laboratory filed Critical Peng Cheng Laboratory
Priority to CN202210616148.2A priority Critical patent/CN114689047B/en
Publication of CN114689047A publication Critical patent/CN114689047A/en
Application granted granted Critical
Publication of CN114689047B publication Critical patent/CN114689047B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/10—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration
    • G01C21/12—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning
    • G01C21/16—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning by integrating acceleration or speed, i.e. inertial navigation
    • G01C21/165—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning by integrating acceleration or speed, i.e. inertial navigation combined with non-inertial navigation instruments
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/20—Instruments for performing navigational calculations
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
    • G01S19/38—Determining a navigation solution using signals transmitted by a satellite radio beacon positioning system
    • G01S19/39—Determining a navigation solution using signals transmitted by a satellite radio beacon positioning system the satellite radio beacon positioning system transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
    • G01S19/42—Determining position
    • G01S19/48—Determining position by combining or switching between position solutions derived from the satellite radio beacon positioning system and position solutions derived from a further system
    • G01S19/49—Determining position by combining or switching between position solutions derived from the satellite radio beacon positioning system and position solutions derived from a further system whereby the further system is an inertial position system, e.g. loosely-coupled

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)
  • Navigation (AREA)

Abstract

The invention discloses a combined navigation method, a device, a system and a storage medium based on deep learning, wherein the method comprises the following steps: an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system; deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of the deep learning model; constructing a deep learning model and a loss function thereof, and learning a nonlinear relation between an input value and an output value; respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model; and carrying out GPS interruption test, generating lost GPS position data, and calibrating the output value of the INS according to the lost data to obtain navigation data after GPS interruption. The method combines the Kalman filter with deep learning, and improves the robustness and the precision of INS and GPS combined navigation.

Description

Deep learning-based integrated navigation method, device, system and storage medium
Technical Field
The invention relates to the technical field of integrated navigation, in particular to an integrated navigation method, device and system based on deep learning and a storage medium.
Background
In recent years, with the development of artificial intelligence technology, mobile robots are applied to executing various intelligent tasks in more and more complex environments, such as security and disinfection of epidemic situation parks, material distribution, autonomous inspection and the like. The prerequisite for the mobile robot to execute these advanced intelligent tasks is to clarify the position and posture of the robot under the world coordinate system of the environment where the robot is located, i.e. the positioning and navigation of the mobile robot.
The most common Positioning and Navigation methods for unmanned vehicles and unmanned vehicles include Inertial Navigation Systems (INS) and Global Positioning Systems (GPS). In the INS, a gyroscope and an accelerometer sensor are used for acquiring angular velocity and acceleration information of a carrier, and the attitude, position and velocity of the carrier are further obtained through integral operation. The inertial navigation system has the advantages of quick updating frequency and high concealment, but has integral accumulated errors. The GPS determines the absolute position of a carrier under a GPS coordinate system by calculating the distance between a satellite and a GPS receiver, and has no accumulated error, but the updating frequency of the GPS is slow and is easy to be interfered. Therefore, in an environment where GPS signals are available, an INS and GPS integrated navigation system formed by integrating an INS and a GPS together is generally used with a kalman filter. However, when the unmanned aerial vehicle or unmanned vehicle equipped with GPS passes through a weak GPS scene, such as a tunnel or a tall building group, the kalman filter cannot update the integrated navigation system using GPS information as an observation value, the INS and GPS integrated navigation system will fall back to the pure INS navigation system, and the inherent noise of the inertial device will cause the precision of the pure INS to rapidly decrease. Therefore, it is of great significance to explore how to improve the combined navigation accuracy of the INS and the GPS in the scene of the GPS short-time interruption.
Therefore, the prior art has yet to be improved.
Disclosure of Invention
The invention provides a combined navigation method, device, system and storage medium based on deep learning, aiming at solving the technical problem of low navigation precision of an INS and GPS combined navigation system under the condition of GPS signal interruption.
The technical scheme adopted by the invention for solving the technical problem is as follows:
in a first aspect, the present invention provides a combined navigation method based on deep learning, including:
an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system;
deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of a deep learning model;
constructing the deep learning model and a loss function thereof, and learning a nonlinear relation between the input value and the output value;
respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model;
and carrying out GPS interruption test by using the trained deep learning model, generating lost GPS position data, and calibrating the output value of the INS according to the lost GPS position data to obtain navigation data after GPS interruption.
In one implementation, the constructing an INS and GPS integrated navigation system by fusing an INS and a GPS through a kalman filter includes:
fusing the INS and the GPS through the Kalman filter;
constructing an error model of the INS;
taking the position increment of the GPS as an observed value of the INS and GPS combined navigation system;
and determining the state vector of the INS and GPS integrated navigation system, and constructing a state equation and an observation equation of the INS and GPS integrated navigation system.
In one implementation, the constructing an error model for an INS includes:
constructing an INS attitude error equation system:
Figure 504543DEST_PATH_IMAGE001
wherein the content of the first and second substances,
Figure 562629DEST_PATH_IMAGE002
an attitude angle error vector is obtained;
Figure 873525DEST_PATH_IMAGE003
and
Figure 924526DEST_PATH_IMAGE004
the rotation angular velocity and the error of the navigation coordinate system relative to the inertial coordinate system are respectively;
Figure 4478DEST_PATH_IMAGE005
a gyroscope drift vector under a navigation coordinate system;
constructing an INS speed error equation set:
Figure 92520DEST_PATH_IMAGE006
wherein the content of the first and second substances,
Figure 766078DEST_PATH_IMAGE007
is oriented from east to,Speed errors in north and sky;
Figure 230557DEST_PATH_IMAGE008
speeds in the east, north and sky directions;
Figure 165015DEST_PATH_IMAGE009
is a specific force value;
Figure 315636DEST_PATH_IMAGE010
and
Figure 335544DEST_PATH_IMAGE011
respectively the rotation speed of the earth in a navigation coordinate system and the angle relative to the earth;
Figure 603715DEST_PATH_IMAGE012
is the deviation of the accelerometer in the navigation coordinate system;
constructing an INS position error equation system:
Figure 268045DEST_PATH_IMAGE013
wherein the content of the first and second substances,
Figure 229048DEST_PATH_IMAGE014
，
Figure 736253DEST_PATH_IMAGE015
and
Figure 667169DEST_PATH_IMAGE016
longitude error, latitude error and altitude error respectively;
Figure 576219DEST_PATH_IMAGE017
、
Figure 317910DEST_PATH_IMAGE018
and
Figure 312411DEST_PATH_IMAGE019
respectively a north direction speed error, a sky direction speed error and an east direction speed error;
Figure 656804DEST_PATH_IMAGE020
and
Figure 312039DEST_PATH_IMAGE021
respectively a meridian curvature radius of the map and a curvature radius of the main vertical line.
In one implementation, the determining a state vector of the INS and GPS integrated navigation system, and constructing a state equation and an observation equation of the INS and GPS integrated navigation system include:
selecting a plurality of system variables to form a state vector of the INS and GPS combined navigation system:
Figure 349265DEST_PATH_IMAGE022
wherein the content of the first and second substances,
Figure 831062DEST_PATH_IMAGE023
is an attitude angle error vector in a navigation coordinate system;
Figure 854513DEST_PATH_IMAGE024
is the speed error;
Figure 738155DEST_PATH_IMAGE025
for longitude errors, latitude errors and altitude errors,
Figure 946283DEST_PATH_IMAGE026
and
Figure 774430DEST_PATH_IMAGE027
respectively representing accelerometer deviation and gyro deviation in a machine body coordinate system;
and constructing a discrete state space equation of the INS and GPS integrated navigation system based on the error model of the INS:
Figure 726206DEST_PATH_IMAGE028
wherein the content of the first and second substances,
Figure 464354DEST_PATH_IMAGE029
representing a state transition matrix;
Figure 718749DEST_PATH_IMAGE030
representing a system noise distribution matrix;
Figure 175139DEST_PATH_IMAGE031
and
Figure 665026DEST_PATH_IMAGE032
process noise and measurement noise, respectively;
Figure 872061DEST_PATH_IMAGE033
is a measurement matrix;
Figure 421991DEST_PATH_IMAGE034
is a measurement vector.
In one implementation, the deriving a relationship between a position increment of the GPS and an output value of the INS and determining an input value and an output value of a deep learning model includes:
substituting the discrete state space equation into the Kalman filter to carry out cyclic update iteration;
and comparing the position increment of the GPS with the output value of the INS through the Kalman filter, and deducing the relation between the position increment of the GPS and the output value of the INS so as to determine the input value and the output value of the deep learning model.
In one implementation, the building a deep learning model and a loss function thereof, and learning a nonlinear relationship between the input value and the output value includes:
constructing the deep learning model;
determining a representation form of the output value of the INS and the position increment of the GPS in feature mapping, and establishing a nonlinear relation between an input layer and an output layer of the deep learning model;
wherein, the output value of the INS comprises an attitude angle value, a specific force value and an angular velocity value.
In one implementation, the building a deep learning model includes:
and taking the attitude angle value, the specific force value and the angular velocity value of the INS as input values of the deep learning model, and taking the position increment value of the GPS as an output value of the deep learning model.
In one implementation, the determining a representation of the INS's output values and the GPS's position deltas in a feature map, establishing a non-linear relationship between an input layer and an output layer of the deep-learning model, includes:
learning sparse features through a convolutional layer by adopting the convolutional layer as a feature extractor;
and modeling an implicit dependency relationship between the INS and the GPS, providing an abstract representation of an output value of the INS in a feature map, and modeling a time dynamic characteristic activated by a feature map by utilizing a GRU layer.
In one implementation, the performing the simulation experiment and the real scene experiment respectively, and adjusting the hyper-parameter of the deep learning model to obtain the trained deep learning model includes:
respectively carrying out the simulation experiment and the real scene experiment;
collecting a data set of the INS and GPS integrated navigation system, and dividing the data set according to a preset proportion to obtain the training set and the test set;
training the deep learning model through the training set, and testing the deep learning model through the testing set;
and adjusting the hyper-parameters of the deep learning model to obtain the trained deep learning model.
In one implementation, the performing a GPS outage test using the trained deep learning model to generate lost GPS location data, and calibrating the output value of the INS according to the lost GPS location data includes:
inputting the attitude angle value, the specific force value and the angular velocity value measured by the INS into the trained deep learning model to generate the lost GPS position data:
Figure 241042DEST_PATH_IMAGE035
wherein the content of the first and second substances,
Figure 534621DEST_PATH_IMAGE036
is the initial position value at the initial time k when the GPS is interrupted,
Figure 981782DEST_PATH_IMAGE037
is the final predicted GPS value;
substituting the lost GPS position data into the Kalman filter, and updating an error covariance matrix to obtain the calibrated INS output value;
and fusing the lost GPS position data and the calibrated INS output value to obtain navigation data after the GPS is interrupted.
In a second aspect, the present invention further provides a deep learning-based integrated navigation device, including: a processor and a memory, the memory storing a deep learning based integrated navigation program, the deep learning based integrated navigation program being executed by the processor for implementing the deep learning based integrated navigation method according to the first aspect.
In a third aspect, the present invention further provides a deep learning-based integrated navigation system, including: a GPS sensor, an INS sensor, a Kalman filter and a combined navigation device based on deep learning according to the second aspect;
the GPS sensor is used for acquiring position data;
the INS sensor is used for acquiring an attitude angle value, a specific force value and an angular velocity value;
the Kalman filter is used for fusing the GPS sensor and the INS sensor to obtain an INS and GPS integrated navigation system; the data acquisition module is used for respectively comparing the data of the GPS sensor with the data of the INS sensor to obtain data difference values of the sensors; the system comprises a GPS module, a correction module and a processing module, wherein the GPS module is used for acquiring GPS position data;
the integrated deep learning based navigation device is used for executing the following steps:
an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system;
deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of a deep learning model;
constructing the deep learning model and a loss function thereof, and learning a nonlinear relation between the input value and the output value;
respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model;
and carrying out GPS interruption test by using the trained deep learning model, generating lost GPS position data, and calibrating the output value of the INS according to the lost GPS position data to obtain navigation data after GPS interruption.
In a fourth aspect, the present invention further provides a storage medium, which is a computer-readable storage medium, and the storage medium stores a deep learning based integrated navigation program, and the deep learning based integrated navigation program is used for implementing the deep learning based integrated navigation method according to the first aspect when executed by a processor.
The invention adopts the technical scheme and has the following effects:
the invention provides a novel INS and GPS neural network framework (GI-NN), which utilizes 4 layers of one-dimensional convolution layers to map IMU input to a high-dimensional space, thereby better extracting IMU characteristics from IMU noise, adopts a gate control recursion unit network to excavate IMU time sequence internal relation, reduces operation load, combines a Kalman filter and deep learning, predicts a GPS position increment value by utilizing a deep learning model, realizes that the Kalman filter can still fuse INS and GPS data under the condition of GPS signal interruption, and improves the robustness and precision of INS and GPS combined navigation.
Drawings
In order to more clearly illustrate the embodiments or technical solutions of the present invention, the drawings used in the embodiments or technical solutions of the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the structures shown in the drawings without creative efforts.
FIG. 1 is a flow chart of a deep learning based integrated navigation method in one implementation of the present invention.
FIG. 2 is a workflow diagram of a deep learning based integrated navigation framework in one implementation of the invention.
FIG. 3 is a schematic diagram of a GI-NN network model structure in an implementation manner of the present invention.
Fig. 4 is a functional schematic diagram of a deep learning based integrated navigation device in one implementation of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Exemplary method
At present, methods for improving the combined navigation accuracy of the INS and the GPS under the condition of short-time GPS interruption can be divided into two categories, one is to add an additional sensor to provide an observation value of a system for a kalman filter, such as a laser radar, a camera, a odometer and the like, and the methods can effectively improve the navigation accuracy, but increase the complexity and the cost of the system. The other is an algorithm based on machine learning, such as an adaptive neural fuzzy inference system, a support vector machine, a random forest and the like, and with the development of artificial intelligence technology, an artificial neural network is also used for improving the combined navigation accuracy of the INS and the GPS. Some research organizations have built a neural network learning model that maps the position error between the INS and the GPS using radial basis function neural networks. Some research institutes have explored a combined method of INS and GPS based on artificial neural networks to fuse uncompensated INS measurements with differential GPS measurements.
Various neural networks, including radial basis function neural networks, back-propagation neural networks, forward back-propagation neural networks, full back-propagation neural networks, adaptive resonance theory-back-propagation neural networks, convolutional neural networks, higher order neural networks, and input delay neural networks have been used to predict INS position errors in environments where GPS signals are unavailable. However, most of the above methods are based on static neural networks, which only use the current and previous step INS information, and do not consider more historical vehicle dynamics information. It is known that it is difficult to obtain an accurate position from time series data due to the lack of a priori dynamic data of the vehicle. The above method cannot maintain high navigation accuracy when the GPS signal is unavailable for a long time.
In recent years, deep learning has achieved satisfactory performance in time series prediction, i.e., natural language processing and speech processing. Deep learning algorithms such as recurrent neural networks have many advantages over machine learning methods in processing time series data and nonlinear system modeling due to their special recurrent units. The stochastic drift of the MEMS gyroscope is estimated and compensated in real time by combining a recurrent neural network with an unscented kalman filter, however, the recurrent neural network may exhibit gradient vanishing and gradient explosion when processing long time sequences.
In view of the above technical problems, the present embodiment provides a combined navigation method based on deep learning, which maps IMU inputs to a high-dimensional space by using 4 layers of one-dimensional convolutional layers, so as to better extract IMU features from IMU noise, and uses a gated recursion unit network to mine the intrinsic relation of IMU time sequences, thereby reducing the computational load, and simultaneously combines a kalman filter and deep learning, predicts a GPS position increment value by using a deep learning model, so that the kalman filter can still fuse INS and GPS data under the condition of GPS signal interruption, and the robustness and accuracy of the combined navigation of INS and GPS are improved.
As shown in fig. 1, an embodiment of the present invention provides a deep learning-based integrated navigation method, including the following steps:
and S100, fusing the INS and the GPS through a Kalman filter to construct an INS and GPS integrated navigation system.
In the embodiment, the integrated navigation method based on deep learning is realized by an INS and GPS integrated navigation system; the INS and GPS integrated navigation system is a combination of traditional INS navigation and GPS navigation, and the traditional INS (namely an inertial navigation system) and the GPS (namely a global positioning system) can be fused through the Kalman filter to construct a loosely coupled integrated navigation system, so that the INS and GPS integrated navigation system is obtained.
Further, in the process of constructing the INS and GPS integrated navigation system, an error model of the INS needs to be determined, that is, the error degree of data acquired by the INS is determined; then, with the error model of the INS and the position increment of the GPS as references, constructing and obtaining a state equation and an observation equation of the INS and GPS combined navigation system; the error model of the INS may be an error model obtained according to an error between actual data and standard data acquired by the INS within a period of time (e.g., 1s, 5s, 10s, etc.); the position increment of the GPS may also be a position difference between position data acquired by the GPS at the current time and position data acquired at a previous time within a period of time.
Specifically, in one implementation manner of the present embodiment, the step S100 includes the following steps:
step S101, fusing the INS and the GPS through the Kalman filter;
step S102, constructing an error model of the INS;
step S103, using the position increment of the GPS as an observed value of the INS and GPS integrated navigation system;
and step S104, determining the state vector of the INS and GPS integrated navigation system, and constructing a state equation and an observation equation of the INS and GPS integrated navigation system.
In this embodiment, according to the dynamic error model equation set of the INS, the GPS is used as a system observation value to construct a state equation set of the INS and GPS integrated navigation system.
Specifically, in an implementation manner of this embodiment, in the process of constructing the error model of the INS, the method includes the following steps:
step S102a, constructing an INS attitude error equation set;
step S102b, constructing an INS speed error equation set;
in step S102c, an INS position error equation set is constructed.
In this embodiment, the constructed error model of the INS includes: an INS attitude error equation set, an INS speed error equation set and an INS position error equation set; wherein the constructed INS attitude error equation set is as follows:
Figure 827248DEST_PATH_IMAGE001
in the INS attitude error equation set,
Figure 992650DEST_PATH_IMAGE002
an attitude angle error vector is obtained;
Figure 89919DEST_PATH_IMAGE003
and
Figure 532532DEST_PATH_IMAGE004
the rotation angular velocity and the error of the navigation coordinate system relative to the inertial coordinate system are respectively;
Figure 158686DEST_PATH_IMAGE005
and the gyroscope drift vector under the navigation coordinate system.
In this embodiment, the constructed INS velocity error equation set is:
Figure 76963DEST_PATH_IMAGE006
in the system of INS velocity error equations,
Figure 338443DEST_PATH_IMAGE007
velocity errors in the east, north and sky directions;
Figure 494617DEST_PATH_IMAGE008
east, north and sky;
Figure 822831DEST_PATH_IMAGE009
is a specific force value;
Figure 572612DEST_PATH_IMAGE010
and
Figure 277263DEST_PATH_IMAGE011
respectively the rotation speed of the earth in a navigation coordinate system and the angle relative to the earth;
Figure 553523DEST_PATH_IMAGE012
is the deviation of the accelerometer in the navigational coordinate system.
In this embodiment, the constructed INS position error equation set is:
Figure 177271DEST_PATH_IMAGE013
in the system of INS position error equations,
Figure 538983DEST_PATH_IMAGE014
，
Figure 657111DEST_PATH_IMAGE015
and
Figure 787878DEST_PATH_IMAGE016
longitude error, latitude error and altitude error respectively;
Figure 192315DEST_PATH_IMAGE017
、
Figure 933000DEST_PATH_IMAGE018
and
Figure 448295DEST_PATH_IMAGE019
north, sky, and east speed errors, respectively;
Figure 964727DEST_PATH_IMAGE020
and
Figure 415431DEST_PATH_IMAGE021
the meridian curvature radius and the main perpendicular curvature radius are respectively shown.
In this embodiment, after obtaining the INS dynamic error equation set, a plurality of system variables are selected to form a state vector of the INS and GPS integrated navigation system according to the INS dynamic error equation set, so as to iteratively update the kalman filter according to the selected state vector.
Specifically, in an implementation manner of this embodiment, when constructing the state equation and the observation equation of the INS and GPS combined navigation system, the method includes the following steps:
step S104a, selecting a plurality of system variables to form a state vector of the INS and GPS integrated navigation system;
and step S104b, constructing a discrete state space equation of the INS and GPS integrated navigation system based on the error model of the INS.
In this embodiment, 15 system variables may be selected to form a system state vector, that is:
Figure 17314DEST_PATH_IMAGE038
wherein the content of the first and second substances,
Figure 336300DEST_PATH_IMAGE023
is the attitude angle error vector in the navigation coordinate system;
Figure 831872DEST_PATH_IMAGE024
is the speed error;
Figure 843690DEST_PATH_IMAGE025
for longitude errors, latitude errors and altitude errors,
Figure 277077DEST_PATH_IMAGE026
and
Figure 665333DEST_PATH_IMAGE027
respectively representing accelerometer bias and gyro bias in the body coordinate system.
Further, according to the selected system variables, a discrete state space equation of the INS and GPS integrated navigation system is constructed:
Figure 625199DEST_PATH_IMAGE028
wherein, the first and the second end of the pipe are connected with each other,
Figure 434017DEST_PATH_IMAGE029
representing a state transition matrix;
Figure 10492DEST_PATH_IMAGE030
representing a system noise distribution matrix;
Figure 546646DEST_PATH_IMAGE031
and
Figure 626598DEST_PATH_IMAGE032
respectively process noise and measurement noise；
Figure 980219DEST_PATH_IMAGE033
Is a measurement matrix;
Figure 903044DEST_PATH_IMAGE034
is a measurement vector.
In this embodiment, the state equation and the observation equation of the INS and GPS integrated navigation system are presented in the form of a discrete state space equation, and the above equations are substituted into a kalman filter to perform cyclic update iteration, where the kalman filter compares data difference values of the GPS sensor over a period of time and compares data difference values of the INS sensor over a period of time, so as to establish an error model to estimate the errors of the INS, and the data results acquired by the INS in real time are corrected by using the errors, so as to obtain the integrated navigation results of speed, position and attitude.
As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the deep learning based integrated navigation method further includes the following steps:
and S200, deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of the deep learning model.
In this embodiment, a deep learning algorithm is incorporated based on the GPS/INS integrated navigation system, and the deep learning algorithm assists the GPS/INS integrated navigation system in navigation, so as to improve the navigation accuracy of the GPS/INS integrated navigation system when signals are interrupted.
In a constructed loosely-coupled GPS/INS integrated navigation system, a deep learning network model needs to be designed, the attitude angle, the specific force value and the angular velocity value of the INS are used as the input of the deep learning network model, the position increment value of the GPS is used as the output of the network model, and the deep learning network model is trained and tested.
Specifically, in one implementation manner of the present embodiment, the step S200 includes the following steps:
step S201, substituting the discrete state space equation into the Kalman filter to perform cycle updating iteration;
step S202, comparing the position increment of the GPS with the output value of the INS through the Kalman filter, and deducing the relation between the position increment of the GPS and the output value of the INS so as to determine the input value and the output value of the deep learning model.
As shown in fig. 2, in the present embodiment, in the process of building the deep learning network model, the method includes: a training stage and a testing stage; in the training stage, the position increment of the GPS is mainly taken as a target, the input (namely the attitude angle, the specific force value and the angular velocity value of the INS) of the deep learning network model is corrected, then the corrected result is output, and iterative training is carried out by using the corrected result until the INS value meets the expected result; and in the testing stage, the corrected INS output value is used as a reference to predict the position increment of the GPS, and the prediction result of the position increment of the GPS and the corrected INS output value are used as the final output result of the deep learning network model.
As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the deep learning based integrated navigation method further includes the following steps:
step S300, the deep learning model and the loss function thereof are constructed, and the nonlinear relation between the input value and the output value is learned.
In this embodiment, the constructed deep learning model is a GI-NN network model, and the structure of the GI-NN network model is shown in fig. 3, which includes: an input layer, a convolutional layer, a GRU layer, a Dropout layer, and a linear output layer.
In an implementation manner of this embodiment, the input layer is an inertial data input layer of the INS; the convolution layer is 4 layers of one-dimensional convolution layers, the number of input channels is 9, the number of output channels is 512, and the size of convolution kernel is 1; the number of input channels of the GRU layer is 512, the number of output channels is 128, and the number of units is 40; the number of input channels of the Dropout layer is 128, the number of output channels is 64, and the discarding rate is 0.25; the number of input channels of the linear output layer is 64, and the number of output channels is 2.
Specifically, in one implementation manner of the present embodiment, the step S300 includes the following steps:
step S301, constructing the deep learning model;
step S302, determining the expression form of the INS output value and the GPS position increment in feature mapping, and establishing a nonlinear relation between an input layer and an output layer of the deep learning model; wherein, the output value of the INS comprises an attitude angle value, a specific force value and an angular velocity value.
In this embodiment, when constructing the GI-NN network model, a convolutional layer is used as a feature extractor to learn sparse feature representation, and model implicit dependency relationships among various sensors, so as to provide an abstract representation of input sensor data in feature mapping, and simultaneously, utilize a time dynamic characteristic activated by a GRU layer modeling feature map.
Specifically, in one implementation manner of the present embodiment, the step S301 includes the following steps:
step S301a, using the attitude angle value, the specific force value, and the angular velocity value of the INS as input values of the deep learning model, and using the position increment value of the GPS as an output value of the deep learning model.
As an implementation manner of this embodiment, a kalman filter may be designed based on a state equation and an observation equation of the INS and the GPS integrated navigation system, the position, speed, and attitude angle errors of the INS and the GPS are input to the kalman filter, the kalman filter updates and iteratively estimates the INS error according to the error covariance matrix, and feeds back an output value of the compensation correction INS; then, designing a deep learning model, mapping the IMU sequence to a high-dimensional space by utilizing a one-dimensional convolutional layer, learning sparse feature representation, excavating time-space features of the IMU sequence through a GRU circulating network layer, and modeling implicit dependency relations among sensors, thereby providing abstract representation of input sensor data in feature mapping and establishing a nonlinear relation between input and output of the network model.
Specifically, in one implementation manner of the present embodiment, the step S302 includes the following steps:
step S302a, adopting the convolution layer as a characteristic extractor, and learning sparse characteristics through the convolution layer;
step S302b, modeling an implicit dependency relationship between the INS and the GPS, providing an abstract representation of an output value of the INS in a feature map, and modeling a time dynamic characteristic activated by a feature map using a GRU layer.
In this embodiment, the input to the deep learning network model may be represented as a time series vector
Figure 367524DEST_PATH_IMAGE039
And each element
Figure 177348DEST_PATH_IMAGE040
The attitude angle, specific force and angular velocity 9 measurements,
Figure 701870DEST_PATH_IMAGE041
indicating the size of the time window. The prediction output is based on the prior
Figure 721779DEST_PATH_IMAGE041
Obtaining IMU measurement value in time step
Figure 616048DEST_PATH_IMAGE042
GPS position increment in
Figure 405012DEST_PATH_IMAGE043
It can be expressed as:
Figure 366015DEST_PATH_IMAGE044
in this embodiment, after the nonlinear relationship between the input layer and the output layer of the deep learning model is established, the deep learning network model can well learn the nonlinear relationship between the input and the output by performing a corresponding simulation experiment and a corresponding real scene experiment, so as to obtain a trained deep learning network model.
As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the deep learning based integrated navigation method further includes the following steps:
and S400, respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain the trained deep learning model.
In this embodiment, after the nonlinear relationship between input and output is determined, a simulation experiment of GPS interrupt and a real scene experiment are performed respectively, a GPS/INS combined navigation data set is collected, the data set is divided into a training set and a test set, a network model is trained using the training set, and a GI-NN network model is optimized by adjusting hyper-parameters.
Specifically, in one implementation manner of the present embodiment, the step S400 includes the following steps:
step S401, respectively performing the simulation experiment and the real scene experiment;
step S402, collecting a data set of the INS and GPS integrated navigation system, and dividing the data set according to a preset proportion to obtain the training set and the test set;
step S403, training the deep learning model through the training set, and testing the deep learning model through the testing set;
and S404, adjusting the hyper-parameters of the deep learning model to obtain the trained deep learning model.
In this embodiment, according to the training process and the testing process shown in fig. 2, a corresponding simulation experiment and a real scene experiment are performed respectively; by collecting the INS and GPS combined navigation data set, the data set is divided into 7: and 3, dividing a training set and a testing set, training the GI-NN network model through a large amount of INS and GPS combined navigation data when the GPS signal is available, and adjusting the training hyper-parameters to enable the GI-NN network model to reach an optimal state, so that the deep learning network model can well learn the nonlinear relation between input and output, and the navigation data can be predicted when the GPS signal is interrupted.
As shown in fig. 1, in an implementation manner of the embodiment of the present invention, the deep learning based integrated navigation method further includes the following steps:
and S500, carrying out GPS interruption test by using the trained deep learning model, generating lost GPS position data, and calibrating the output value of the INS according to the lost GPS position data to obtain navigation data after GPS interruption.
In this embodiment, when the GPS is interrupted, the trained model is used to predict the GPS position increment, and the position increment value at each time is accumulated to generate lost GPS position data, which is input to the kalman filter to be fused with the INS sensor data, so that the INS and GPS integrated navigation system still maintains better navigation accuracy under the condition of the GPS interruption, and the output value of the INS is calibrated by the lost GPS position data to obtain accurate navigation data after the GPS interruption.
Specifically, in one implementation manner of the present embodiment, the step S500 includes the following steps:
step S501, inputting the attitude angle value, the specific force value and the angular velocity value measured by the INS into the trained deep learning model to generate the lost GPS position data;
step S502, substituting the lost GPS position data into the Kalman filter, updating an error covariance matrix, and obtaining the calibrated INS output value;
and S503, fusing the lost GPS position data and the calibrated INS output value to obtain navigation data after the GPS is interrupted.
In this embodiment, when the GPS signal is interrupted, the attitude angle, specific force and angular velocity measured by the INS are input into the deep learning model GI-NN trained in the previous step, and a GPS position increment value is predicted, and then a pseudo GPS position value is generated by adding up the position increment value at each time, which can be specifically expressed as:
Figure 748586DEST_PATH_IMAGE035
wherein the content of the first and second substances,
Figure 554868DEST_PATH_IMAGE036
is the initial position value at the initial time k when the GPS is interrupted,
Figure 463918DEST_PATH_IMAGE037
is the final predicted GPS value.
By generating
Figure 454877DEST_PATH_IMAGE037
And substituting the updated error covariance matrix into the Kalman filter in the second step to correct the INS error, so that the combined navigation of the INS and the GPS can still continue to work under the condition of GPS interruption, and better navigation accuracy is maintained.
The embodiment achieves the following technical effects through the technical scheme:
aiming at the scene of GPS interruption, the embodiment designs a unique and novel deep learning model framework under the condition of not additionally adding an additional sensor, and predicts lost GPS data, so that the INS and the GPS can continuously and normally work and output high-precision and reliable navigation data. The method comprises the steps of firstly mapping an input IMU sequence to a high-dimensional space by introducing one-dimensional convolution in a deep learning model, and then extracting the characteristics of the sequence, thereby achieving the effect of noise reduction. The implicit association characteristic between the IMU historical moment data and the current moment is fully considered by utilizing the GRU circulating network layer, and the calculation load is reduced compared with that of a long-short-term memory neural network.
Exemplary device
Based on the above embodiments, the present invention further provides a combined navigation device based on deep learning, and its functional block diagram may be as shown in fig. 4.
The integrated navigation device based on deep learning comprises: the system comprises a processor, a memory, an interface, a display screen and a communication module which are connected through a system bus; wherein the processor of the deep learning based integrated navigation device is used for providing calculation and control capability; the memory of the combined navigation device based on deep learning comprises a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of an operating system and a computer program in the storage medium; the interface is used for connecting external equipment, such as mobile terminals, computers and the like; the display screen is used for displaying corresponding combined navigation information based on deep learning; the communication module is used for communicating with a cloud server or a mobile terminal.
The computer program is adapted to be executed by a processor to implement a deep learning based integrated navigation method.
It will be understood by those skilled in the art that the schematic block diagram shown in fig. 4 is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation of the combined navigation device based on deep learning to which the solution of the present invention is applied, and a specific combined navigation device based on deep learning may include more or less components than those shown in the figure, or combine some components, or have different arrangements of components.
In one embodiment, a deep learning based integrated navigation device is provided, which includes: the combined navigation method based on the deep learning comprises a processor and a memory, wherein the memory stores a combined navigation program based on the deep learning, and the combined navigation program based on the deep learning is used for realizing the combined navigation method based on the deep learning when being executed by the processor.
Exemplary System
Based on the embodiment, the invention also provides a combined navigation system based on deep learning, which comprises a GPS sensor, an INS sensor, a Kalman filter and the combined navigation device based on deep learning;
the GPS sensor is used for acquiring position data;
the INS sensor is used for acquiring an attitude angle value, a specific force value and an angular velocity value;
the Kalman filter is used for fusing the GPS sensor and the INS sensor to obtain an INS and GPS integrated navigation system; the data acquisition module is used for respectively comparing the data of the GPS sensor with the data of the INS sensor to obtain data difference values of the sensors; the system comprises a GPS module, a calibration module, a GPS module and a control module, wherein the GPS module is used for acquiring GPS position data;
the integrated deep learning based navigation device is used for executing the following steps:
an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system;
deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of a deep learning model;
constructing the deep learning model and a loss function thereof, and learning a nonlinear relation between the input value and the output value;
respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model;
and carrying out GPS interruption test by using the trained deep learning model, generating lost GPS position data, and calibrating the output value of the INS according to the lost GPS position data to obtain navigation data after GPS interruption.
In one embodiment, a storage medium is provided, wherein the storage medium stores a deep learning based integrated navigation program, and the deep learning based integrated navigation program is used for implementing the deep learning based integrated navigation method when being executed by a processor.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by instructing relevant hardware by a computer program, and the computer program may be stored in a non-volatile storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory.
In summary, the present invention provides a method, an apparatus, a system and a storage medium for integrated navigation based on deep learning, wherein the method comprises: an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system; deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of the deep learning model; constructing a deep learning model and a loss function thereof, and learning a nonlinear relation between an input value and an output value; respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model; and carrying out GPS interruption test, generating lost GPS position data, and calibrating the output value of the INS according to the lost data to obtain navigation data after GPS interruption. According to the method, the Kalman filter and the deep learning are combined, and the robustness and the precision of the INS and GPS integrated navigation are improved.
It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.

Claims (10)

1. A deep learning based integrated navigation method is characterized in that the deep learning based integrated navigation method comprises the following steps:
an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system;
deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of a deep learning model;
constructing the deep learning model and a loss function thereof, and learning a nonlinear relation between the input value and the output value;
respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model;
carrying out GPS interruption test by using the trained deep learning model, generating lost GPS position data, and calibrating the output value of the INS according to the lost GPS position data to obtain navigation data after GPS interruption;
the building of the deep learning model and the loss function thereof, and the learning of the nonlinear relationship between the input value and the output value include:
constructing the deep learning model;
determining a representation form of the output value of the INS and the position increment of the GPS in feature mapping, and establishing a nonlinear relation between an input layer and an output layer of the deep learning model;
wherein, the output value of the INS comprises an attitude angle value, a specific force value and an angular velocity value;
the building of the deep learning model comprises the following steps:
taking the attitude angle value, the specific force value and the angular velocity value of the INS as input values of the deep learning model, and taking the position increment value of the GPS as an output value of the deep learning model;
determining a representation of the INS output values and the GPS position increments in a feature map, establishing a non-linear relationship between an input layer and an output layer of the deep learning model, comprising:
learning sparse features through convolutional layers by adopting the convolutional layers as feature extractors;
modeling an implicit dependency relationship between the INS and the GPS, providing an abstract representation of an output value of the INS in a feature mapping, and modeling a time dynamic characteristic activated by a feature map by using a GRU layer;
the constructed deep learning model is a GI-NN network model, and the structure of the GI-NN network model comprises: an input layer, a convolution layer, a GRU layer, a Dropout layer, and a linear output layer;
mapping the IMU sequence to a high-dimensional space by utilizing a one-dimensional convolutional layer, learning sparse feature representation, excavating time-space features of the IMU sequence through the GRU layer, and modeling an implicit dependency relationship between the INS and the GPS.
2. The integrated navigation method based on deep learning of claim 1, wherein the constructing of the integrated navigation system of INS and GPS by fusing INS and GPS through a kalman filter comprises:
fusing the INS and the GPS through the Kalman filter;
constructing an error model of the INS;
taking the position increment of the GPS as an observed value of the INS and GPS combined navigation system;
and determining the state vector of the INS and GPS integrated navigation system, and constructing a state equation and an observation equation of the INS and GPS integrated navigation system.
3. The integrated navigation method based on deep learning of claim 2, wherein the constructing the error model of the INS comprises:
constructing an INS attitude error equation set:
Figure DEST_PATH_IMAGE002
wherein, the first and the second end of the pipe are connected with each other,
Figure DEST_PATH_IMAGE004
an attitude angle error vector is obtained;
Figure DEST_PATH_IMAGE006
and
Figure DEST_PATH_IMAGE008
the rotation angular velocity and the error of the navigation coordinate system relative to the inertial coordinate system are respectively;
Figure DEST_PATH_IMAGE010
a gyroscope drift vector under a navigation coordinate system;
constructing an INS speed error equation set:
Figure DEST_PATH_IMAGE012
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE014
velocity errors in the east, north and sky directions;
Figure DEST_PATH_IMAGE016
speeds in the east, north and sky directions;
Figure DEST_PATH_IMAGE018
is a specific force value;
Figure DEST_PATH_IMAGE020
and
Figure DEST_PATH_IMAGE022
respectively the rotation speed of the earth in a navigation coordinate system and the angle relative to the earth;
Figure DEST_PATH_IMAGE024
is the deviation of the accelerometer in the navigation coordinate system;
constructing an INS position error equation set:
Figure DEST_PATH_IMAGE026
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE028
，
Figure DEST_PATH_IMAGE030
and
Figure DEST_PATH_IMAGE032
longitude error, latitude error and altitude error respectively;
Figure DEST_PATH_IMAGE034
、
Figure DEST_PATH_IMAGE036
and
Figure DEST_PATH_IMAGE038
respectively a north direction speed error, a sky direction speed error and an east direction speed error;
Figure DEST_PATH_IMAGE040
and
Figure DEST_PATH_IMAGE042
the meridian curvature radius and the main perpendicular curvature radius are respectively shown.
4. The integrated navigation method based on deep learning of claim 2, wherein the determining the state vector of the INS and GPS integrated navigation system, and the constructing the state equation and the observation equation of the INS and GPS integrated navigation system comprise:
selecting a plurality of system variables to form a state vector of the INS and GPS combined navigation system:
Figure DEST_PATH_IMAGE044
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE046
is the attitude angle error vector in the navigation coordinate system;
Figure DEST_PATH_IMAGE048
is the speed error;
Figure DEST_PATH_IMAGE050
as a longitude error, a latitude error, and an altitude error,
Figure DEST_PATH_IMAGE052
and
Figure DEST_PATH_IMAGE054
respectively representing an accelerometer deviation and a gyro deviation in a body coordinate system;
and constructing a discrete state space equation of the INS and GPS integrated navigation system based on the error model of the INS:
Figure DEST_PATH_IMAGE056
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE058
representing a state transition matrix;
Figure DEST_PATH_IMAGE060
representing a system noise distribution matrix;
Figure DEST_PATH_IMAGE062
and
Figure DEST_PATH_IMAGE064
process noise and measurement noise, respectively;
Figure DEST_PATH_IMAGE066
is a measurement matrix;
Figure DEST_PATH_IMAGE068
is a measurement vector.
5. The integrated navigation method based on deep learning of claim 4, wherein the deriving the relationship between the position increment of the GPS and the output value of the INS to determine the input value and the output value of the deep learning model comprises:
substituting the discrete state space equation into the Kalman filter to perform cycle updating iteration;
and comparing the position increment of the GPS with the output value of the INS through the Kalman filter, and deducing the relation between the position increment of the GPS and the output value of the INS so as to determine the input value and the output value of the deep learning model.
6. The integrated navigation method based on deep learning of claim 1, wherein the performing of the simulation experiment and the real scene experiment, respectively, and the adjusting of the hyper-parameters of the deep learning model to obtain the trained deep learning model comprises:
respectively carrying out the simulation experiment and the real scene experiment;
collecting a data set of the INS and GPS integrated navigation system, and dividing the data set according to a preset proportion to obtain a training set and a test set;
training the deep learning model through the training set, and testing the deep learning model through the testing set;
and adjusting the hyper-parameters of the deep learning model to obtain the trained deep learning model.
7. The integrated navigation method based on deep learning of claim 1, wherein the performing a GPS outage test using the trained deep learning model to generate missing GPS location data and calibrating the INS output values according to the missing GPS location data comprises:
inputting the attitude angle value, the specific force value and the angular velocity value measured by the INS into the trained deep learning model to generate the lost GPS position data:
Figure DEST_PATH_IMAGE070
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE072
for the initiation of the initial time k at GPS interruptionThe value of the position is set to be,
Figure DEST_PATH_IMAGE074
is the final predicted GPS value;
substituting the lost GPS position data into the Kalman filter, and updating an error covariance matrix to obtain the calibrated INS output value;
and fusing the lost GPS position data and the calibrated INS output value to obtain navigation data after the GPS is interrupted.
8. A combined navigation device based on deep learning is characterized by comprising: a processor and a memory, the memory storing a deep learning based integrated navigation program, the deep learning based integrated navigation program when executed by the processor being adapted to implement the deep learning based integrated navigation method according to any one of claims 1 to 7.
9. A deep learning based integrated navigation system, comprising: a GPS sensor, an INS sensor, a kalman filter, and the deep learning based integrated navigation device of claim 8;
the GPS sensor is used for acquiring position data;
the INS sensor is used for acquiring an attitude angle value, a specific force value and an angular velocity value;
the Kalman filter is used for fusing the GPS sensor and the INS sensor to obtain an INS and GPS integrated navigation system; the data acquisition module is used for respectively comparing the data of the GPS sensor with the data of the INS sensor to obtain data difference values of the sensors; the system comprises a GPS module, a correction module and a processing module, wherein the GPS module is used for acquiring GPS position data;
the integrated deep learning based navigation device is used for executing the following steps:
an INS and a GPS are fused through a Kalman filter to construct an INS and GPS integrated navigation system;
deducing the relation between the position increment of the GPS and the output value of the INS, and determining the input value and the output value of a deep learning model;
constructing the deep learning model and a loss function thereof, and learning a nonlinear relation between the input value and the output value;
respectively carrying out a simulation experiment and a real scene experiment, and adjusting the hyper-parameters of the deep learning model to obtain a trained deep learning model;
carrying out GPS interruption test by using the trained deep learning model, generating lost GPS position data, and calibrating the output value of the INS according to the lost GPS position data to obtain navigation data after GPS interruption;
the constructing the deep learning model and the loss function thereof, and learning the nonlinear relationship between the input value and the output value includes:
constructing the deep learning model;
determining a representation form of the output value of the INS and the position increment of the GPS in feature mapping, and establishing a nonlinear relation between an input layer and an output layer of the deep learning model;
wherein the output values of the INS comprise an attitude angle value, a specific force value and an angular velocity value;
the building of the deep learning model comprises the following steps:
taking the attitude angle value, the specific force value and the angular velocity value of the INS as input values of the deep learning model, and taking the position increment value of the GPS as an output value of the deep learning model;
determining a representation of the INS output values and the GPS position increments in a feature map, establishing a non-linear relationship between an input layer and an output layer of the deep learning model, comprising:
learning sparse features through a convolutional layer by adopting the convolutional layer as a feature extractor;
modeling an implicit dependency relationship between the INS and the GPS, providing an abstract representation of an output value of the INS in a feature map, and modeling a time dynamic characteristic activated by a feature map by using a GRU layer;
the constructed deep learning model is a GI-NN network model, and the structure of the GI-NN network model comprises: an input layer, a convolutional layer, a GRU layer, a Dropout layer, and a linear output layer;
mapping the IMU sequence to a high-dimensional space by utilizing a one-dimensional convolutional layer, learning sparse feature representation, excavating time-space features of the IMU sequence through the GRU layer, and modeling an implicit dependency relationship between the INS and the GPS.
10. A storage medium, which is a computer-readable storage medium, and which stores a deep learning based integrated navigation program, and when the deep learning based integrated navigation program is executed by a processor, the deep learning based integrated navigation program is used for implementing the deep learning based integrated navigation method according to any one of claims 1 to 7.
CN202210616148.2A 2022-06-01 2022-06-01 Deep learning-based integrated navigation method, device, system and storage medium Active CN114689047B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210616148.2A CN114689047B (en) 2022-06-01 2022-06-01 Deep learning-based integrated navigation method, device, system and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210616148.2A CN114689047B (en) 2022-06-01 2022-06-01 Deep learning-based integrated navigation method, device, system and storage medium

Publications (2)

Publication Number Publication Date
CN114689047A CN114689047A (en) 2022-07-01
CN114689047B true CN114689047B (en) 2022-09-13

Family

ID=82131049

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210616148.2A Active CN114689047B (en) 2022-06-01 2022-06-01 Deep learning-based integrated navigation method, device, system and storage medium

Country Status (1)

Country Link
CN (1) CN114689047B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115290067A (en) * 2022-07-09 2022-11-04 哈尔滨工程大学 Big data navigation method supporting vector clustering neural network
CN116518981B (en) * 2023-06-29 2023-09-22 中国人民解放军国防科技大学 Aircraft visual navigation method based on deep learning matching and Kalman filtering
CN117091457B (en) * 2023-08-03 2024-02-13 南京理工大学 Guided projectile navigation method and system based on deep learning

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108763493A (en) * 2018-05-30 2018-11-06 深圳市思迪信息技术股份有限公司 A kind of recommendation method based on deep learning
CN112578419A (en) * 2020-11-24 2021-03-30 南京邮电大学 GPS data reconstruction method based on GRU network and Kalman filtering
CN113221758A (en) * 2021-05-16 2021-08-06 西北工业大学 Underwater acoustic target identification method based on GRU-NIN model
CN113447021A (en) * 2021-07-15 2021-09-28 北京理工大学 MEMS inertial navigation system positioning enhancement method based on LSTM neural network model
CN114239376A (en) * 2021-10-29 2022-03-25 武汉大学 Algorithm for assisting navigation by using GRU neural network during GNSS signal interruption

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2447809C (en) * 2001-06-04 2011-08-02 Novatel Inc. An inertial/gps navigation system
CN101871782B (en) * 2010-05-19 2011-11-02 北京航空航天大学 Position error forecasting method for GPS (Global Position System)/MEMS-INS (Micro-Electricomechanical Systems-Inertial Navigation System) integrated navigation system based on SET2FNN
CN106980133A (en) * 2017-01-18 2017-07-25 中国南方电网有限责任公司超高压输电公司广州局 The GPS INS Combinated navigation methods and system for being compensated and being corrected using neural network algorithm
CN107390246A (en) * 2017-07-06 2017-11-24 电子科技大学 A kind of GPS/INS Combinated navigation methods based on genetic neural network
CN107655472B (en) * 2017-08-30 2019-11-01 杨华军 A kind of high-precision inertial navigation set error compensating method based on deep learning
CN109521454B (en) * 2018-12-06 2020-06-09 中北大学 GPS/INS integrated navigation method based on self-learning volume Kalman filtering
CN110346821B (en) * 2019-07-17 2022-11-29 贵州理工学院 SINS/GPS combined attitude-determining and positioning method and system for solving long-time GPS unlocking problem
CN110487271A (en) * 2019-09-26 2019-11-22 哈尔滨工程大学 Elman neural network aiding tight integration air navigation aid when a kind of GNSS signal is obstructed
CN111275969B (en) * 2020-02-15 2022-02-25 湖南大学 Vehicle track filling method based on intelligent identification of road environment
CN111854741B (en) * 2020-06-16 2022-08-09 中国人民解放军战略支援部队信息工程大学 GNSS/INS tight combination filter and navigation method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108763493A (en) * 2018-05-30 2018-11-06 深圳市思迪信息技术股份有限公司 A kind of recommendation method based on deep learning
CN112578419A (en) * 2020-11-24 2021-03-30 南京邮电大学 GPS data reconstruction method based on GRU network and Kalman filtering
CN113221758A (en) * 2021-05-16 2021-08-06 西北工业大学 Underwater acoustic target identification method based on GRU-NIN model
CN113447021A (en) * 2021-07-15 2021-09-28 北京理工大学 MEMS inertial navigation system positioning enhancement method based on LSTM neural network model
CN114239376A (en) * 2021-10-29 2022-03-25 武汉大学 Algorithm for assisting navigation by using GRU neural network during GNSS signal interruption

Also Published As

Publication number Publication date
CN114689047A (en) 2022-07-01

Similar Documents

Publication Publication Date Title
CN114689047B (en) Deep learning-based integrated navigation method, device, system and storage medium
CN111721289B (en) Vehicle positioning method, device, equipment, storage medium and vehicle in automatic driving
CN112146655B (en) Elastic model design method for BeiDou/SINS tight integrated navigation system
Doostdar et al. INS/GNSS integration using recurrent fuzzy wavelet neural networks
CN108387236B (en) Polarized light SLAM method based on extended Kalman filtering
CN111189442B (en) CEPF-based unmanned aerial vehicle multi-source navigation information state prediction method
CN115143954B (en) Unmanned vehicle navigation method based on multi-source information fusion
US20230366680A1 (en) Initialization method, device, medium and electronic equipment of integrated navigation system
CN112985391B (en) Multi-unmanned aerial vehicle collaborative navigation method and device based on inertia and binocular vision
CN105547300A (en) All-source navigation system and method used for AUV (Autonomous Underwater Vehicle)
CN112595313A (en) Vehicle-mounted navigation method and device based on machine learning and computer equipment
CN114061570A (en) Vehicle positioning method and device, computer equipment and storage medium
CN114061611A (en) Target object positioning method, apparatus, storage medium and computer program product
US20220057517A1 (en) Method for constructing point cloud map, computer device, and storage medium
CN111964675A (en) Intelligent aircraft navigation method for blackout area
CN113566850B (en) Method and device for calibrating installation angle of inertial measurement unit and computer equipment
CN113375664B (en) Autonomous mobile device positioning method based on dynamic loading of point cloud map
CN104634348B (en) Attitude angle computational methods in integrated navigation
Glavine et al. Gps integrated inertial navigation system using interactive multiple model extended kalman filtering
CN109655057B (en) Filtering optimization method and system for accelerator measurement value of six-push unmanned aerial vehicle
Sharaf et al. Merits and limitations of using fuzzy inference system for temporal integration of INS/GPS in vehicular navigation
CN114088104B (en) Map generation method under automatic driving scene
CN114001730B (en) Fusion positioning method, fusion positioning device, computer equipment and storage medium
EP3855117A1 (en) Terrain referenced navigation system with generic terrain sensors for correcting an inertial navigation solution
CN115711616A (en) Indoor and outdoor unmanned aerial vehicle penetrating smooth positioning method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant