CN116026325A - Navigation method and related device based on neural process and Kalman filtering - Google Patents
Navigation method and related device based on neural process and Kalman filtering Download PDFInfo
- Publication number
- CN116026325A CN116026325A CN202310125904.6A CN202310125904A CN116026325A CN 116026325 A CN116026325 A CN 116026325A CN 202310125904 A CN202310125904 A CN 202310125904A CN 116026325 A CN116026325 A CN 116026325A
- Authority
- CN
- China
- Prior art keywords
- navigation
- data
- training
- model
- layer
- 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.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Navigation (AREA)
Abstract
The invention discloses a navigation method and a related device based on a nerve process and Kalman filtering, wherein the navigation method comprises the following steps: solving navigation data of the speed and the gesture of the motion carrier, and combining the navigation data with the gyroscope and the accelerometer data to form a training data set; grouping data in the training data set, and establishing an NP model of a navigation method; training the NP model by selecting training data until the training result converges to obtain a navigation parameter prediction model; updating the attitude information of the motion carrier, inputting the updated attitude information into a navigation parameter prediction model, and estimating navigation parameters at the later moment; and repeatedly updating the attitude information of the motion carrier until the navigation task is completed or the GNSS is stabilized again. The invention combines LSTM with NP, and reduces the dependency of NP on data format. The invention combines NP and KF, and inputs the navigation parameters corrected by KF into NP for the prediction of the next period.
Description
Technical Field
The invention belongs to the technical field of inertial navigation, and relates to a navigation method and a related device based on a neural process and Kalman filtering.
Background
Global navigation satellite system (Global Navigation Satellite System, GNSS) -strapdown inertial navigation system (Strapdown Inertial Navigation System, SINS) integrated navigation technology is one of the most widely used integrated navigation technologies at present. However, due to the vulnerability of the GNSS system, the complexity and diversity of the application environment, the GNSS signals are easily unlocked or disturbed, thereby seriously affecting the accuracy and stability of the GNSS/SINS integrated navigation system. In the case of GNSS out-of-lock, how to obtain navigation information with long time and high accuracy has been a problem that is desired to be solved in both academia and industry. Aiming at the problem, a great deal of research work is carried out by students at home and abroad, and two solutions of adding auxiliary sensors and estimating navigation parameters of a motion carrier by using a machine learning algorithm such as a neural network are provided.
Compared with the scheme of adding auxiliary sensors, the motion carrier pose estimation method based on the neural network has the advantages of simple structure, good applicability, easiness in upgrading and reconstruction on the existing navigation system and the like, and is a hot research direction for obtaining high-precision navigation information under the condition of GNSS unlocking. For this scheme, the currently customary approach is a motion vector pose estimation scheme based on neural networks, kalman Filters (KF) and SINS. When the GNSS stably works, the neural network is trained by using the sensor data and the navigation parameters, the parameters of the neural network are adjusted to minimize errors, and the neural network is substantially approximated to the function by adjusting the parameters of the network. And when the GNSS fails, the sensor data and the navigation parameters are directly input into the neural network for prediction. Although the scheme has a certain progress in predicting the position and the speed of the motion carrier, a certain gap exists from the practical engineering application, and the problem is mainly reflected in that after GNSS is out of lock, the existing method is often difficult to obtain high-precision estimation under the influence of measurement noise and model uncertainty, and when the gesture of the motion carrier changes, the estimation precision of the position and the speed can be rapidly reduced, and only the two aspects of short-time application scenes with unchanged gestures and the motion track after the gesture of the motion carrier can not be estimated can be dealt with.
Disclosure of Invention
The invention aims to overcome the defects of the traditional neural network estimation algorithm, provides a navigation method and a related device based on a neural process and Kalman filtering, utilizes output data of a gyroscope and an accelerometer to realize high-precision estimation of the position information probability distribution of a motion carrier based on the advantages of NP on random variable estimation, and utilizes KF to realize joint estimation of the pose of the motion carrier by means of a dynamic equation of a system to realize long-time and high-precision estimation prediction of multidimensional coupling navigation parameters of the motion carrier under the condition of only inputting gyroscope and acceleration information.
In order to achieve the above purpose, the invention is realized by adopting the following technical scheme:
in a first aspect, the present invention provides a navigation method based on neural process and kalman filtering, comprising the steps of:
solving navigation data of the speed and the gesture of the motion carrier, and combining the navigation data with the gyroscope and the accelerometer data to form a training data set;
grouping data in the training data set, and establishing an NP model of a navigation method;
training the NP model by selecting training data until the training result converges to obtain a navigation parameter prediction model;
updating the attitude information of the motion carrier, inputting the updated attitude information into a navigation parameter prediction model, and estimating navigation parameters at the later moment; and repeatedly updating the attitude information and estimating navigation parameters at the later moment until the navigation task is completed or the GNSS is stabilized again.
In a second aspect, the present invention provides a navigation system method based on neural process and kalman filtering, comprising the steps of:
the data set generation module is used for solving navigation data of the speed and the gesture of the motion carrier, and combining the navigation data, the gyroscope and the accelerometer data into a training data set;
the NP model building module is used for grouping data in the training data set and building an NP model of the navigation method;
the NP model training module is used for selecting training data to train the NP model until the training result converges to obtain a navigation parameter prediction model;
the navigation parameter estimation module is used for updating the gesture information of the motion carrier, inputting the updated gesture information into the navigation parameter prediction model and estimating the navigation parameter at the later moment; and repeatedly updating the attitude information of the motion carrier until the navigation task is completed or the GNSS is stabilized again.
In a third aspect, the present invention provides a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method as described above when executing the computer program.
In a fourth aspect, the present invention provides a computer readable storage medium storing a computer program which when executed by a processor performs the steps of a method as described above.
Compared with the prior art, the invention has the following beneficial effects:
aiming at the limitations of the existing combined navigation system and Neural network estimation method, the invention estimates the probability distribution information of the carrier navigation parameters through a Neural Process (NP) algorithm, thereby realizing the estimation of the key parameters. And updating the estimated parameters through KF, and realizing navigation parameter estimation by using a gyroscope and an accelerometer under GNSS unlocking. The method has higher prediction precision, is an effective method for predicting navigation parameters, and is influenced by factors such as strong magnetic field environment, topography and the like, so that the integrated navigation system is easy to be interfered; the existing neural network-based method only learns data of a small number of routes, is easily influenced by measurement noise and model uncertainty, and has poor anti-interference capability of predicting navigation parameters. According to the invention, the characteristics of the NP capable of adaptively learning the random variable data set and the advantage of KF on noise processing are fully utilized, and navigation parameter estimation can be realized with higher precision only by relying on a gyroscope and an accelerometer under the condition of GNSS unlocking.
The invention combines LSTM with NP to make LSTM pre-process sensor data, and reduces NP dependence on data format by utilizing LSTM processing capability of time sequence. Secondly, the invention combines NP and KF, takes the state parameter and variance predicted by NP as the observed value and observed value variance of KF, and inputs the navigation parameter corrected by KF into NP for the prediction of the next period. Finally, the invention estimates the position and speed information by utilizing NP, estimates the variance of the attitude information and the attitude information by utilizing the predicted value and the variance, and takes the estimated value and the variance of the navigation information as KF observation values.
Drawings
For a clearer description of the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of the system of the present invention.
FIG. 3 is a flow chart illustrating a navigation method according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of NP model.
Fig. 5 is a schematic diagram of KF calculation.
Fig. 6 is a diagram illustrating an experimental platform according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
In the description of the embodiments of the present invention, it should be noted that, if the terms "upper," "lower," "horizontal," "inner," and the like indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, or the azimuth or the positional relationship in which the inventive product is conventionally put in use, it is merely for convenience of describing the present invention and simplifying the description, and does not indicate or imply that the apparatus or element to be referred to must have a specific azimuth, be configured and operated in a specific azimuth, and thus should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like, are used merely to distinguish between descriptions and should not be construed as indicating or implying relative importance.
Furthermore, the term "horizontal" if present does not mean that the component is required to be absolutely horizontal, but may be slightly inclined. As "horizontal" merely means that its direction is more horizontal than "vertical", and does not mean that the structure must be perfectly horizontal, but may be slightly inclined.
In the description of the embodiments of the present invention, it should also be noted that, unless explicitly specified and limited otherwise, the terms "disposed," "mounted," "connected," and "connected" should be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
The invention is described in further detail below with reference to the attached drawing figures:
referring to fig. 1, the embodiment of the invention discloses a navigation method based on a neural process and kalman filtering, which comprises the following steps:
step S1: for a motion carrier with a certain motion mode, when the GNSS works normally, a Kalman filter KF is utilized to solve navigation information of the speed and the gesture of the motion carrier, the navigation data, a gyroscope and an accelerometer are cooperated to form a training data set, the data set takes sensor data and navigation parameters at the same time as variables, and the navigation parameters at the later time are taken as target values;
step S2: and (3) randomly taking 200 groups of data from the training data set in the S1 as a group, dividing each group of data into a context set and a target set according to the proportion of 3:1, taking K groups of data, and establishing an NP model of the navigation method, wherein the NP model combines the advantages of a neural network and a Gaussian process, and the NP model is integrally formed by an encoder and a decoder. The encoder respectively applies target features and uncertainty of deterministic path and hidden path learning data, and firstly, the encoder pre-processes sensor data by using a Long Short-Term Memory (LSTM) model to acquire a feature sequence g. The deterministic path processes the feature sequence g and navigation parameters using a Multi-Layer Perceptron (MLP), and calculates high-dimensional feature values V of the sensor data sequence using Multi-head attention. The pre-processed long-short-term memory model LSTM is a single-layer 128-node network. The multilayer perceptron MLP of the processing characteristic sequence has the same number of neurons of an input layer as that of nodes of each layer of the preprocessing LSTM, and comprises two hidden layers, each hidden layer comprises 128 nerve units, and the output layer also comprises 128 neurons. The multi-head attention calculates an attention head by utilizing a plurality of points to multiply an attention matrix and a softmax function, and outputs a high-dimensional characteristic value V of the multi-head attention through linear calculation of the attention head. The hidden path utilizes a multi-layer perceptron MLP to process a feature sequence g and navigation parameters to generate a group of high-dimensional feature tensors s, averages the high-dimensional feature tensors s, respectively obtains a mean vector mu and a variance vector sigma of a Gao Weiyin variable z through two double-layer 128-node full-connection multi-layer perceptron MLPs, and randomly samples to obtain the Gao Weiyin variable z. The Gao Weiyin variable z follows a multidimensional gaussian distribution. The structure of the hidden path multi-layer perceptron MLP is the same as the deterministic path multi-layer perceptron MLP, and the decoder utilizes the multi-layer perceptron MLP to process the characteristic sequence g, the high-dimensional characteristic value V and the Gao Weiyin variable z to estimate navigation parameters. The number of the MLP input layer nerve units of the multi-layer perceptron of the decoder is determined by input data, 128 units are included in the multi-layer perceptron, each hidden layer comprises 128 nerve cells, the output layer consists of 2 nerve cells, and a predicted value and a predicted variance are respectively output;
step S3: training an NP model, randomly acquiring K groups of samples from a training set, wherein the samples cover all motion modes, the number of the samples is not less than 20% of the data volume of a testing set, randomly acquiring Q groups of samples from a testing machine, the number of the samples is not less than 30% of the data volume of the testing set, inputting selected training data into the NP for training, and adopting the evidence infinitesimal to confirm as a loss function; repeating training until the training result converges to obtain a navigation parameter prediction model;
step S4: updating the attitude information by utilizing KF by means of a system dynamics equation to realize the joint estimation of the pose of the motion carrier, inputting the updated navigation parameters into a navigation parameter prediction model trained in the step S4, and estimating the navigation parameters at the later moment; and repeatedly updating the attitude information and estimating navigation parameters at the later moment until the navigation task is completed or the GNSS is stabilized again.
The principle of the invention is as follows:
according to the invention, the LSTM algorithm is used for processing the time sequence data, so that important features in the original data are primarily extracted, the influence of the data structure on the NP prediction result is reduced, the robustness of the system on the data structure is enhanced, and the navigation precision is improved; dividing a training set theory into a training content set and a training target set, training an NP by utilizing the data characteristics related between the content set and the training target set, establishing a navigation parameter prediction model based on sensor data, fully utilizing the correlation of a motion mode and better overcoming the characteristic of less reliable data when GNSS is out of lock; the NP is used for understanding the navigation parameters as a group of random variables, the advantages of the random variables processed by the Gaussian process and the advantages of the Gaussian process kernel function can be adaptively learned by the NP, and various information of sensor data can be fully extracted to predict the navigation parameters.
The method can be used for online prediction of the navigation parameters of the flight platform, for the real-time sensor data of the flight platform, the training data and the nearby data set are used as a content set, the new data and the navigation parameters are combined to be used as a target set in the step S3, and the content set and the target set are input into an NP model to realize online prediction of the navigation parameters of the flight platform.
The invention uses a multi-head attention mechanism in the deterministic path of the encoder to process the feature sequence g output by the LSTM network and the high-dimensional feature value V output by the MLP, the attention mechanism can fully learn the global feature and the local feature of the feature sequence, and give more weight to more important features in the process of learning the content set and the target set, and simultaneously improve the learning efficiency and the prediction effect.
The LSTM network can extract key characteristics of data of the target set and the content set, and when similar characteristics appear, the multi-head attention mechanism can prompt the network to focus on learning and processing the characteristics, so that the processing capacity of NP on the characteristics is improved, and further, the network learning efficiency and the prediction effect are improved.
As shown in fig. 2, the embodiment of the invention discloses a navigation system method based on a neural process and kalman filtering, which comprises the following steps:
the data set generation module is used for solving navigation data of the speed and the gesture of the motion carrier, and combining the navigation data, the gyroscope and the accelerometer data into a training data set;
the NP model building module is used for grouping data in the training data set and building an NP model of the navigation method;
the NP model training module is used for selecting training data to train the NP model until the training result converges to obtain a navigation parameter prediction model;
the navigation parameter estimation module is used for updating the gesture information of the motion carrier, inputting the updated gesture information into the navigation parameter prediction model and estimating the navigation parameter at the later moment; and repeatedly updating the attitude information of the motion carrier until the navigation task is completed or the GNSS is stabilized again.
Examples:
the embodiment discloses a navigation method based on a neural process and Kalman filtering, the implementation flow is shown in fig. 3, and the specific method comprises the following steps:
step S1: for a motion carrier with a certain motion mode, when a GNSS works normally, the navigation information of the speed and the gesture of the motion carrier is solved by utilizing KF, the navigation data, the gyroscope and the acceleration data are cooperated to form a training data set, the data set has N groups of samples, the data set takes sensor data and navigation parameters at the same time as variables, and the navigation parameters at the later time are taken as target values;
aiming at a certain aircraft platform, the embodiment records various sensor data including a gyroscope, an accelerometer, a GNSS and the like at the frequency of 50Hz, solves navigation parameters such as attitude, speed and the like of the aircraft platform by using KF after downward alignment, acquires data for 10 minutes, and can acquire 30000 groups of data, and takes all the data as an original training data set.
Step S2: and randomly taking 200 groups of data as a group in the training data set in the S1, dividing each group of data into a content set and a target set according to the proportion of 3:1, taking K groups of data, and establishing an NP model of the navigation method.
As shown in fig. 4, the NP model combines the advantages of the neural network and the gaussian process, and the NP model is integrally formed by an encoder and a decoder, the encoder respectively applies the target features and uncertainties of deterministic path and hidden path learning data, the encoder firstly processes sensor data using Long Short-Term Memory (LSTM) to obtain a feature sequence g, the deterministic path processes the feature sequence g and navigation parameters using Multi-Layer Perceptron (MLP) and calculates to obtain a high-dimensional feature value V of the sensor data sequence using Multi-head attention, the preprocessing LSTM is a single-Layer 128-node network, the MLP of the processed feature sequence, the number of neurons of an input Layer is the same as the number of nodes of each Layer of preprocessing LSTM, each hidden Layer comprises two hidden layers, each hidden Layer comprises 128 nerve units, the output Layer also comprises 128 neurons, the Multi-head attention calculates attention heads by utilizing a plurality of point multiplication attention matrixes and softmax functions, the Multi-head attention high-dimensional characteristic values V are output by utilizing linear calculation of the attention heads, the hidden paths process characteristic sequences g and navigation parameters by utilizing MLP to generate a group of high-dimensional characteristic tensors s, average the s and respectively acquire average vectors and variance vectors of Gao Weiyin variable z through two double-Layer 128-node full-connection MLPs, the Gao Weiyin variable z is acquired by random sampling, the z obeys multidimensional Gaussian distribution, the hidden path MLP structure is identical with a deterministic path MLP, the decoder utilizes MLP to process the g, V and z to estimate navigation parameters, the number of MLP input Layer nerve units of the decoder is determined by input data, the hidden layers are 128 units, each hidden Layer comprises 128 neurons, the output layer is composed of 2 neurons and respectively outputs a predicted value and a predicted variance;
the existing neural network-based method is often difficult to obtain high-precision estimation under the influence of measurement noise and model uncertainty, and when the gesture of a moving carrier changes, the estimation precision of the position and the speed can be rapidly reduced. The NP converts the modeling process of the Gaussian process into the calculation of the high-dimensional random variable by calculating the high-dimensional random variable, the advantage of the Gaussian process on the random variable estimation is utilized, the process of learning a kernel function of the Gaussian process is improved, the relation between navigation parameters and sensor data can be adaptively learned, and accurate regression estimation is realized.
Step S3: training an NP model, randomly collecting K groups of samples from a training set, wherein the number of the samples is not less than 30% of the data amount of the training set, randomly collecting Q groups of samples from a testing machine, wherein the samples cover all motion modes, the number of the samples is not less than 20% of the data amount of the testing set, inputting selected training data into the NP for training, and adopting the evidence lower-boundary as a loss function;
the original training data set in the step S1 is subjected to data screening, 20 groups of training data sets which are different in similar movement modes are selected to serve as training data, data in each training set are sequentially connected front and back, the first 150 groups are context points, the second 50 groups are target sets, variables of the NP network comprise sensor data and aligned navigation parameters, the target values are navigation parameters at the later moment, and a loss function is calculated according to a evidence lower boundary formula:
ELBO(q)＝E[logp(z,x)]-E[logq(z)]
step S4: repeating the step S3 until the training result converges to obtain a navigation parameter prediction model;
when the loss function is trained to be more than 3, the NP network can obtain better estimation capability, the error maximum value of the independent estimated speed of the NP network is less than 0.6m/s, and the error is less than 0.2m/s when the NP network is stably estimated.
Step S5: by means of a system dynamics equation, the KF shown in fig. 5 is utilized to update gesture information, so that joint estimation of the gesture of the motion carrier is realized, updated navigation parameters are input into the NP after training in the step S4, and the navigation parameters at the later moment are estimated;
during estimation, the mean value and the variance of navigation parameters estimated by the NP network are input into KF by means of a system dynamics equation, updating of navigation parameters such as gestures and speeds is completed, a content set and a target set are updated by means of the corrected navigation parameters, online prediction can be achieved by means of reasonable utilization of the updated content set and the updated target set, and subsequent estimation is completed.
Step S6: and repeating the step S5 until the navigation task is completed or the GNSS is stabilized again.
The embodiment of the invention provides computer equipment. The computer device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. The steps of the various method embodiments described above are implemented when the processor executes the computer program. Alternatively, the processor may implement the functions of the modules/units in the above-described device embodiments when executing the computer program. Fig. 6 is a diagram illustrating an experimental platform according to an embodiment of the present invention.
The computer program may be divided into one or more modules/units, which are stored in the memory and executed by the processor to accomplish the present invention.
The computer equipment can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing equipment. The computer device may include, but is not limited to, a processor, a memory.
The processor may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like.
The memory may be used to store the computer program and/or modules, and the processor may implement various functions of the computer device by running or executing the computer program and/or modules stored in the memory, and invoking data stored in the memory.
The modules/units integrated with the computer device may be stored in a computer readable storage medium if implemented in the form of software functional units and sold or used as stand alone products. Based on such understanding, the present invention may implement all or part of the flow of the method of the above embodiment, or may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A navigation method based on a neural process and kalman filtering, comprising the steps of:
solving navigation data of the speed and the gesture of the motion carrier, and combining the navigation data with the gyroscope and the accelerometer data to form a training data set;
grouping data in the training data set, and establishing an NP model of a navigation method;
training the NP model by selecting training data until the training result converges to obtain a navigation parameter prediction model;
updating the attitude information of the motion carrier, inputting the updated attitude information into a navigation parameter prediction model, and estimating navigation parameters at the later moment; and repeatedly updating the attitude information and estimating navigation parameters at the later moment until the navigation task is completed or the GNSS is stabilized again.
2. The navigation method based on neural process and kalman filtering according to claim 1, wherein the navigation data for solving the velocity and the posture of the motion carrier is solved by using a kalman filter KF.
3. The neural process and kalman filter based navigation method of claim 1, wherein the grouping of data in the training dataset comprises:
200 data are randomly selected from the training data set to be a group, and each group of data is divided into a context set and a target set according to a ratio of 3:1.
4. The navigation method based on neural process and kalman filtering according to claim 1, wherein the NP model is composed of an encoder and a decoder;
the encoder learns target features and uncertainty of navigation data by using deterministic paths and hidden paths, and pre-processes sensor data by using a long-short-term memory model LSTM to obtain a feature sequence;
the deterministic path utilizes a multi-layer perceptron MLP to process the characteristic sequence and navigation data, and utilizes multi-head attention to calculate to obtain a high-dimensional characteristic value of a sensor data sequence; the long-term memory model LSTM for preprocessing is a single-layer 128-node network; the number of neurons of an input layer of the multi-layer perceptron MLP for the deterministic path is the same as that of nodes of each layer of a long-short-period memory model LSTM for preprocessing, the multi-layer perceptron MLP for the deterministic path comprises two hidden layers, each hidden layer comprises 128 nerve units, and an output layer comprises 128 nerve units; the multi-head attention utilizes a plurality of points to multiply an attention matrix and a softmax function to calculate an attention head, and a high-dimensional characteristic value of the multi-head attention is output through linear calculation of the attention head;
the hidden path utilizes a multi-layer perceptron MLP to process a feature sequence and navigation parameters, a group of high-dimensional feature tensors are generated, the high-dimensional feature tensors are averaged, a mean value vector and a variance vector are respectively obtained through two double-layer 128-node full-connection multi-layer perceptron MLPs, and random sampling is carried out to obtain Gao Weiyin variables; the Gao Weiyin variable obeys a multidimensional gaussian distribution; the structure of the multi-layer perceptron MLP for the hidden path is the same as that of the multi-layer perceptron MLP for the deterministic path;
the decoder utilizes a multi-layer perceptron MLP to process the characteristic sequence, the high-dimensional characteristic value and the Gao Weiyin variable to estimate navigation parameters; the number of input layer neural units of the multi-layer perceptron MLP for the decoder is determined by the input data.
5. The neural process and kalman filter based navigation method of claim 1, wherein the training data comprises samples randomly selected from a training set and a test set, in particular as follows:
randomly selecting a plurality of groups of samples from the training set, wherein the plurality of groups of samples cover all the motion modes and the number of the samples is not less than 20% of the data volume of the test set;
a plurality of groups of samples are randomly selected from the test set, and the number of the groups of samples is not less than 30% of the data volume of the test set.
6. The neural process and kalman filter based navigation method of claim 1 or 5, wherein the NP model is trained using a loss function of evidence infinitesimal bounds when the NP model is trained.
7. The navigation method based on neural process and kalman filter according to claim 1, wherein the updating of the pose information of the motion carrier is performed by using a kalman filter KF.
8. A navigation system method based on neural process and kalman filtering, comprising the steps of:
the data set generation module is used for solving navigation data of the speed and the gesture of the motion carrier, and combining the navigation data, the gyroscope and the accelerometer data into a training data set;
the NP model building module is used for grouping data in the training data set and building an NP model of the navigation method;
the NP model training module is used for selecting training data to train the NP model until the training result converges to obtain a navigation parameter prediction model;
the navigation parameter estimation module is used for updating the gesture information of the motion carrier, inputting the updated gesture information into the navigation parameter prediction model and estimating the navigation parameter at the later moment; and repeatedly updating the attitude information of the motion carrier until the navigation task is completed or the GNSS is stabilized again.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1-7 when the computer program is executed.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310125904.6A CN116026325A (en) | 2023-02-16 | 2023-02-16 | Navigation method and related device based on neural process and Kalman filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310125904.6A CN116026325A (en) | 2023-02-16 | 2023-02-16 | Navigation method and related device based on neural process and Kalman filtering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116026325A true CN116026325A (en) | 2023-04-28 |
Family
ID=86075994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310125904.6A Pending CN116026325A (en) | 2023-02-16 | 2023-02-16 | Navigation method and related device based on neural process and Kalman filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116026325A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116878535A (en) * | 2023-09-05 | 2023-10-13 | 杭州宇谷科技股份有限公司 | Intelligent power conversion guiding method and system based on hybrid time sequence network |
-
2023
- 2023-02-16 CN CN202310125904.6A patent/CN116026325A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116878535A (en) * | 2023-09-05 | 2023-10-13 | 杭州宇谷科技股份有限公司 | Intelligent power conversion guiding method and system based on hybrid time sequence network |
| CN116878535B (en) * | 2023-09-05 | 2023-12-12 | 杭州宇谷科技股份有限公司 | Intelligent power conversion guiding method and system based on hybrid time sequence network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | A hybrid prediction method for bridging GPS outages in high-precision POS application | |
| CN111366156A (en) | Transformer substation inspection robot navigation method and system based on neural network assistance | |
| CN109827579A (en) | The method and system of Filtering Model real time correction in a kind of integrated positioning | |
| CN115451952B (en) | Multi-system integrated navigation method and device for fault detection and robust adaptive filtering | |
| US11928813B2 (en) | Method and system for detecting change to structure by using drone | |
| CN111913175A (en) | Water surface target tracking method with compensation mechanism under transient failure of sensor | |
| CN116026325A (en) | Navigation method and related device based on neural process and Kalman filtering | |
| CN116381753B (en) | Neural network assisted navigation method of GNSS/INS integrated navigation system during GNSS interruption | |
| Han et al. | Quadratic-Kalman-filter-based sensor fault detection approach for unmanned aerial vehicles | |
| Li et al. | Novel hybrid algorithm of improved CKF and GRU for GPS/INS | |
| Yan et al. | An adaptive nonlinear filter for integrated navigation systems using deep neural networks | |
| Shi et al. | Tilt angle monitoring by using sparse residual LSTM network and grid search | |
| Jingsen et al. | Integrating extreme learning machine with Kalman filter to bridge GPS outages | |
| Qian et al. | An INS/DVL integrated navigation filtering method against complex underwater environment | |
| Li et al. | A new adaptive unscented Kalman filter based on covariance matching technique | |
| CN116608859A (en) | Navigation method, storage medium and device of self-adaptive unscented Kalman filtering based on threshold processing | |
| Yang et al. | Adaptively constrained Kalman filtering for navigation applications | |
| CN114018250B (en) | Inertial navigation method, electronic device, storage medium and computer program product | |
| CN114719850A (en) | Attitude estimation fusion method and system of multi-layer perception and volume Kalman filtering | |
| Zug et al. | An approach towards smart fault-tolerant sensors | |
| Li et al. | Neural network application on ship motion prediction | |
| Guo et al. | A novel self-learning gnss/ins integrated navigation method | |
| Malleswaran et al. | Real-time INS/GPS data fusion using hybrid adaptive network based fuzzy inference | |
| Xin et al. | Research on dual-threshold detection based adaptive fault-tolerant kalman filtering algorithm for dynamic 6-DOF measurement | |
| CN116499469B (en) | GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation |
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 |