CN116499469B - GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation - Google Patents
GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation Download PDFInfo
- Publication number
- CN116499469B CN116499469B CN202310769700.6A CN202310769700A CN116499469B CN 116499469 B CN116499469 B CN 116499469B CN 202310769700 A CN202310769700 A CN 202310769700A CN 116499469 B CN116499469 B CN 116499469B
- Authority
- CN
- China
- Prior art keywords
- neural network
- gnss
- ins
- data
- training
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000003062 neural network model Methods 0.000 title claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 claims abstract description 116
- 238000012549 training Methods 0.000 claims abstract description 70
- 230000008569 process Effects 0.000 claims abstract description 24
- 230000009131 signaling function Effects 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 17
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000004806 packaging method and process Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000011428 standard deviation standardization method Methods 0.000 claims description 2
- 238000011160 research Methods 0.000 description 8
- 239000004973 liquid crystal related substance Substances 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G01C21/20—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
- G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/13—Receivers
- G01S19/35—Constructional details or hardware or software details of the signal processing chain
- G01S19/37—Hardware or software details of the signal processing chain
-
- 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
- G01S19/45—Determining position by combining measurements of signals from the satellite radio beacon positioning system with a supplementary measurement
-
- 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
Abstract
The application relates to a GNSS/INS integrated navigation method utilizing neural network model on-line learning and compensation, belonging to the technical field of integrated navigation; comprising the following steps: when the GNSS is normal in the GNSS/INS integrated navigation process, alternately training the first neural network and the second neural network in a set period; one of the first neural network and the second neural network is always in a training state of a training process and the other is always in a trained standby state through cycle by cycle; the trained neural network is used for predicting a pseudo GNSS signal function according to the input INS data; when GNSS is lost in the navigation process, the neural network in a standby state is accessed into a combined structure of combined navigation; and (3) inputting INS data into an accessed neural network, and compensating INS original data by using pseudo GNSS signals predicted by the neural network through a Kalman filter to output a positioning result. The application can make predictions meeting the current environment of the user when the GNSS is lost, and has strong instantaneity and flexibility.
Description
Technical Field
The application relates to the technical field of integrated navigation, in particular to a GNSS/INS integrated navigation method for online learning and compensation by utilizing a neural network model.
Background
The GNSS/INS integrated navigation becomes a hot spot for the research of the navigation field due to the higher positioning precision and universality. However, when the GNSS signal is unlocked, the navigation accuracy is drastically reduced due to the accumulated error of the INS when the INS is used for the navigation service. Therefore, how to compensate the accumulated error of INS when the GNSS signal is out of lock and maintain the accuracy of integrated navigation in a high range is an important issue for improving the usability of integrated navigation of GNSS/INS.
In recent years, an artificial intelligence technology is raised, and a new direction is provided for solving the problem of reduced integrated navigation precision when GNSS signals lose lock. At present, two methods for solving the problem of GNSS signal unlocking in integrated navigation by using a neural network are mainly two methods, namely errors of pseudo GNSS signals output by the neural network and INS signals output by the neural network. Most of the current researches have defects, some researches do not consider the influence of the historical navigation information, most of the researches which consider the historical navigation information adopt more complex neural networks (such as CNN-LSTM series neural networks and the like), the training cost is high, the real-time training cannot be carried out, the existing data sets must be used for training before the current researches are put into use, the neural network models of the current researches cannot be predicted according to the current environment of users, and the accuracy and the flexibility of the current researches in practical application are reduced. Therefore, neural network-assisted integrated navigation systems capable of real-time training have value in exploration and research.
Disclosure of Invention
In view of the above analysis, the present application aims to disclose a GNSS/INS integrated navigation method using neural network model online learning and compensation; the method is used for improving the precision of GNSS/INS integrated navigation when the GNSS is out of lock.
The application discloses a GNSS/INS integrated navigation method utilizing neural network model on-line learning and compensation, comprising the following steps:
step S1, alternately training a first neural network and a second neural network in a set period when GNSS is normal in the GNSS/INS integrated navigation process; one of the first neural network and the second neural network is always in a training state of a training process and the other is always in a trained standby state through cycle by cycle; the trained neural network is used for predicting a pseudo GNSS signal function according to the input INS data;
step S2, when GNSS is lost in the navigation process, the neural network in a standby state is accessed into a combined structure of combined navigation; and (3) inputting INS data into an accessed neural network, and compensating INS original data by using pseudo GNSS signals predicted by the neural network through a Kalman filter to output a positioning result.
Further, the step S1 includes:
step S101, collecting INS original output data and position change data output by a GNSS when the GNSS works normally, and packaging the collected data in a set period;
step S102, denoising and standardizing data including specific force and angular velocity in the INS original data after the bagging treatment to obtain INS data for training a neural network;
step S103, training the first neural network or the second neural network which is set to be in a training state; in training, INS data for training a neural network is used as input characteristics of the training neural network, and the packed GNSS position variation is used as target data of output characteristics of the neural network; the training process does not exceed half a period;
step S104, setting the trained neural network to be in a standby state, and setting the other neural network to be in a training state for training; the training process is also not more than half a cycle;
after the next cycle is entered, steps S101-S104 are repeated, alternately training two neural networks.
Further, in step S101,
the collected INS original data consists of the specific force, angular velocity and speed of the triaxial at the current moment and the specific force, angular velocity and speed of the triaxial at the previous momentVector x (t);
the collected GNSS output data is composed of GNSS-resolved three-axis position variationVector y (t);
the GNSS data y (1) to y (m) of m continuous moments collected in a set period are packed into oneA matrix Y;
packaging INS data of m times corresponding to GNSS data into oneMatrix X.
Further, in step S102, the specific force f and the angular velocity ω in X are denoised by a wavelet threshold denoising algorithm.
Further, in a wavelet threshold denoising algorithm, a sym8 function is selected as a wavelet base, 3-level wavelet base decomposition is adopted, and a wavelet threshold is set by adopting a soft and hard threshold compromise method.
Further, wavelet thresholdThe method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>The average value of the absolute value of the wavelet coefficient of the high-frequency part after the first-stage wavelet decomposition is taken as m, and the m is the number of rows of the matrix X.
Further, after denoising, the data of each row of the matrix X is standardized by adopting a standard deviation standardization method to obtain a new matrix X new The method comprises the steps of carrying out a first treatment on the surface of the The new matrix X new The medium specific force and angular velocity data are denoising and normalization results, and the velocity data are normalization results.
Further, the first neural network and the second neural network are both MLP neural networks with the same structure.
Further, the MLP neural network comprises an input layer, two hidden layers and an output layer;
wherein the input layer is 18 neurons, respectively receives 18-dimensional input features, and the input features are formed by a matrix X new A line of data provided;
the output layer is 3 neurons and is used for outputting 3-dimensional output characteristics;
the two hidden layers are respectively 100-dimension and 50-dimension, the activation functions are both Relu functions, the optimizer selects an SGD optimizer, the learning rate is set to be 0.002, the root mean square error calculation loss function is adopted, and the model iteration training round number epoch variable is set to be 10000.
Further, a Dropout layer is added before the input layer and the first hidden layer, and 50% of weight is randomly ignored during training;
a momentum variable is added, set to a value of 0.9, to help the neural network jump out of local maximum.
The application can realize one of the following beneficial effects:
according to the GNSS/INS integrated navigation method utilizing the neural network model for online learning and compensation, the lightweight neural network is adopted for training, and the neural network can be trained based on the current environment and the motion state of the user in the use process of the user, so that the neural network can make predictions which more accord with the current environment of the user when the GNSS is lost, and the GNSS integrated navigation method has strong instantaneity and flexibility in application;
according to the application, the denoising algorithm is adopted to denoise the INS original signal, and the denoised INS signal is used as the input of the neural network, so that the prediction accuracy of the neural network is improved, and the problem that the prediction accuracy of the lightweight neural network is inferior to that of a complex neural network is solved.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the application, like reference numerals being used to designate like parts throughout the drawings;
FIG. 1 is a flowchart of a GNSS/INS integrated navigation method using neural network model online learning and compensation in an embodiment of the application;
FIG. 2 is a flowchart of a method for GNSS normal time data acquisition and neural network training in an embodiment of the present application;
FIG. 3 is a block diagram of a wavelet decomposition flow chart in an embodiment of the present application;
FIG. 4 is a block diagram of training patterns in an embodiment of the present application;
FIG. 5 is a block diagram illustrating the structure of data compensation in an embodiment of the present application.
Detailed Description
Preferred embodiments of the present application are described in detail below with reference to the attached drawing figures, which form a part of the present application and are used in conjunction with embodiments of the present application to illustrate the principles of the present application.
One embodiment of the present application discloses a GNSS/INS integrated navigation method for online learning and compensation by using a neural network model, as shown in FIG. 1, comprising:
step S1, alternately training a first neural network and a second neural network in a set period when GNSS is normal in the GNSS/INS integrated navigation process; one of the first neural network and the second neural network is always in a training state of a training process and the other is always in a trained standby state through cycle by cycle; the trained neural network is used for predicting a pseudo GNSS signal function according to the input INS data;
step S2, when GNSS is lost in the navigation process, the neural network in a standby state is accessed into a combined structure of combined navigation; and (3) inputting INS data into an accessed neural network, and compensating INS original data by using pseudo GNSS signals predicted by the neural network through a Kalman filter to output a positioning result.
Specifically, as shown in fig. 2, the step S1 includes:
step S101, collecting INS original output data and position change data output by a GNSS when the GNSS works normally, and packaging the collected data in a set period;
for example, the set period is 3 minutes, and the INS raw output data and the position change amount data of the GNSS output are collected at intervals of 3 minutes and packaged.
Wherein the collected INS original data consists of the specific force, angular velocity and speed of the triaxial at the current moment and the specific force, angular velocity and speed of the triaxial at the previous momentVector x (t);
namely:
；
wherein f represents the specific force, ω represents the angular velocity, v represents the velocity, t represents the current time, the subscripts x, y, z of the specific force and the angular velocity respectively represent the three axes of the carrier coordinate system, and the subscripts x, y, z of the velocity respectively represent the three axes of the northeast day coordinate system.
The collected GNSS output data is composed of GNSS-resolved three-axis position variationVector y (t);
namely:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,the index x, y, z represent the warp, weft, and high triaxial, respectively.
The GNSS data Y (1) to Y (m) collected in a set period and at m continuous moments are packed into a matrix Y, Y being oneIs a matrix of (a); the p-th row is the data y (p) output by the GNSS at the p-th moment, wherein p is an integer and p is E [1, m]。
；
Packaging INS data of m times corresponding to GNSS data into matrix X, wherein X is oneIs a matrix of (a) in the matrix.
The q-th row of the matrix X is the data X (q) output by the INS at the q-th moment of GNSS operation, wherein q is an integer and q is E [1, m ].
；
Since the data output frequency of INS is often much higher than that of GNSS. Therefore, the amount of data collected by INS over a set period is typically much greater than m. In this step, the m INS data corresponding to the time of the GNSS data need to be taken and packaged.
Step S102, denoising and standardizing data including specific force and angular velocity in the INS original data after the bagging treatment to obtain INS data for training a neural network;
preferably, a wavelet threshold denoising algorithm is adopted to denoise the specific force f and the angular velocity omega in X.
In the wavelet threshold denoising algorithm, a Mallet algorithm is adopted for wavelet decomposition, and a flow chart is shown in fig. 3. The flow of the Malet algorithm is to decompose the signal into two parts of low-frequency approximation and high-frequency detail, and then to decompose the low-frequency part again to obtain the low-frequency approximation and high-frequency detail of the stage, and the like, to subdivide the signals step by step, so that wavelet decomposition of the required layer number can be performed according to the resolution requirement in practical application.
Classical wavelet basis functions are mainly haar wavelets, dbN wavelets, coifN wavelets, symlet wavelets, meyer wavelets, etc.
In the wavelet threshold denoising algorithm of this embodiment, sym8 function is selected as wavelet base, 3-level wavelet base decomposition is adopted, and a wavelet threshold is set by adopting a method of soft-hard threshold compromise.
The sym8 function is a symlet8 wavelet basis function, and is a wavelet basis function in a symlet wavelet system.
The threshold function formula is as follows:
；
wherein W represents a wavelet coefficient before denoising, Q represents a wavelet coefficient after denoising, k represents coordinates of a wavelet domain, alpha is a parameter, the value range is [0,1], in the example, alpha takes 0.5, lambda is a selected threshold value,
in this embodiment, the formula for calculating the wavelet threshold is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,the average value of the absolute value of the wavelet coefficient of the high-frequency part after the first-stage wavelet decomposition is taken as m, and the m is the original data quantity, namely the number of rows of the matrix X.
Specifically, after denoising, a Standard deviation standardization (Standard Scale) method is adopted to perform standardization processing on each data line, and the step is to improve the training speed of the neural network and simultaneously has the effect of avoiding the overfitting of the neural network. The standard deviation normalized formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,represents normalized values, ++>Representing the mean value of the dataset, +.>Representing the standard deviation of the dataset.
With denoised and normalized results f new 、ω new 、v new Replacing f, omega and v in original matrix X to obtain new matrix X new In (a):
；
wherein f new 、ω new The results of f, omega after denoising and normalization are respectively v new The result is obtained by normalizing v.
The three-axis specific force and angular velocity of the current moment and the last moment after wavelet threshold denoising and the velocity of the current moment and the last moment obtained directly from INS output are oneIs a matrix of (a) in the matrix. The neural network is trained by adopting the INS data after denoising and standardization processing, the data volume is smaller, the real-time performance is better, and the neural network training under the condition of on-line acquisition is facilitated.
Step S103, training the first neural network or the second neural network which is set to be in a training state; in training, INS data for training a neural network is used as input characteristics of the training neural network, and the packed GNSS position variation is used as target data of output characteristics of the neural network; the training process does not exceed half a set period;
step S104, setting the trained neural network to be in a standby state, and setting the other neural network to be in a training state for training; the training process also does not exceed half a set period;
upon entering the next cycle, steps S101-S104 are repeated, alternately training two neural networks.
The first neural network and the second neural network are MLP neural networks with the same structure. The MLP neural network model has the advantages of simple structure, high training speed and strong generalization capability, and meets the requirement of training the neural network based on the current environment and motion state of the user in the use process of the user.
Specifically, the MLP neural network adopted in the embodiment includes an input layer, two hidden layers, and an output layer;
wherein the input layer is 18 neurons, respectively receives 18-dimensional input features, and the input features are formed by a matrix X new Providing;
the output layer is 3 neurons and is used for outputting 3-dimensional output characteristics; during training, the target data of the output characteristics are provided by a matrix Y;
the two hidden layers are respectively 100-dimension and 50-dimension, the activation function adopts a Relu function, the optimizer selects an SGD optimizer, the learning rate is set to be 0.002, the root Mean Square Error (MSE) is adopted to calculate the loss function, and the model iteration training round number epoch variable is set to be 10000.
In order to avoid overfitting, in addition to normalizing the data, the neural network of the present embodiment further includes:
adding a Dropout layer before the input layer and the first hidden layer, and randomly ignoring 50% of weight during training;
a momentum variable is added, set to a value of 0.9, to help the neural network jump out of local maximum.
Through the improvement, the training speed can be increased while the overfitting is avoided, namely the Dropout layer reduces the weight quantity in calculation, and the momentum increases the convergence speed of the loss zone number.
Through the neural network training process of the embodiment, the newly trained neural network is in a standby state, and the previously trained neural network is in a retraining state, so that when the GNSS works normally, one neural network is always in the standby state, the other neural network is in the training state, the neural network in the standby state is always trained by using the latest data, and when the neural network in the standby state is used for prediction, the environment where a user is currently located is more consistent with strong instantaneity and flexibility in application.
As shown in FIG. 4, a block diagram of a training pattern for implementing the above steps in a GNSS/INS integrated navigation system is presented. In the figure, the denoising module is configured to implement the denoising method in step S102, where the neural network includes a first neural network and a second neural network, and the KF module is an existing kalman filter structure.
Similarly, the training mode in the above steps can be applied to a GNSS/INS tightly integrated navigation system, and the technical scheme of tightly integrated navigation by using GNSS information and INS information can refer to the existing tightly integrated navigation technical scheme.
In step S2, when the GNSS is lost in the navigation process, the neural network in the standby state is accessed to the combined structure of the combined navigation, the INS data is input to the neural network in the standby state to predict the pseudo GNSS signals, the neural network input is the same as the input used in training, the three-axis position variation of the predicted pseudo GNSS signals is output, and the three-axis position variation is output to the kalman filter to compensate the INS raw data and then output the positioning result.
As shown in fig. 5, a block diagram of a structure for implementing data compensation in a GNSS/INS integrated navigation system is provided. The denoising module in the figure is used for implementing the denoising method in step S102, and the neural network adopts a combined structure of neural network access combination navigation in a standby state.
Similarly, the data compensation in the above steps can be applied to a GNSS/INS integrated navigation system, and the technical scheme of performing integrated navigation by using GNSS information and INS information can refer to the existing integrated navigation technical scheme.
In summary, the GNSS/INS integrated navigation method using the neural network model for online learning and compensation disclosed by the embodiment of the application adopts the lightweight neural network for training, and can train the neural network based on the current environment and the motion state of the user in the use process of the user, so that the neural network can make predictions more in line with the current environment of the user when the GNSS is lost, and has strong instantaneity and flexibility in application;
denoising the INS original signal by adopting a denoising algorithm, and taking the denoised INS signal as the input of the neural network, so that the prediction accuracy of the neural network is improved, and the problem that the prediction accuracy of the lightweight neural network is inferior to that of a complex neural network is solved.
The present application is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present application are intended to be included in the scope of the present application.
Claims (5)
1. The GNSS/INS integrated navigation method for online learning and compensation by utilizing the neural network model is characterized by comprising the following steps of:
step S1, alternately training a first neural network and a second neural network in a set period when GNSS is normal in the GNSS/INS integrated navigation process; one of the first neural network and the second neural network is always in a training state of a training process and the other is always in a trained standby state through cycle by cycle; the trained neural network is used for predicting a pseudo GNSS signal function according to the input INS data;
step S2, when GNSS is lost in the navigation process, the neural network in a standby state is accessed into a combined structure of combined navigation; the INS data are input into an accessed neural network, and the pseudo GNSS signals predicted by the neural network are utilized to compensate INS original data through a Kalman filter to output a positioning result;
step S101, collecting INS original output data and position change data output by a GNSS when the GNSS works normally, and packaging the collected data in a set period;
step S102, denoising and standardizing data including specific force and angular velocity in the INS original data after the bagging treatment to obtain INS data for training a neural network;
step S103, training the first neural network or the second neural network which is set to be in a training state; in training, INS data for training a neural network is used as input characteristics of the training neural network, and the packed GNSS position variation is used as target data of output characteristics of the neural network; the training process does not exceed half a period;
step S104, setting the trained neural network to be in a standby state, and setting the other neural network to be in a training state for training; the training process is also not more than half a cycle;
repeating the steps S101-S104 after entering the next period, and alternately training two neural networks;
in the step S101 of the process,
the collected INS original data consists of the specific force, angular velocity and speed of the triaxial at the current moment and the specific force, angular velocity and speed of the triaxial at the previous momentVector x (t);
the collected GNSS output data is composed of GNSS-resolved three-axis position variationVector y (t);
the GNSS data y (1) to y (m) of m continuous moments collected in a set period are packed into oneA matrix Y;
packaging INS data of m times corresponding to GNSS data into oneA matrix X;
in step S102, a wavelet threshold denoising algorithm is adopted to denoise the specific force f and the angular velocity ω in X;
in a wavelet threshold denoising algorithm, a sym8 function is selected as a wavelet base, 3-level wavelet base decomposition is adopted, and a wavelet threshold is set by adopting a soft and hard threshold compromise method;
wavelet thresholdThe method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>The average value of the absolute value of the wavelet coefficient of the high-frequency part after the first-stage wavelet decomposition is taken as m, and the m is the number of rows of the matrix X.
2. The method for integrated navigation of GNSS/INS using neural network model learning and compensation in accordance with claim 1, wherein,
after denoising, carrying out standardization processing on each column of data of the matrix X by adopting a standard deviation standardization method to obtain a new matrix X new The method comprises the steps of carrying out a first treatment on the surface of the The new matrix X new The medium specific force and angular velocity data are denoising and normalization results, and the velocity data are normalization results.
3. The method for integrated navigation of GNSS/INS using neural network model learning and compensation in accordance with claim 2, wherein,
the first neural network and the second neural network are MLP neural networks with the same structure.
4. The method for integrated GNSS/INS navigation using neural network model online learning and compensation as claimed in claim 3, wherein,
the MLP neural network comprises an input layer, two hidden layers and an output layer;
wherein the input layer is 18 neurons, respectively receives 18-dimensional input features, and the input features are formed by a matrix X new A line of data provided;
the output layer is 3 neurons and is used for outputting 3-dimensional output characteristics;
the two hidden layers are respectively 100-dimension and 50-dimension, the activation functions are both Relu functions, the optimizer selects an SGD optimizer, the learning rate is set to be 0.002, the root mean square error calculation loss function is adopted, and the model iteration training round number epoch variable is set to be 10000.
5. The method for integrated navigation of GNSS/INS using neural network model online learning and compensation of claim 4, wherein,
adding a Dropout layer before the input layer and the first hidden layer, and randomly ignoring 50% of weight during training;
a momentum variable is added, set to a value of 0.9, to help the neural network jump out of local maximum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310769700.6A CN116499469B (en) | 2023-06-28 | 2023-06-28 | GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310769700.6A CN116499469B (en) | 2023-06-28 | 2023-06-28 | GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116499469A CN116499469A (en) | 2023-07-28 |
| CN116499469B true CN116499469B (en) | 2023-09-08 |
Family
ID=87320630
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310769700.6A Active CN116499469B (en) | 2023-06-28 | 2023-06-28 | GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116499469B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104898148A (en) * | 2015-06-02 | 2015-09-09 | 北京航空航天大学 | Low-cost INS/GPS seamless navigation method based on data compression and neural network |
| CN109883547A (en) * | 2019-01-10 | 2019-06-14 | 武汉大学 | A kind of wide-band spectrum signal antinoise method based on wavelet threshold difference |
| 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 |
| CN110501169A (en) * | 2019-08-27 | 2019-11-26 | 北理慧动（常熟）车辆科技有限公司 | Diagnostic method, device and the electronic equipment of vehicle trouble |
| CN111127646A (en) * | 2019-12-26 | 2020-05-08 | 西南林业大学 | Construction method and system of rasterized elevation curved surface for measuring height difference of landform |
| WO2021038485A1 (en) * | 2019-08-27 | 2021-03-04 | Indian Institute Of Science | System and method for autonomous navigation of unmanned aerial vehicle (uav) in gps denied environment |
| CN112505737A (en) * | 2020-11-16 | 2021-03-16 | 东南大学 | GNSS/INS combined navigation method based on Elman neural network online learning assistance |
-
2023
- 2023-06-28 CN CN202310769700.6A patent/CN116499469B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104898148A (en) * | 2015-06-02 | 2015-09-09 | 北京航空航天大学 | Low-cost INS/GPS seamless navigation method based on data compression and neural network |
| CN109883547A (en) * | 2019-01-10 | 2019-06-14 | 武汉大学 | A kind of wide-band spectrum signal antinoise method based on wavelet threshold difference |
| CN110501169A (en) * | 2019-08-27 | 2019-11-26 | 北理慧动（常熟）车辆科技有限公司 | Diagnostic method, device and the electronic equipment of vehicle trouble |
| WO2021038485A1 (en) * | 2019-08-27 | 2021-03-04 | Indian Institute Of Science | System and method for autonomous navigation of unmanned aerial vehicle (uav) in gps denied environment |
| 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 |
| CN111127646A (en) * | 2019-12-26 | 2020-05-08 | 西南林业大学 | Construction method and system of rasterized elevation curved surface for measuring height difference of landform |
| CN112505737A (en) * | 2020-11-16 | 2021-03-16 | 东南大学 | GNSS/INS combined navigation method based on Elman neural network online learning assistance |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116499469A (en) | 2023-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Noureldin et al. | GPS/INS integration utilizing dynamic neural networks for vehicular navigation | |
| Abdolkarimi et al. | Wavelet-adaptive neural subtractive clustering fuzzy inference system to enhance low-cost and high-speed INS/GPS navigation system | |
| CN113486574B (en) | Sound velocity profile completion method and device based on historical data and machine learning | |
| CN108334987B (en) | Wavelet decomposition-neural network-based sea wave height prediction method | |
| CN110926466A (en) | Multi-scale data blocking algorithm for unmanned ship combined navigation information fusion | |
| Hasan et al. | Optimizing of ANFIS for estimating INS error during GPS outages | |
| CN113411744A (en) | High-precision indoor positioning and tracking method | |
| Shi et al. | Tilt angle monitoring by using sparse residual LSTM network and grid search | |
| Lu et al. | Fusion algorithm-based temperature compensation method for High-G MEMS accelerometer | |
| CN114777771A (en) | Outdoor unmanned vehicle combined navigation positioning method | |
| CN116499469B (en) | GNSS/INS combined navigation method utilizing neural network model on-line learning and compensation | |
| Li et al. | An improved cyclic multi model-eXtreme gradient boosting (CMM-XGBoost) forecasting algorithm on the GNSS vertical time series | |
| CN110501686A (en) | A kind of method for estimating state based on NEW ADAPTIVE high-order Unscented kalman filtering | |
| Liu et al. | Navigability analysis of local gravity map with projection pursuit-based selection method by using gravitation field algorithm | |
| CN116026325A (en) | Navigation method and related device based on neural process and Kalman filtering | |
| CN108120981B (en) | Shallow water depth radar remote sensing detection method | |
| CN111008549A (en) | UUV platform DVL signal distortion reconstruction method based on sample entropy and IFOA-GRNN | |
| CN115859116A (en) | Marine environment field reconstruction method based on radial basis function regression interpolation method | |
| Akyılmaz et al. | GPS monitoring of the Fatih Sultan Mehmet suspension bridge by using assessment methods of neural networks | |
| CN108960406B (en) | MEMS gyroscope random error prediction method based on BFO wavelet neural network | |
| WO2020016454A1 (en) | Method and apparatus for actively tuning a predictor to an input signal | |
| CN109740111A (en) | Electric field value prediction algorithm over the ground | |
| Zheng et al. | A multi-factor data-driven prediction model for cyanobacteria blooms in lakes and reservoirs | |
| Patel et al. | Development of Wiener-Hammerstein Models Parameterized using Orthonormal Basis Filters and Deep Neural Network | |
| CN114577204B (en) | Anti-interference self-alignment method and device for strapdown inertial navigation system based on neural network |
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 |