CN110044350A - The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network - Google Patents
The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network Download PDFInfo
- Publication number
- CN110044350A CN110044350A CN201910300780.4A CN201910300780A CN110044350A CN 110044350 A CN110044350 A CN 110044350A CN 201910300780 A CN201910300780 A CN 201910300780A CN 110044350 A CN110044350 A CN 110044350A
- Authority
- CN
- China
- Prior art keywords
- output
- random drift
- neural network
- recurrent neural
- 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
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
-
- 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
Abstract
The invention discloses a kind of methods modeled using dynamic recurrent neural network is improved to the random drift of MEMS gyro, this method be introduced into Denoising Algorithm by MEMS gyro export in high frequency white noise and low frequency random drift separate, the random drift after denoising is trained using dynamic recurrent neural network is improved, establish relationship model of the non-stationary wander sequences in the past between current time, increase output layer node feeding back to improve network structure, it realizes the real-time prediction to Modelling of Random Drift of Gyroscopes variation tendency, effectively improves the precision of MEMS inertial navigation system.
Description
Technical field
The present invention relates to technical field of inertial, more particularly to a kind of application enhancements dynamic recurrent neural network
MEMS gyro random drift modeling method.
Background technique
Currently, the important branch as inertial navigation field, the inertial navigation system based on MEMS (MEMS) has
Have the advantages that it is small in size, at low cost, be easily installed, be light-weight, high reliability and impact resistance, it is military in normal domestic and strategy
The fields such as navigation have broad application prospects.However, being influenced by manufacturing process and use environment, MEMS inertia device and biography
Inertia device of uniting is lower compared to precision, wherein the lower signal-to-noise ratio of MEMS gyro, which becomes, restricts MEMS inertial navigation system precision improvement
One of principal element.MEMS gyro error is broadly divided into ascertainment error and random drift two parts, ascertainment error parameter
It can be obtained by calibration experiment, establish accurate mathematical model and be compensated；Random drift is non-stationary, irregular slow time-varying
Signal, it is difficult to obtain its true model.As the important errors source of MEMS gyro, the research of the modeling compensation method of random drift
It is particularly important for MEMS inertial navigation system precision improvement.In existing random drift modeling compensation method, floated to Gyro Random
The methods of time series analysis, wavelet theory and neural network, and common random drift model can be generally used when moving modeling
It is the arma modeling based on stable time rank analysis.
But arma modeling is difficult to completely describe the time-varying and nonlinear characteristic of gyroscopic drift, is based on stationary time series
Modeling method inevitably result in the inaccuracy of model.Therefore, seeking the modeling method based on nonstationary time series becomes
Promote the important research direction of gyroscopic drift modeling accuracy.
As one of common intelligent optimization method, neural network have unique adaptive learning, nonlinear transformation with
And the features such as parallel information processing capacity, before the fields such as pattern-recognition, signal processing, System Discrimination all have a wide range of applications
Scape, neural network provide a kind of effective way by Nonlinear Modeling of its outstanding advantage, also model in the random drift of gyro
In obtain extensive concern.Currently, counterpropagation network (BP), radial base neural net (RBF), support vector machines (SVM) etc. are calculated
Method is widely used in the random drift modeling of gyro, but exists and easily fall into that locally optimal solution, convergence rate be slow and over-fitting
Etc. limitations.
Therefore, how a kind of MEMS gyro random drift modeling method that can effectively promote MEMS inertial navigation system precision is provided
The problem of as those skilled in the art's urgent need to resolve.
Summary of the invention
In view of this, the present invention provides a kind of MEMS gyro random drift modeling sides of application enhancements dynamic recurrent neural network
Method, this method be introduced into Denoising Algorithm by MEMS gyro export in high frequency white noise and low frequency random drift separate, use
It improves dynamic recurrent neural network neural network to be trained the random drift after denoising, establishes non-stationary wander sequences and go over
Relationship model between current time increases output layer node feeding back to improve network structure, realizes to Modelling of Random Drift of Gyroscopes
The real-time prediction of variation tendency, effectively improves the precision of MEMS inertial navigation system.
To achieve the goals above, the present invention adopts the following technical scheme:
A kind of MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network, this method include following step
It is rapid:
The output error models of MEMS gyro are established, the main error source of analyzing influence gyro output obtains high frequency white noise
Sound and low frequency random drift the two main error sources；
MEMS gyro Static output data are acquired, high frequency white noise and low frequency random drift are separated using Denoising Algorithm, mentioned
Take the random drift in gyro output signals；
The random drift extracted is inputted as training data and improves dynamic recurrent neural network, a large amount of training improve dynamic and pass
Return network, obtains optimal network model；
Untrained test data is inputted into optimal network model, obtains the random drift prediction output of MEMS gyro, it is complete
The dynamic modeling of pairs of MEMS gyro random drift.
On the basis of above scheme, specific explanations explanation is done to scheme provided by the invention.
Further, the output error models of the MEMS gyro are as follows:
ω (t)=ωiesinL+εd+Dr+W(t) (1)
Wherein, ωieFor earth rotation angular speed；L is local latitude；εdFor constant value zero bias, i.e., when input angular velocity is zero
When gyro export constant, can be compensated by calibration experiment；DrFor drift error, there is randomness, tendency and period
Property；W (t) is the zero-mean white noise with time correlation.
Further, dynamic recurrent neural network is inputted using the random drift extracted as training data, a large amount of training improve
Dynamic recurrent neural network obtains optimal network model, specifically includes the following steps:
1) using the random drift that extracts as the sample data for improving dynamic recurrent neural network, network inputs are last time
MEMS gyro random drift, setting input layer include 4 nodes, respectively represent the random drift data at preceding 4 moment；Network
Output is the random drift of the MEMS gyro at current time, and setting output layer includes 1 node, represents the random drift at current time
Move data；Hidden layer output is inputted by the delay and storage feedback of articulamentum to hidden layer, and it is 2 that hidden layer, which is arranged, according to heuristic
Layer, each hidden layer include 10 nodes, improve four layers of neural network that dynamic recurrent neural network is four inputs, singly exports, number
Learn model are as follows:
X (k)=f (wl1xc(k)+wl2u(k-1)+wl4yc(k)) (2)
xc(k)=α xc(k-1)+x(k-1) (3)
yc(k)=γ yc(k-1)+y(k-1) (4)
Y (k)=g (wl3x(k)) (5)
Wherein, wl1For the connection weight matrix for connecting node layer and hidden node, wl2For the company of input unit and Hidden unit
Meet weight matrix, wl3For the connection weight matrix of hidden node and output unit, wl4For the articulamentum section with output layer node feeding back
The connection weight matrix of point and hidden node, xc(k) and x (k) respectively indicates the output for connecting node layer and hidden node, yc(k) and
Y (k) respectively indicate the connection node layer and export node layer output, α, γ (0≤α < 1,0≤γ < 1) be respectively hidden layer and
The connection feedback oscillator factor certainly of output layer；
2) random drift after denoising is inputted as training sample and improves dynamic recurrent neural network, according to formula (3) and (4) point
Hidden layer section Ji Suan not be calculated separately according to formula (2) and (5) from the connection node layer output for being linked to hidden layer feedback and output layer feedback
Point and output node layer output, according to the error function for substituting into error function formula calculating reality output and desired output；
3) when error is not up to setting accuracy, network is constantly trained using back-propagation algorithm, updates network parameter；When
When error drops to setting accuracy, optimal network model is obtained.
Further, if the network reality output of kth step is yd(k), then the error function of reality output and desired output
Formula are as follows:
Further, the Denoising Algorithm selects small wave converting method.Wavelet analysis (wavelet analysis) or small
Wave conversion (wavelet transform), which refers to use, limit for length or rapid decay, referred to as " morther wavelet " (mother
Wavelet waveform) indicates signal, which is scaled and translates to match the signal of input.The present invention chooses small
Wave conversion method is used for the extraction of random drift.
It can be seen via above technical scheme that compared with prior art, it is dynamic using improving that the present disclosure provides a kind of
The method that state Recursive Networks model the random drift of MEMS gyro, this method introduce Denoising Algorithm and export MEMS gyro
In high frequency white noise and low frequency random drift separated, using improve dynamic recurrent neural network to the random drift after denoising
Shifting is trained, and is established relationship model of the non-stationary wander sequences in the past between current time, is increased output layer node feeding back
To improve network structure, realizes the real-time prediction to Modelling of Random Drift of Gyroscopes variation tendency, effectively improve MEMS inertial navigation system
Precision.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 attached drawing is a kind of MEMS gyro random drift modeling side of application enhancements dynamic recurrent neural network provided by the invention
Method method flow schematic diagram；
Fig. 2 attached drawing is the random drift data statistics schematic diagram of MEMS gyro in the embodiment of the present invention；
Fig. 3 attached drawing is the structure chart that dynamic recurrent neural network is improved in the embodiment of the present invention；
Fig. 4 attached drawing is the training flow diagram that dynamic recurrent neural network is improved in the embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to attached drawing 1, the embodiment of the invention discloses a kind of MEMS gyros of application enhancements dynamic recurrent neural network to float at random
Modeling method is moved, method includes the following steps:
S1: establishing the output error models of MEMS gyro, and it is white to obtain high frequency for the main error source of analyzing influence gyro output
Noise and low frequency random drift the two main error sources；
S2: acquisition MEMS gyro Static output data separate high frequency white noise and low frequency random drift using Denoising Algorithm,
Extract the random drift in gyro output signals；
S3: the random drift extracted being inputted as training data and improves dynamic recurrent neural network, and a large amount of training improve dynamic
State Recursive Networks obtain optimal network model；
S4: inputting optimal network model for untrained test data, obtains the random drift prediction output of MEMS gyro,
Complete the dynamic modeling to MEMS gyro random drift.
Specifically, gyro sensitive angular information, in a static condition, signal input come from rotational-angular velocity of the earth.
In practical applications, due to the influence of the factors such as working principle, manufacturing process and use environment, except additional angular velocity information it
Outside, MEMS gyro output would generally contain error.By taking the measurement axis perpendicular to horizontal plane as an example, the output error mould of MEMS gyro
Type are as follows:
ω (t)=ωiesinL+εd+Dr+W(t) (1)
Wherein, ωieFor earth rotation angular speed；L is local latitude；εdFor constant value zero bias, i.e., when input angular velocity is zero
When gyro export constant, can be compensated by calibration experiment；DrFor drift error, there is randomness, tendency and period
Property；W (t) is the zero-mean white noise with time correlation.
Since earth rate is very small relative to other parameters, so that the MEMS gyro of low precision is difficult to sensitivity and turns to the earth
Speed.Therefore, by analyzing the output error models of MEMS gyro it is found that drift error present in gyro output and white noise are
Promote the research object that MEMS gyro measurement accuracy needs to pay close attention to.
The embodiment of the present invention using the MTi-1 series inertial navigation system of Xsens company as research object, place by static system
In northeast day position, Z axis gyro main shaft is in local ground vertical line direction, and continuous acquisition 1 hour static gyro data repeats 3 groups
Test.Low frequency random drift is separated with high frequency white noise using Denoising Algorithm, is modeled, is had just for random drift
Effect promotes modeling accuracy.The random drift data of MEMS gyro are as shown in Figure 2.
It is to realize the important link of MEMS gyro random drift modeling to the training for improving dynamic recurrent neural network, is chosen at dynamic
Improvement network on the basis of state Recursive Networks is as network model.The structure of dynamic recurrent neural network include input layer, output layer,
Hidden layer and a special articulamentum improve dynamic recurrent neural network in addition to the feedback of hidden node, introduce output node layer
Feedback, with hidden layer output jointly from hidden layer input is linked to, structure chart is as shown in Figure 3.
Specifically, dynamic recurrent neural network is inputted using the random drift extracted as training data, a large amount of training improve dynamic
State Recursive Networks obtain optimal network model, specifically includes the following steps:
S31: using the random drift that extracts as the sample data for improving dynamic recurrent neural network, network inputs are past tense
The MEMS gyro random drift at quarter, setting input layer includes 4 nodes, respectively represents the random drift data at preceding 4 moment；Net
Network output is the random drift of the MEMS gyro at current time, and setting output layer includes 1 node, represents the random of current time
Drift data；Hidden layer output is inputted by the delay and storage feedback of articulamentum to hidden layer, and it is 2 that hidden layer, which is arranged, according to heuristic
Layer, each hidden layer include 10 nodes, improve four layers of neural network that dynamic recurrent neural network is four inputs, singly exports, number
Learn model are as follows:
X (k)=f (wl1xc(k)+wl2u(k-1)+wl4yc(k)) (2)
xc(k)=α xc(k-1)+x(k-1) (3)
yc(k)=γ yc(k-1)+y(k-1) (4)
Y (k)=g (wl3x(k)) (5)
Wherein, wl1For the connection weight matrix for connecting node layer and hidden node, wl2For the company of input unit and Hidden unit
Meet weight matrix, wl3For the connection weight matrix of hidden node and output unit, wl4For the articulamentum section with output layer node feeding back
The connection weight matrix of point and hidden node, xc(k) and x (k) respectively indicates the output for connecting node layer and hidden node, yc(k) and
Y (k) respectively indicate the connection node layer and export node layer output, α, γ (0≤α < 1,0≤γ < 1) be respectively hidden layer and
The connection feedback oscillator factor certainly of output layer；
If the network reality output of kth step is yd(k), then the error function formula of reality output and desired output are as follows:
S32: the random drift after denoising is inputted as training sample and improves dynamic recurrent neural network, according to formula (3) and (4)
It calculates separately from the connection node layer output for being linked to hidden layer feedback and output layer feedback, hidden layer is calculated separately according to formula (2) and (5)
Node and output node layer output, the error function of reality output and desired output is calculated according to formula (6)；
S33: when error is not up to setting accuracy, network is constantly trained using back-propagation algorithm, updates network parameter；
When error drops to setting accuracy, optimal network model is obtained.
The process being specifically trained to improvement dynamic recurrent neural network can be found in attached drawing 4.
Specifically, the Denoising Algorithm in the present embodiment selects small wave converting method.In order to grasp MEMS gyro output signal
Characteristic, analyzed it in terms of time domain and frequency domain two.Correlation analysis is used in time domain, uses power spectrum on frequency domain
Density analysis method.It is found by correlation analysis and power spectral-density analysis, the relevance function of gyro output signals has tight
There is spike in the trailing phenomenon of weight, power spectral density, illustrate in signal comprising white noise and coloured noise.In general, white noise
Sound distribution is wider, concentrates on medium-high frequency part, and coloured noise concentrates on low frequency part, is easy the useful signal phase with low frequency
Obscure, so as to cause the lower signal-to-noise ratio of gyro output signals.It will be in gyro output signals therefore, it is necessary to be introduced into Denoising Algorithm
White noise and random drift separated, to be promoted to the modeling accuracy of random drift.Common signal antinoise method master
There are wavelet transformation, empirical mode decomposition, principal component decomposition and adaptive-filtering etc., comprehensively considers the spy of various Denoising Algorithms
Point, the present invention choose the extraction that small wave converting method is used for random drift.Wavelet analysis (wavelet analysis) or small echo
Transformation (wavelet transform) refer to have limit for length or rapid decay, be known as " morther wavelet " (mother wavelet)
Waveform indicate signal, which is scaled and translates to match the signal of input.The present invention chooses wavelet transformation side
Method is used for the extraction of random drift.
After optimal network model foundation is good, other 2 groups of MEMS gyros static test data are subjected to denoising, will be gone
The prediction of random drift can be obtained as trained optimal network model in test sample input S3 in random drift after making an uproar
The dynamic modeling to MEMS gyro random drift is realized in output, to provide accurate mathematics for the compensation of subsequent random drift
Model.The modeling effect of MEMS gyro random drift is as shown in table 1,
The modeling effect recording table of 1 MEMS gyro random drift of table
| Precision index | Mean value (°/s) | Mean square error (°/s) | Relative error (%) | Operation time (s) |
| 1st group | -0.00331411 | 7.0647e-11 | 0.1794 | 4.927579 |
| 2nd group | -0.00328247 | 3.45964e-11 | 0.1531 | 6.56113 |
As shown in Table 1, modeling method provided by the invention is with arithmetic speed is fast, modeling procedure is simple, modeling accuracy is high
The features such as, a kind of effective method is provided for the modeling of MEMS gyro random drift.
Method provided in an embodiment of the present invention, dynamic recurrent neural network increase interior on the basis of counterpropagation network structure
Portion's feedback element, hidden layer output are inputted by the delay and storage feedback of articulamentum to hidden layer, and storage internal state makes it have
The function of behavioral characteristics is mapped, so that system be made to have the ability for adapting to time-varying characteristics.Meanwhile dynamic recurrent neural network is improved except hidden
Except the feedback of node layer, the feedback of output node layer is introduced, is linked to hidden layer input certainly jointly with hidden layer output.Output feedback
Introducing enhances the ability of network processes multidate information, can more preferably realize the dynamic modeling of nonlinear system, has preferable
Time series forecasting performance.The accurate compensation that can be used for MEMS gyro random error after modeling reaches and promotes MEMS inertial navigation system
The purpose of precision.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For device disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (5)
1. a kind of MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network, which is characterized in that including following
Step:
Establish the output error models of MEMS gyro, the main error source of analyzing influence gyro output, obtain high frequency white noise and
The two main error sources of low frequency random drift；
MEMS gyro Static output data are acquired, high frequency white noise and low frequency random drift are separated using Denoising Algorithm, extract top
Random drift in spiral shell output signal；
The random drift extracted is inputted as training data and improves dynamic recurrent neural network, a large amount of training improve Dynamic Recurrent Neural Networks
Network obtains optimal network model；
Untrained test data is inputted into optimal network model, obtains the random drift prediction output of MEMS gyro, completion pair
The dynamic modeling of MEMS gyro random drift.
2. a kind of MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network according to claim 1,
It is characterized in that, the output error models of the MEMS gyro are as follows:
ω (t)=ωiesinL+εd+Dr+W(t) (1)
Wherein, ωieFor earth rotation angular speed；L is local latitude；εdFor constant value zero bias, i.e., when input angular velocity is zero
Gyro exports constant, can be compensated by calibration experiment；DrFor drift error, there is randomness, tendency and periodicity；W
It (t) is the zero-mean white noise with time correlation.
3. a kind of MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network according to claim 1,
It is characterized in that, inputting dynamic recurrent neural network for the random drift extracted as training data, a large amount of training improve dynamic and pass
Return network, obtains optimal network model, specifically includes the following steps:
1) using the random drift that extracts as the sample data for improving dynamic recurrent neural network, network inputs are last time
MEMS gyro random drift, setting input layer includes 4 nodes, respectively represents the random drift data at preceding 4 moment；Network is defeated
It is out the random drift of the MEMS gyro at current time, setting output layer includes 1 node, represents the random drift at current time
Data；Hidden layer output is inputted by the delay and storage feedback of articulamentum to hidden layer, and it is 2 layers that hidden layer, which is arranged, according to heuristic,
Each hidden layer includes 10 nodes, improves four layers of neural network that dynamic recurrent neural network is four inputs, singly exports, mathematical modulo
Type are as follows:
X (k)=f (wl1xc(k)+wl2u(k-1)+wl4yc(k)) (2)
xc(k)=α xc(k-1)+x(k-1) (3)
yc(k)=γ yc(k-1)+y(k-1) (4)
Y (k)=g (wl3x(k)) (5)
Wherein, wl1For the connection weight matrix for connecting node layer and hidden node, wl2For the connection weight of input unit and Hidden unit
Matrix, wl3For the connection weight matrix of hidden node and output unit, wl4For with output layer node feeding back connection node layer with
The connection weight matrix of hidden node, xc(k) and x (k) respectively indicates the output for connecting node layer and hidden node, yc(k) and y (k)
It respectively indicates the connection node layer and exports the output of node layer, α, γ (0≤α < 1,0≤γ < 1) are respectively hidden layer and output
The connection feedback oscillator factor certainly of layer；
2) random drift after denoising is inputted as training sample and improves dynamic recurrent neural network, counted respectively according to formula (3) and (4)
Calculate from be linked to hidden layer feedback and output layer feedback connection node layer output, according to formula (2) and (5) calculate separately hidden node with
Node layer output is exported, the error function that error function formula calculates reality output and desired output is substituted into；
3) when error is not up to setting accuracy, network is constantly trained using back-propagation algorithm, updates network parameter；Work as error
When dropping to setting accuracy, optimal network model is obtained.
4. a kind of MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network according to claim 3,
It is characterized in that, setting the network reality output of kth step as yd(k), then the error function formula are as follows:
5. a kind of MEMS gyro random drift of application enhancements dynamic recurrent neural network according to claim 1-3 is built
Mould method, which is characterized in that the Denoising Algorithm selects small wave converting method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910300780.4A CN110044350A (en) | 2019-04-15 | 2019-04-15 | The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910300780.4A CN110044350A (en) | 2019-04-15 | 2019-04-15 | The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110044350A true CN110044350A (en) | 2019-07-23 |
Family
ID=67277307
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910300780.4A Pending CN110044350A (en) | 2019-04-15 | 2019-04-15 | The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110044350A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102353375A (en) * | 2011-05-06 | 2012-02-15 | 微迈森惯性技术开发（北京）有限公司 | Dynamic parameter adjustment method of motion attitude data and device thereof |
| CN202177381U (en) * | 2011-05-06 | 2012-03-28 | 微迈森惯性技术开发（北京）有限公司 | Motion gesture data acquisition device and human body motion gesture tracking system |
| CN202472729U (en) * | 2011-12-20 | 2012-10-03 | 西北师范大学 | SIMF Elman network structure |
| CN103593538A (en) * | 2013-11-28 | 2014-02-19 | 东南大学 | Fiber optic gyroscope temperature drift modeling method by optimizing dynamic recurrent neural network through genetic algorithm |
| CN104035037A (en) * | 2014-05-12 | 2014-09-10 | 广东电网公司电力科学研究院 | On-line estimating method for SOH of new energy automobile power battery |
| CN105446821A (en) * | 2015-11-11 | 2016-03-30 | 哈尔滨工程大学 | Improved neural network based fault diagnosis method for intelligent underwater robot propeller |
| CN108168577A (en) * | 2017-12-22 | 2018-06-15 | 清华大学 | MEMS gyro random error compensation method based on BP neural network |
| CN108764475A (en) * | 2018-08-01 | 2018-11-06 | 北斗航天卫星应用科技集团有限公司 | The Gyro Random error compensating method and system of genetic wavelet neural network |
| CN109460621A (en) * | 2018-11-14 | 2019-03-12 | 上海交通大学 | Based on the gas turbine blades fault detection method for improving neural network |
-
2019
- 2019-04-15 CN CN201910300780.4A patent/CN110044350A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102353375A (en) * | 2011-05-06 | 2012-02-15 | 微迈森惯性技术开发（北京）有限公司 | Dynamic parameter adjustment method of motion attitude data and device thereof |
| CN202177381U (en) * | 2011-05-06 | 2012-03-28 | 微迈森惯性技术开发（北京）有限公司 | Motion gesture data acquisition device and human body motion gesture tracking system |
| CN202472729U (en) * | 2011-12-20 | 2012-10-03 | 西北师范大学 | SIMF Elman network structure |
| CN103593538A (en) * | 2013-11-28 | 2014-02-19 | 东南大学 | Fiber optic gyroscope temperature drift modeling method by optimizing dynamic recurrent neural network through genetic algorithm |
| CN104035037A (en) * | 2014-05-12 | 2014-09-10 | 广东电网公司电力科学研究院 | On-line estimating method for SOH of new energy automobile power battery |
| CN105446821A (en) * | 2015-11-11 | 2016-03-30 | 哈尔滨工程大学 | Improved neural network based fault diagnosis method for intelligent underwater robot propeller |
| CN108168577A (en) * | 2017-12-22 | 2018-06-15 | 清华大学 | MEMS gyro random error compensation method based on BP neural network |
| CN108764475A (en) * | 2018-08-01 | 2018-11-06 | 北斗航天卫星应用科技集团有限公司 | The Gyro Random error compensating method and system of genetic wavelet neural network |
| CN109460621A (en) * | 2018-11-14 | 2019-03-12 | 上海交通大学 | Based on the gas turbine blades fault detection method for improving neural network |
Non-Patent Citations (2)
| Title |
|---|
| 汤霞清等: "小波分析和灰色神经网络融合的光纤陀螺误差建模与补偿", 《中国激光》 * |
| 罗姣妍等: "Elman 动态递归神经网络在陀螺仪系统建模中的应用", 《中国惯性技术学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103499345B (en) | A kind of Fiber Optic Gyroscope Temperature Drift compensation method based on wavelet analysis and BP neural network | |
| CN103593538B (en) | Fiber optic gyroscope temperature drift modeling method by optimizing dynamic recurrent neural network through genetic algorithm | |
| CN101105126B (en) | Logging-while-drilling orientation measurement error compensation method based on micro-quartz angular rate sensor | |
| CN102779238B (en) | Brushless DC (Direct Current) motor system identification method on basis of adaptive Kalman filter | |
| CN104992225B (en) | A kind of compensation method of large deformation flexible body dynamic measuring data temperature and humidity | |
| CN103743402B (en) | A kind of underwater intelligent self adaptation Approach of Terrain Matching of topographic information based amount | |
| CN107832575A (en) | Band feedback maneuvering target Asynchronous Track Fusion based on pseudo-measurement | |
| CN103454677B (en) | Based on the earthquake data inversion method that population is combined with linear adder device | |
| CN106483559B (en) | A kind of construction method of subsurface velocity model | |
| CN104101344A (en) | MEMS (micro electro mechanical system) gyroscope random error compensation method based on particle swarm wavelet network | |
| CN106093849A (en) | A kind of Underwater Navigation method based on range finding with neural network algorithm | |
| CN106022471A (en) | Wavelet neural network model ship rolling real-time prediction method based on particle swarm optimization algorithm | |
| CN105353789A (en) | Continuous vibration signal time history replication control method | |
| CN104915534A (en) | Deformation analysis and decision-making method of electric power tower based on sequence learning | |
| CN109460874B (en) | Sense wave height prediction method based on deep learning | |
| CN102819611B (en) | Local community digging method of complicated network | |
| Zhang et al. | An improved UKF algorithm for extracting weak signals based on RBF neural network | |
| Liu et al. | Gravity aided positioning based on real-time ICCP with optimized matching sequence length | |
| CN104778260B (en) | A kind of dynamic radar environmental knowledge storehouse modeling method | |
| CN105698799B (en) | A kind of optimal FIR filter of pretreatment for improving Strapdown Inertial Navigation System attitude accuracy | |
| CN110044350A (en) | The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network | |
| CN107765259A (en) | A kind of transmission line of electricity laser ranging Signal denoising algorithm that threshold value is improved based on Lifting Wavelet | |
| CN115906559B (en) | Geoelectromagnetic adaptive finite element forward modeling method based on mixed grid | |
| CN103925939A (en) | Method for calibration and compensation of full measuring range of photoelectric encoder | |
| CN111008549A (en) | UUV platform DVL signal distortion reconstruction method based on sample entropy and IFOA-GRNN |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190723 |
|
| RJ01 | Rejection of invention patent application after publication |