CN108764475A - The Gyro Random error compensating method and system of genetic wavelet neural network - Google Patents
The Gyro Random error compensating method and system of genetic wavelet neural network Download PDFInfo
- Publication number
- CN108764475A CN108764475A CN201810866207.5A CN201810866207A CN108764475A CN 108764475 A CN108764475 A CN 108764475A CN 201810866207 A CN201810866207 A CN 201810866207A CN 108764475 A CN108764475 A CN 108764475A
- Authority
- CN
- China
- Prior art keywords
- neural network
- wavelet
- wavelet neural
- random error
- gyro random
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C25/00—Manufacturing, calibrating, cleaning, or repairing instruments or devices referred to in the other groups of this subclass
- G01C25/005—Manufacturing, calibrating, cleaning, or repairing instruments or devices referred to in the other groups of this subclass initial alignment, calibration or starting-up of inertial devices
Abstract
The present invention provides the Gyro Random error compensating method and system of a kind of genetic wavelet neural network, this method includes:Determine the input quantity and output quantity of wavelet neural network;The weights and wavelet basis function coefficient of wavelet neural network are trained by gradient modification method;The weights and wavelet basis function coefficient that wavelet neural network is further trained by genetic algorithm, are used for wavelet neural network, establish the Gyro Random error model based on genetic algorithm and wavelet neural network;Gyro Random error is compensated according to the Gyro Random error model based on genetic algorithm and wavelet neural network.The method and system of the present invention, by combining the optimizing feature of overall importance of genetic algorithm that can make up the partial analysis characteristic of wavelet function, the two is subjected to unification, and then it obtains having the characteristics that global optimizing and takes into account local optimal searching, so as to which effectively Gyro Random error is predicted and compensated, to achieve the purpose that eliminate error.
Description
Technical field
The present invention relates to field of inertia technology more particularly to a kind of Gyro Random error compensations of genetic wavelet neural network
Method and system.
Background technology
Micro inertial measurement unit be it is a kind of carrying out integrated inertial measurement combination device to sextuple inertia parameter, use
The design method and processing technology of MEMS sensor (Microelectro Mechanical Systems, MEMS), energy
Enough while multiaxis acceleration value and magnitude of angular velocity are detected, therefore the volume of device and quality are all lighter, in every field
It is widely applied.Navigation system and the validity of aviation attitude system depend primarily on MEMS-Inertial Measurement Unit
The precision of (microelectro mechanical system-inertial measurement unit, MEMS-IMU), so
And due to the limitation of MEMS inertia devices self structure and processing technology, it exports there are the low drawback of signal-to-noise ratio, causes to lead
Precision of navigating is low.The Random Drift Error of MEMS inertia devices mainly by the disturbance torque of randomness generate, generally include angle with
Machine migration, quantizing noise, zero bias unstability equal error item.Therefore, the random error of MEMS inertia devices is effectively mended
It repays and the application range for improving its precision, extension MEMS inertial navigation systems is had a very important significance.
Wavelet neural network (Wavelet Neural Network, WNN) is by wavelet transformation and neural network theory knot
A kind of neural network altogether, thought are that wavelet function is used to be obtained in this way as the excitation function of hidden layer in network
Network has more degree of freedom and more flexible efficient function approximation capability and stronger fault-tolerant ability.Made using wavelet function
For the excitation function of hidden layer in network so that wavelet transformation can not only be described signal local feature but also simultaneous in time-frequency domain
Care for neural network possessed by autonomous learning and it is adaptive the advantages that.Wavelet Analysis Theory is the development of Fourier's analysis method
With continuation, handled by small echo translation and stretching, it can be achieved that carrying out Local Features Analysis to signal simultaneously in time-frequency domain.
Artificial neural network (Artificial Neural Network, ANN) is from mathematics, physical method and the angle of information processing
The simplified model that human nerve's network is abstracted and is established, have MPP ability, distributed storage ability,
The features such as adaptive ability.Neural network has the best approximation advantage to nonlinear problem and is more suitable for compared to conventional method
Optimal Nonlinear Modeling.Therefore, it in pattern-recognition, signal processing, predictive estimation, the fields such as automatically control and have and widely answer
With.Wavelet neural network be one kind based on BP neural network topological structure, using wavelet basis function as hidden layer node
Transmission function, the neural network of error back propagation, topological structure are as shown in Figure 1 while signal propagated forward.Heredity is calculated
Method (Genetic Algorithm, GA) is a kind of adaptive global optimization probability search method, can directly to structure objects into
Row operation, has inherent Implicit Parallelism and better global optimizing ability.Genetic algorithm is a kind of reference living nature natural selection
With the random search algorithm of natural genetic mechanism, it is insoluble complicated and non-linear to be highly suitable for processing conventional search algorithm
Optimization problem.It is a kind of parallel random search optimal method, is not required to the specifying information of understanding problem, can be to specific objective
Automatic Optimal is not limited by search space.
Invention content
In view of the above problems, it is proposed that the present invention overcoming the above problem in order to provide one kind or solves at least partly
The Gyro Random error compensating method and system of the genetic wavelet neural network of problem are stated, solving MEMS gyro random error has
Uncertainty, time variation, and the problem of be difficult to compensate using mathematical models.
According to an aspect of the present invention, a kind of Gyro Random error compensating method of genetic wavelet neural network is provided,
Including:
Determine the input quantity and output quantity of wavelet neural network;
The weights and wavelet basis function coefficient of wavelet neural network are trained by gradient modification method;
The weights and wavelet basis function coefficient that wavelet neural network is further trained by genetic algorithm, are used for wavelet neural
Network establishes the Gyro Random error model based on genetic algorithm and wavelet neural network;
Gyro Random error is mended according to the Gyro Random error model based on genetic algorithm and wavelet neural network
It repays.
Further, wavelet neural network includes three layers, is input layer, hidden layer, output layer respectively, hidden layer and input
It is wavelet basis function to have connection weight, transmission function between layer, output layer respectively, and the wavelet basis function of the wavelet neural network is
Following formula:
Wherein, ψ (j) is j-th of node output valve of hidden layer, ωijFor the connection weight of input layer and hidden layer, k is defeated
Enter node layer number, I is node in hidden layer, bjFor the shift factor of wavelet basis function, ajFor the contraction-expansion factor of wavelet basis function, ψ
(x) Morlet small echos are taken,
The output layer formula of wavelet neural network is as follows:
Wherein, y (k) is k-th of node output valve of output layer, ωjkFor hidden layer to output layer weights, I is hidden layer section
Points, m are output layer number of nodes, and ψ (j) is j-th of node output valve of hidden layer.
Further, the input quantity of wavelet neural network is the prediction time corresponding previous moment of Gyro Random error
Gyro Random error amount;The output quantity of wavelet neural network is the Gyro Random error prediction of the prediction time of Gyro Random error
Value.
Further, the weights and wavelet basis function coefficient of wavelet neural network are trained by gradient modification method, it is specific real
It is now as follows:
The first step calculates the prediction error of wavelet neural network
Wherein,For desired output, y (k) is prediction output valve, and m is the number of nodes of output layer;
Second step updates the weights and wavelet basis function coefficient of wavelet neural network according to prediction error e
Wherein, n is node in hidden layer,WithIt respectively updates preceding and updated wavelet neural network
Weights,WithThe respectively preceding wavelet basis function coefficient with updated wavelet neural network of update,With
With the wavelet basis function coefficient of updated wavelet neural network respectively before update, η is learning rate.
Further, the weights and wavelet basis function coefficient of wavelet neural network are further trained by genetic algorithm, are used
In wavelet neural network, the Gyro Random error model based on genetic algorithm and wavelet neural network is established, is implemented as follows:
The weights and wavelet basis function coefficient of the wavelet neural network after the training of gradient modification method are obtained, and to wavelet neural
The weights and wavelet basis function coefficient of network are pre-processed;
Calculate the fitness value of each Population in Genetic Algorithms individual;
If the fitness value meets end condition, best weight value and the wavelet basis function system of wavelet neural network are obtained
Number;
If not satisfied, then according to the fitness value of each Population in Genetic Algorithms individual, selected, intersected inside population,
Mutation operation, until meeting end condition, to obtain the best weight value and wavelet basis function coefficient of wavelet neural network;
Best weight value and wavelet basis function coefficient to wavelet neural network carry out reverse process, assign Wavelet Neural Network
Network.
Further, it pre-processes as real coding.
Further, it is calculated by the following formula the fitness value of each Population in Genetic Algorithms individual:
Wherein, f is fitness value, and E is target error functional value, and target error function is as follows:
Wherein, E is target error functional value, and p is input chromosome number,For target output value,For reality output
Value.
Further, it is calculated by the following formula the select probability of each individual:
Wherein, PiFor select probability, f is fitness value, and p is input chromosome number.
Further, random error predicted value is subtracted by the actual random error value of gyro to carry out Gyro Random error
Compensation.
According to another aspect of the present invention, a kind of Gyro Random error according to above-mentioned genetic wavelet neural network is provided
The Gyro Random error compensation system that compensation method is realized, including:
Wavelet neural network parameter setting module, input quantity and output quantity for determining wavelet neural network;
Calculating parameter initial training module, weights and wavelet basis for training wavelet neural network by gradient modification method
Function coefficients;
Calculating parameter second training module, for further training the weights of wavelet neural network and small by genetic algorithm
Wave basic function coefficient is used for wavelet neural network, establishes the Gyro Random error mould based on genetic algorithm and wavelet neural network
Type;
Gyro Random error compensation module, for according to the Gyro Random error based on genetic algorithm and wavelet neural network
Model compensates Gyro Random error.
Compared with prior art, the present invention has the following advantages:
1. the Gyro Random error compensating method and system of the genetic wavelet neural network of the present invention, by combining heredity to calculate
The optimizing feature of overall importance of method can make up the partial analysis characteristic of wavelet function, the two be carried out unification, and then had
Global optimizing simultaneously takes into account local optimal searching feature, so as to which effectively Gyro Random error is predicted and compensated, to reach
To the purpose for eliminating error.
2. the Gyro Random error compensating method and system of the genetic wavelet neural network of the present invention not only there is small echo to become
The partial analysis characteristic and neural network autonomous learning, adaptive ability changed, but also take into account genetic algorithm global optimizing,
Adaptability and robustness so that network convergence rate is fast and prediction is more accurate, to achieve the purpose that accurate compensation.
3. the Gyro Random error compensating method and system of the genetic wavelet neural network of the present invention pass through gradient modification method
The weights and wavelet basis function coefficient of training wavelet neural network, and wavelet neural network is further trained by genetic algorithm
Weights and wavelet basis function coefficient so that network convergence rate is fast and prediction is more accurate, to reach the mesh of accurate compensation
's.
Above description is only the general introduction of technical solution of the present invention, in order to better understand the technical means of the present invention,
And can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage can
It is clearer and more comprehensible, below the special specific implementation mode for lifting the present invention.
Description of the drawings
By reading the detailed description of hereafter preferred embodiment, various other advantages and benefit are common for this field
Technical staff will become clear.Attached drawing only for the purpose of illustrating preferred embodiments, and is not considered as to the present invention
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 is the topological structure of wavelet neural network;
Fig. 2 is the Gyro Random error compensating method flow chart of the genetic wavelet neural network of the embodiment of the present invention;
Fig. 3 is the stochastic error modeling flow chart of the MEMS gyroscope of the embodiment of the present invention;
Fig. 4 is weights and the wavelet basis function system that wavelet neural network is trained by gradient modification method of the embodiment of the present invention
Number block diagram;
Fig. 5 is the foundation of the embodiment of the present invention based on the Gyro Random error model of genetic algorithm and wavelet neural network
Block diagram;
Fig. 6 is the Gyro Random error compensation system block diagram of the genetic wavelet neural network of the embodiment of the present invention;
Fig. 7 is the operation principle stream of the Gyro Random error compensation system of the genetic wavelet neural network of the embodiment of the present invention
Cheng Tu.
Specific implementation mode
The exemplary embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
Completely it is communicated to those skilled in the art.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific terminology), there is meaning identical with the general understanding of the those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, it should be understood that have in the context of the prior art
The consistent meaning of meaning, and unless by specific definitions, otherwise will not be explained with the meaning of idealization or too formal.
Fig. 2 is the Gyro Random error compensating method flow chart of the genetic wavelet neural network of the embodiment of the present invention, such as Fig. 2
It is shown, the Gyro Random error compensating method of genetic wavelet neural network provided by the invention, including:S1 determines wavelet neural
The input quantity and output quantity of network;S2 trains the weights and wavelet basis function coefficient of wavelet neural network by gradient modification method;
S3 further trains the weights and wavelet basis function coefficient of wavelet neural network by genetic algorithm, is used for wavelet neural network,
Establish the Gyro Random error model based on genetic algorithm and wavelet neural network;S4, according to based on genetic algorithm and small echo god
Gyro Random error model through network compensates Gyro Random error.
The Gyro Random error compensating method of the genetic wavelet neural network of the present invention, by the overall situation for combining genetic algorithm
Property optimizing feature can make up the partial analysis characteristic of wavelet function, the two is subjected to unification, and then obtain with global optimizing
And local optimal searching feature is taken into account, so as to which effectively Gyro Random error is predicted and compensated, missed to reach to eliminate
The purpose of difference.
The Gyro Random error compensating method of the genetic wavelet neural network of the present invention not only has the part of wavelet transformation
Analytical characteristics and neural network autonomous learning, adaptive ability, but also take into account the global optimizing of genetic algorithm, adaptability and
Robustness so that network convergence rate is fast and prediction is more accurate, to achieve the purpose that accurate compensation.
The Gyro Random error compensating method of the genetic wavelet neural network of the present invention trains small echo by gradient modification method
The weights and wavelet basis function coefficient of neural network, and by genetic algorithm further train wavelet neural network weights and
Wavelet basis function coefficient so that network convergence rate is fast and prediction is more accurate, to achieve the purpose that accurate compensation.
Referring to Fig. 1, wavelet neural network includes three layers, is input layer, hidden layer, output layer respectively, hidden layer and input
It is wavelet basis function to have connection weight, transmission function between layer, output layer respectively, and the wavelet basis function of the wavelet neural network is
Following formula:
Wherein, x1, x2... ..., xkIt is the input parameter of wavelet neural network, y1, y2... ..., ymIt is wavelet neural network
Prediction output, ωijAnd ωjkFor wavelet neural network weights, when list entries is xi(i=1ij, 2 ... ..., kjk) when, ψ (j)
For j-th of node output valve of hidden layer, ωijFor the connection weight of input layer and hidden layer, k is input layer number, and I is implicit
Node layer number, bjFor the shift factor of wavelet basis function, ajFor the contraction-expansion factor of wavelet basis function, ψ (x) can take Morlet small
Wave,
The output layer formula of wavelet neural network is as follows:
Wherein, y (k) is k-th of node output valve of output layer, ωjkFor hidden layer to output layer weights, I is hidden layer section
Points, m are output layer number of nodes, and ψ (j) is j-th of node output valve of hidden layer.Herein, m is more than 1, this is because m is by accidentally
Poor item number determines.
Herein, ψ (x) can also take the wavelet function other than Morlet small echos, if for example, the Fourier of function ψ (x) becomes
It changesMeet the following conditions simultaneously:
ψ (x) is then selected as wavelet basis function, for arbitrary signal f (x) ∈ L2(R), continuous wavelet transform can be with table
It is shown as:
Wherein, a, b are respectively flexible and translation scale factor,For the conjugation of ψ.
The accuracy of MEMS gyro stochastic error modeling is influenced by randomness, and randomness causes more greatly modeling accuracy lower,
Referring to Fig. 3, the random error of MEMS gyroscope is modeled using Wavelet Neural Network Method, the main structure for including network
It builds, three parts of network training and network test.Although the random error of MEMS gyroscope has very big randomness, three layers
Network structure has good fitting effect to nonlinear function, and network structure is simple, so this method selects three layers
Wavelet network predicts the stochastic error modeling of MEMS gyroscope.
The input quantity of wavelet neural network is the Gyro Random of the prediction time corresponding previous moment of Gyro Random error
Error amount;The output quantity of wavelet neural network is the Gyro Random error prediction value of the prediction time of Gyro Random error so that
The Gyro Random error prediction value and the actual measured value of prediction time of prediction time is subtracted each other, and prediction error is obtained, which misses
Difference trains the weights and wavelet basis function coefficient of wavelet neural network by gradient modification method, to adjust wavelet neural network in real time
Weights and wavelet basis function coefficient, it is pre- that the weights and wavelet basis function coefficient of the wavelet neural network after adjustment are directly used in this
The prediction for surveying the subsequent time Gyro Random error prediction value at moment, forms weights and the wavelet basis function system of wavelet neural network
Several real-time adjustment.
Specifically, referring to Fig. 4, the weights of wavelet neural network and wavelet basis function are trained to be by gradient modification method
Number, is implemented as follows:
S21 calculates the prediction error of wavelet neural network
Wherein,For desired output, y (k) is prediction output valve, and m is the number of nodes of output layer;
S22 updates the weights and wavelet basis function coefficient of wavelet neural network according to prediction error e
Wherein, n is node in hidden layer,WithIt respectively updates preceding and updated wavelet neural network
Weights,WithThe respectively preceding wavelet basis function coefficient with updated wavelet neural network of update,With
With the wavelet basis function coefficient of updated wavelet neural network respectively before update, η is learning rate.Using gradient modification method
The weights and wavelet basis function coefficient for correcting wavelet neural network, to make wavelet neural network prediction output constantly approach expectation
Output.
Referring to Fig. 5, the weights and wavelet basis function coefficient of wavelet neural network are further trained by genetic algorithm, are used for
Wavelet neural network is established the Gyro Random error model based on genetic algorithm and wavelet neural network, is implemented as follows:
S31 obtains the weights and wavelet basis function coefficient of the wavelet neural network after the training of gradient modification method, and to wavelet neural network
Weights and wavelet basis function coefficient pre-processed;S32 calculates the fitness value of each Population in Genetic Algorithms individual;S33, if
The fitness value meets end condition, then obtains the best weight value and wavelet basis function coefficient of wavelet neural network;S34, if not
Meet, then according to the fitness value of each Population in Genetic Algorithms individual, is selected, intersected inside population, mutation operation, until
Meet end condition, to obtain the best weight value and wavelet basis function coefficient of wavelet neural network;S35, to wavelet neural network
Best weight value and wavelet basis function coefficient carry out reverse process, assign wavelet neural network.Wherein, it pre-processes and is compiled for real number
Code, reverse process decode for real number.Due to genetic algorithm cannot directly process problem space parameter, it is therefore necessary to pass through coding
The chromosome or individual for requiring the feasible solution of problem to be expressed as hereditary space, numerical value conversion need not be carried out in view of real coding,
Operatings of genetic algorithm can be directly carried out in the phenotype of solution, therefore the present invention uses real coding.Therefore, something lost of the invention
Propagation algorithm will need the individual optimized to encode first, then be grasped according to set fitness function, selection operation, intersection
To make and mutation operation screens individual, the preferable individual of fitness value is retained, and the bad individual of fitness value is rejected,
Optimal solution is selected by the iterative cycles of certain number.
It is calculated by the following formula the fitness value of each Population in Genetic Algorithms individual:
Wherein, f is fitness value, and E is target error functional value, and target error function is as follows:
Wherein, E is target error functional value, and p is input chromosome number,For target output value,For reality output
Value.
The select probability of each individual can be calculated by the following formula:
Wherein, PiFor select probability, f is fitness value, and p is input chromosome number.
Random error predicted value is subtracted by the actual random error value of gyro to compensate Gyro Random error.Specifically
Ground, the input quantity of wavelet neural network are the Gyro Random error of the prediction time corresponding previous moment of Gyro Random error
Value;The output quantity of wavelet neural network is the Gyro Random error prediction value of the prediction time of Gyro Random error, therefore, prediction
The Gyro Random error prediction value and the actual measured value of prediction time at moment are subtracted each other, and prediction error are obtained, when the prediction error
When meeting end condition, the actual measured value of the subsequent time of prediction time subtracts the random error predicted value, as to prediction
The actual measured value of the subsequent time at moment compensates.
Fig. 6 is the Gyro Random error compensation system block diagram of the genetic wavelet neural network of the embodiment of the present invention, such as Fig. 6 institutes
Show, the Gyro Random provided by the invention realized according to the Gyro Random error compensating method of above-mentioned genetic wavelet neural network is missed
Poor compensation system, including:Wavelet neural network parameter setting module, the input quantity for determining wavelet neural network and output
Amount;Calculating parameter initial training module, weights and wavelet basis function for training wavelet neural network by gradient modification method
Coefficient;Calculating parameter second training module, weights and small echo for further training wavelet neural network by genetic algorithm
Basic function coefficient is used for wavelet neural network, establishes the Gyro Random error model based on genetic algorithm and wavelet neural network;
Gyro Random error compensation module, for Gyro Random error model of the basis based on genetic algorithm and wavelet neural network to top
Spiral shell random error compensates.
For system embodiments, since it is basically similar to the method embodiment, so fairly simple, the correlation of description
Place illustrates referring to the part of embodiment of the method.
Fig. 7 is the operation principle stream of the Gyro Random error compensation system of the genetic wavelet neural network of the embodiment of the present invention
Cheng Tu, as shown in fig. 7, first, determining wavelet neural network structure, the i.e. input quantity, output quantity of wavelet neural network, transmission letter
Connection weight initial value, hidden layer between number, hidden layer and input layer and the connection weight initial value between output layer transmit
Wavelet basis function coefficient in function etc.;It is calculated according to gradient modification method, i.e., to predict error as object function, to learn speed
Rate is step-length, calculates the best initial weights for making prediction minimize the error and wavelet basis function coefficient;Gradient is repaiied using genetic algorithm
It executes best initial weights and wavelet basis function coefficient after training to be encoded, and calculates best initial weights and wavelet basis letter in each population
The fitness value of number system number, selection are intersected and are made a variation to the population more than the corresponding population of fitness value of predetermined value, and
Fitness value is calculated to the population after cross and variation, if the fitness value meets the first end condition, after obtaining second training
Best initial weights and wavelet basis function coefficient, and corresponding Gyro Random is calculated according to the best initial weights and wavelet basis function coefficient
It predicts error, and updates weights and wavelet basis function coefficient, it is defeated if Gyro Random prediction error meets the second end condition
Go out newer weights and wavelet basis function coefficient, selection, intersection and the variation of next step is used for, if being unsatisfactory for the first termination item
Part, then reselect population and the population is intersected and made a variation etc., if being unsatisfactory for the second end condition, recalculate
Corresponding Gyro Random prediction error of other populations etc..
The Gyro Random error compensation system of the genetic wavelet neural network of the present invention, by the overall situation for combining genetic algorithm
Property optimizing feature can make up the partial analysis characteristic of wavelet function, the two is subjected to unification, and then obtain with global optimizing
And local optimal searching feature is taken into account, so as to which effectively Gyro Random error is predicted and compensated, missed to reach to eliminate
The purpose of difference.
The Gyro Random error compensation system of the genetic wavelet neural network of the present invention not only has the part of wavelet transformation
Analytical characteristics and neural network autonomous learning, adaptive ability, but also take into account the global optimizing of genetic algorithm, adaptability and
Robustness so that network convergence rate is fast and prediction is more accurate, to achieve the purpose that accurate compensation.
The Gyro Random error compensation system of the genetic wavelet neural network of the present invention trains small echo by gradient modification method
The weights and wavelet basis function coefficient of neural network, and by genetic algorithm further train wavelet neural network weights and
Wavelet basis function coefficient so that network convergence rate is fast and prediction is more accurate, to achieve the purpose that accurate compensation.
In addition, the embodiment of the present invention additionally provides a kind of computer readable storage medium, it is stored thereon with computer program,
The program realizes the function of the Gyro Random error compensating method of genetic wavelet neural network as described above when being executed by processor.
In the present embodiment, the integrated module/unit of the Gyro Random error compensation system of the genetic wavelet neural network
If being realized in the form of SFU software functional unit and when sold or used as an independent product, a computer can be stored in
In read/write memory medium.Based on this understanding, the present invention realizes all or part of flow in above-described embodiment method,
Relevant hardware can be instructed to complete by computer program, it is computer-readable that the computer program can be stored in one
In storage medium, the computer program is when being executed by processor, it can be achieved that the step of above-mentioned each embodiment of the method.Wherein,
The computer program includes computer program code, and the computer program code can be source code form, object identification code
Form, executable file or certain intermediate forms etc..The computer-readable medium may include:The computer can be carried
Any entity or device of program code, USB flash disk, mobile hard disk, magnetic disc, CD, computer storage, read-only are deposited at recording medium
Reservoir (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier wave letter
Number, telecommunication signal and software distribution medium etc..It should be noted that the content that the computer-readable medium includes can root
Increase and decrease appropriate is carried out according to legislation in jurisdiction and the requirement of patent practice, such as in certain jurisdictions, according to vertical
Method and patent practice, computer-readable medium do not include electric carrier signal and telecommunication signal.
It will be appreciated by those of skill in the art that although some embodiments in this include included in other embodiments
Certain features rather than other feature, but the combination of the feature of different embodiments means to be within the scope of the present invention simultaneously
And form different embodiments.For example, in the following claims, the one of arbitrary of embodiment claimed all may be used
It is used in a manner of in any combination.
Finally it should be noted that:The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
Present invention has been described in detail with reference to the aforementioned embodiments, it will be understood by those of ordinary skill in the art that:It still may be used
With technical scheme described in the above embodiments is modified or equivalent replacement of some of the technical features;
And these modifications or replacements, various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of Gyro Random error compensating method of genetic wavelet neural network, which is characterized in that including:
Determine the input quantity and output quantity of wavelet neural network;
The weights and wavelet basis function coefficient of wavelet neural network are trained by gradient modification method;
The weights and wavelet basis function coefficient that wavelet neural network is further trained by genetic algorithm, are used for Wavelet Neural Network
Network establishes the Gyro Random error model based on genetic algorithm and wavelet neural network;
Gyro Random error is compensated according to the Gyro Random error model based on genetic algorithm and wavelet neural network.
2. the Gyro Random error compensating method of genetic wavelet neural network according to claim 1, which is characterized in that small
Wave neural network includes three layers, is input layer, hidden layer, output layer respectively, has respectively between hidden layer and input layer, output layer
Connection weight, transmission function are wavelet basis function, and the wavelet basis function of the wavelet neural network is following formula:
Wherein, ψ (j) is j-th of node output valve of hidden layer, ωijFor the connection weight of input layer and hidden layer, k is input layer
Number of nodes, I are node in hidden layer, bjFor the shift factor of wavelet basis function, ajFor the contraction-expansion factor of wavelet basis function, ψ (x)
Morlet small echos are taken,
The output layer formula of wavelet neural network is as follows:
Wherein, y (k) is k-th of node output valve of output layer, ωjkFor hidden layer to output layer weights, I is node in hidden layer,
M is output layer number of nodes, and ψ (j) is j-th of node output valve of hidden layer.
3. the Gyro Random error compensating method of genetic wavelet neural network according to claim 2, which is characterized in that small
The input quantity of wave neural network is the Gyro Random error amount of the prediction time corresponding previous moment of Gyro Random error;Small echo
The output quantity of neural network is the Gyro Random error prediction value of the prediction time of Gyro Random error.
4. the Gyro Random error compensating method of genetic wavelet neural network according to claim 3, which is characterized in that logical
The weights and wavelet basis function coefficient for crossing gradient modification method training wavelet neural network, are implemented as follows:
The first step calculates the prediction error of wavelet neural network
Wherein,For desired output, y (k) is prediction output valve, and m is the number of nodes of output layer;
Second step updates the weights and wavelet basis function coefficient of wavelet neural network according to prediction error e
Wherein, n is node in hidden layer,WithThe respectively preceding power with updated wavelet neural network of update
Value,WithThe respectively preceding wavelet basis function coefficient with updated wavelet neural network of update,WithPoint
With the wavelet basis function coefficient of updated wavelet neural network before Wei not updating, η is learning rate.
5. the Gyro Random error compensating method of genetic wavelet neural network according to claim 4, which is characterized in that logical
Weights and wavelet basis function coefficient that genetic algorithm further trains wavelet neural network are crossed, wavelet neural network is used for, establishes
Gyro Random error model based on genetic algorithm and wavelet neural network, is implemented as follows:
The weights and wavelet basis function coefficient of the wavelet neural network after the training of gradient modification method are obtained, and to wavelet neural network
Weights and wavelet basis function coefficient pre-processed;
Calculate the fitness value of each Population in Genetic Algorithms individual;
If the fitness value meets end condition, the best weight value and wavelet basis function coefficient of wavelet neural network are obtained;
If not satisfied, then according to the fitness value of each Population in Genetic Algorithms individual, is selected, is intersected, made a variation inside population
Operation, until meeting end condition, to obtain the best weight value and wavelet basis function coefficient of wavelet neural network;
Best weight value and wavelet basis function coefficient to wavelet neural network carry out reverse process, assign wavelet neural network.
6. the Gyro Random error compensating method of genetic wavelet neural network according to claim 5, which is characterized in that pre-
Processing is real coding.
7. the Gyro Random error compensating method of genetic wavelet neural network according to claim 6, which is characterized in that logical
Cross the fitness value that following formula calculates each Population in Genetic Algorithms individual:
Wherein, f is fitness value, and E is target error functional value, and target error function is as follows:
Wherein, E is target error functional value, and p is input chromosome number,For target output value,For real output value.
8. the Gyro Random error compensating method of genetic wavelet neural network according to claim 7, which is characterized in that logical
Cross the select probability that following formula calculates each individual:
Wherein, PiFor select probability, f is fitness value, and p is input chromosome number.
9. the Gyro Random error compensating method of genetic wavelet neural network according to claim 3, which is characterized in that logical
It crosses the actual random error value of gyro and subtracts random error predicted value and Gyro Random error is compensated.
10. a kind of gyro of Gyro Random error compensating method that realizing genetic wavelet neural network described in claim 1 with
Machine error compensation system, which is characterized in that including:
Wavelet neural network parameter setting module, input quantity and output quantity for determining wavelet neural network;
Calculating parameter initial training module, weights and wavelet basis function for training wavelet neural network by gradient modification method
Coefficient;
Calculating parameter second training module, weights and wavelet basis for further training wavelet neural network by genetic algorithm
Function coefficients are used for wavelet neural network, establish the Gyro Random error model based on genetic algorithm and wavelet neural network;
Gyro Random error compensation module, for according to the Gyro Random error model based on genetic algorithm and wavelet neural network
Gyro Random error is compensated.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810866207.5A CN108764475A (en) | 2018-08-01 | 2018-08-01 | The Gyro Random error compensating method and system of genetic wavelet neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810866207.5A CN108764475A (en) | 2018-08-01 | 2018-08-01 | The Gyro Random error compensating method and system of genetic wavelet neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108764475A true CN108764475A (en) | 2018-11-06 |
Family
ID=63968461
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810866207.5A Pending CN108764475A (en) | 2018-08-01 | 2018-08-01 | The Gyro Random error compensating method and system of genetic wavelet neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108764475A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110007235A (en) * | 2019-03-24 | 2019-07-12 | 天津大学青岛海洋技术研究院 | A kind of accumulator of electric car SOC on-line prediction method |
CN110044350A (en) * | 2019-04-15 | 2019-07-23 | 北京航空航天大学 | The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network |
CN110107231A (en) * | 2019-06-06 | 2019-08-09 | 吉林大学 | A kind of automatic cat road of adaptive neural network and its control method |
CN110440827A (en) * | 2019-08-01 | 2019-11-12 | 北京神导科讯科技发展有限公司 | A kind of scaling method of parameter error, device and storage medium |
CN112284366A (en) * | 2020-10-26 | 2021-01-29 | 中北大学 | Method for correcting course angle error of polarized light compass based on TG-LSTM neural network |
CN114347018A (en) * | 2021-12-20 | 2022-04-15 | 上海大学 | Mechanical arm disturbance compensation method based on wavelet neural network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103512569A (en) * | 2013-09-29 | 2014-01-15 | 北京理工大学 | Discrete wavelet multiscale analysis based random error compensation method for MEMS (Micro Electro Mechanical system) gyroscope |
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 |
CN104101344A (en) * | 2014-07-11 | 2014-10-15 | 哈尔滨工程大学 | MEMS (micro electro mechanical system) gyroscope random error compensation method based on particle swarm wavelet network |
CN108168577A (en) * | 2017-12-22 | 2018-06-15 | 清华大学 | MEMS gyro random error compensation method based on BP neural network |
-
2018
- 2018-08-01 CN CN201810866207.5A patent/CN108764475A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103512569A (en) * | 2013-09-29 | 2014-01-15 | 北京理工大学 | Discrete wavelet multiscale analysis based random error compensation method for MEMS (Micro Electro Mechanical system) gyroscope |
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 |
CN104101344A (en) * | 2014-07-11 | 2014-10-15 | 哈尔滨工程大学 | MEMS (micro electro mechanical system) gyroscope random error compensation method based on particle swarm wavelet network |
CN108168577A (en) * | 2017-12-22 | 2018-06-15 | 清华大学 | MEMS gyro random error compensation method based on BP neural network |
Non-Patent Citations (4)
Title |
---|
专祥涛: "《最优化方法基础》", 30 April 2018, 武汉大学出版社 * |
卢海曦 等: "基于遗传小波神经网络的MEMS陀螺误差建模", 《中国惯性技术学报》 * |
孙伟 等: "基于GA-WNN神经网络的MEMS陀螺随机误差补偿", 《第九届中国卫星导航学术年会论文集——S02 导航与位置服务》 * |
牛春峰 等: "基于小波神经网络的MEMS陀螺输出预测方法", 《海军工程大学学报》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110007235A (en) * | 2019-03-24 | 2019-07-12 | 天津大学青岛海洋技术研究院 | A kind of accumulator of electric car SOC on-line prediction method |
CN110044350A (en) * | 2019-04-15 | 2019-07-23 | 北京航空航天大学 | The MEMS gyro random drift modeling method of application enhancements dynamic recurrent neural network |
CN110107231A (en) * | 2019-06-06 | 2019-08-09 | 吉林大学 | A kind of automatic cat road of adaptive neural network and its control method |
CN110107231B (en) * | 2019-06-06 | 2024-03-29 | 吉林大学 | Self-adaptive neural network automatic catwalk and control method thereof |
CN110440827A (en) * | 2019-08-01 | 2019-11-12 | 北京神导科讯科技发展有限公司 | A kind of scaling method of parameter error, device and storage medium |
CN110440827B (en) * | 2019-08-01 | 2022-05-24 | 北京神导科讯科技发展有限公司 | Parameter error calibration method and device and storage medium |
CN112284366A (en) * | 2020-10-26 | 2021-01-29 | 中北大学 | Method for correcting course angle error of polarized light compass based on TG-LSTM neural network |
CN112284366B (en) * | 2020-10-26 | 2022-04-12 | 中北大学 | Method for correcting course angle error of polarized light compass based on TG-LSTM neural network |
CN114347018A (en) * | 2021-12-20 | 2022-04-15 | 上海大学 | Mechanical arm disturbance compensation method based on wavelet neural network |
CN114347018B (en) * | 2021-12-20 | 2024-04-16 | 上海大学 | Mechanical arm disturbance compensation method based on wavelet neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108764475A (en) | The Gyro Random error compensating method and system of genetic wavelet neural network | |
CN103593538B (en) | Fiber optic gyroscope temperature drift modeling method by optimizing dynamic recurrent neural network through genetic algorithm | |
CN108805287A (en) | The Gyro Random error compensating method and system of genetic wavelet neural network | |
Hajian et al. | Artificial neural networks | |
CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
US11080586B2 (en) | Neural network reinforcement learning | |
US20180046915A1 (en) | Compression of deep neural networks with proper use of mask | |
US11182676B2 (en) | Cooperative neural network deep reinforcement learning with partial input assistance | |
KR102596158B1 (en) | Reinforcement learning through dual actor critical algorithm | |
US20130151460A1 (en) | Particle Methods for Nonlinear Control | |
Sushchenko et al. | Processing of redundant information in airborne electronic systems by means of neural networks | |
CN109978283A (en) | Photovoltaic power generation power prediction method based on branch evolution neural network | |
CN106529185B (en) | A kind of combination forecasting method and system of ancient building displacement | |
CN114199248B (en) | AUV co-location method for optimizing ANFIS based on mixed element heuristic algorithm | |
CN108168577A (en) | MEMS gyro random error compensation method based on BP neural network | |
CN109165737B (en) | Porosity prediction method based on conditional random field and BP neural network | |
CN115935834A (en) | History fitting method based on deep autoregressive network and continuous learning strategy | |
CN108460462A (en) | A kind of Interval neural networks learning method based on interval parameter optimization | |
Stopar et al. | GPS-derived geoid using artificial neural network and least squares collocation | |
CN112388628A (en) | Apparatus and method for training a gaussian process regression model | |
CN114154401A (en) | Soil erosion modulus calculation method and system based on machine learning and observation data | |
KR102120150B1 (en) | Learning method and learning device for variational interference using neural network and test method and test device for variational interference using the same | |
WO2020257263A1 (en) | Systems and methods for solving geosteering inverse problems in downhole environments using a deep neural network | |
Li | Comparing the Kalman filter with a Monte Carlo-based artificial neural network in the INS/GPS vector gravimetric system | |
CN114996947A (en) | Three-dimensional oil reservoir numerical simulation method |
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: 20181106 |
|
RJ01 | Rejection of invention patent application after publication |