CN109034202A - A kind of avionics system mode identification method of deepness belief network - Google Patents
A kind of avionics system mode identification method of deepness belief network Download PDFInfo
- Publication number
- CN109034202A CN109034202A CN201810693984.4A CN201810693984A CN109034202A CN 109034202 A CN109034202 A CN 109034202A CN 201810693984 A CN201810693984 A CN 201810693984A CN 109034202 A CN109034202 A CN 109034202A
- Authority
- CN
- China
- Prior art keywords
- model
- dbnc
- avionics system
- training
- artificial fish
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of avionics system mode identification method based on artificial fish school optimization method deepness belief network, and steps are as follows: the building of one, DBNC model, and setting model parameter initializes;Two, typical fault mode is chosen, the historical failure data for being used for DBNC model training is obtained from avionics system;Three, pre-training is carried out to DBNC model using to sdpecific dispersion method;The bias of the connection weight of each node and each node layer between layers is initialized;Four, the reversed fine tuning of DBNC model is trained;Five, to DBNC model training outcome evaluation;Six, pattern-recognition is carried out to avionics system operation data using DBNC model;By above step, the error of DBNC model is constantly reduced, and is goed deep into the number of plies, and accuracy obviously rises, and shows that the training process of the model is rationally effective;The model is used for the pattern-recognition of avionics system, has higher accuracy rate of diagnosis, is very suitable for the valid model as avionics system pattern-recognition.
Description
Technical field:
The present invention provides a kind of avionics system mode identification method of deepness belief network, it relates to a kind of based on artificial
The avionics system mode identification method of fish school optimization method deepness belief network, belongs to the technical field of pattern-recognition.
Background technique:
Avionics system is the core processing system of modern aircraft, mainly serve for ensuring aircraft takeoff, navigation and
The safe navigation of the mission phases such as land.On the one hand, with the high integrity of avionics system, the hardware capability in system is by software
Substitution realizes that the performance of aircraft is continuously improved;But on the other hand, due to integrated avionics resource shared resources,
The features such as task activity crosslinking, the comprehensive complexity of system, so that the small fault in system is likely to cause catastrophic loss.Boat
Electric system has received widespread attention as China's universal avionics system, which has the characteristics that Highgrade integration, and
And intercouple, be cross-linked with each other between modules, while performance improves but also daily maintenance and pattern-recognition more
It is complicated.In order to improve the pattern-recognition rate and system reliability of avionics system, typical avionics system pattern-recognition side is studied
Method is necessary.
Deepness belief network (Deep Belief Network, DBN) is a generative probabilistic model, passes through its mind of training
Through the weight between member, we can allow entire neural network to generate training data according to maximum probability.The constituent element of DBN is
Limited Boltzmann machine (Restricted Boltzmann Machines, RBM), the process of training DBN be layer by layer into
It is capable, in each layer, infer hidden layer with data vector, then this hidden layer as next layer of data vector.softmax
Function has very extensive application in machine learning, it is a classification function, main function be by the element in array by
Classify according to probability size, it is assumed that we have an array V, a shared n element, ViIndicate i-th of element in V, then
The Softmax value of this element is exactlyAvionics system has many data volumes, in order to more effectively to these data
Processing, we combine DBN and softmax function, building deepness belief network classification (Deep Belief Network
Classifier, DBNC) model, does data processing to avionics system with DBNC model.With traditional method to avionics system
Carrying out data processing will appear that learning efficiency is low, convergence rate is slow, easily falls into local minimum state, is difficult to converge to optimal weight
The problems such as solution, can obtain higher identification accuracy with DBNC model with faster rate, so as to improve the knowledge of avionics system mode
Not existing deficiency.
With the continuous development of deep learning theory, application field has been expanded in the pattern-recognition of complication system.But
Be the existing pattern recognition model based on deep learning be generally adopted by traditional backpropagation (BackPropagation,
BP) method carries out reversed tuning, although BP method achieves many achievements in the evolutionary process of deep learning model,
That there are learning efficiencies is low for BP method, convergence rate is slow, easily falls into local minimum state, is difficult to converge to optimal weight solution etc. and ask
Topic.Therefore the method for being badly in need of a suitable reversed tuning of deep learning model optimizes it.
Summary of the invention:
For the problem that avionics system pattern-recognition precision is inadequate, the invention proposes one kind to be based on artificial fish school optimization side
The mode identification method of the avionics system of method (Artificial Fish Swarm Algorithm, AFSA) deepness belief network,
Purpose is to construct DBNC model by DBN and softmax classification function, by DBNC model, according to data input and output, judgement
The state of avionics system, providing one kind with this, more preferably method to carry out pattern-recognition to avionics system.
The invention proposes a kind of avionics system mode identification methods of deepness belief network, i.e., a kind of to be based on artificial fish-swarm
The avionics system mode identification method of optimization method deepness belief network, the specific steps of which are as follows:
Step 1, the building of DBNC model, setting model parameter initialize;
Step 2 chooses typical fault mode, and the historical failure for being used for DBNC model training is obtained from avionics system
Data;
Step 3 carries out DBNC model using to sdpecific dispersion (Contrastive Divergence, CD) method
Pre-training;The bias of the connection weight of each node and each node layer between layers is initialized;
Step 4 trains the reversed fine tuning of DBNC model, i.e., by artificial fish school optimization method in DBNC model
Connection weight and node bias situation optimize;
Step 5, to DBNC model training outcome evaluation, according to the study situation of error judgment model, if error reaches
It is required that step 6 is then carried out, and if not up to accuracy requirement, resume at step two;
Step 6 carries out pattern-recognition to avionics system operation data using trained DBNC model;
By above step, the error of DBNC model is constantly reduced, and is goed deep into the number of plies, on accuracy is obvious
It rises, shows that the training process of the model is rationally effective;The model is used for the pattern-recognition of avionics system, has higher examine
Disconnected accuracy rate, is very suitable for the valid model as avionics system pattern-recognition.
Wherein, in " building of DBNC model " described in step 1, specific practice is as follows: the bottom of building DBNC model
Layer is made of several layers RBM, and top layer is the output layer of Softmax classification function composition;Training data is by several layers RBM training
Later by classification layer output category result.
Wherein, in the specific practice of " setting model parameter initializes " described in step 1 are as follows: setting hidden layer number,
Input layer and output layer neuron number of nodes, the neuron node number of hidden layer, the connection weight W between each layer, each neuron
Node is bias initial value, initial learning rate ρ size and study momentum m.
Wherein, " choosing typical fault mode " described in step 2, specific practice is as follows: therefrom choosing ten kinds
Typical fault mode is respectively: 74 failure of GDC, GDC74 configuration module failure, GDC74 route or power failure, GDC
74 dynamic and static pressure interfaces and pipeline blockage, atmospheric air temperature probe failure, atmospheric air temperature probe route and joint failure, GPS signal event
Barrier, GRS77 failure, GRS77 configuration module failure, GMU44 failure.
Wherein, " historical failure that operation is used for DBNC model training is obtained from avionics system described in step 2
The specific practice of data " has: collecting the data of avionics system in the process of running, these data reflect the operation of avionics system
Time, air pressure, ground level, cabin temperature, delivery temperature, communication/navigation frequency, amount of fuel, oil temperature, aircraft flight
Roll angle, pitch angle and course angle etc..
Wherein, described in above-mentioned steps three " using to sdpecific dispersion method (ContrastiveDivergence,
CD method) to DBNC model carry out pre-training " specific practice are as follows:
The initial state value ν of the first step, initialization network parameter θ and visual layer unit0=x0, setting RBM maximum is trained to change
Generation number is 400;
Second step calculates all hiding layer unitsFrom condition distribution p
(h0j|ν0) in extract h0j∈{0,1};
In formula:For activation primitive;h0jFor hidden layer neuron vector;WijTo connect visual layers
The weight matrix of neuron and hidden layer neuron;bjFor hidden layer bias vector;ν is visual layers neuron vector;
Third step calculates all visible elementsFrom p (ν1i|h0) in take out
Take ν1i∈{0,1};
In formula: wherein ajFor visual layers bias vector;
4th step calculates all hiding layer units
In formula: h0jFor hidden layer neuron vector;WijFor the weight square for connecting visual layers neuron and hidden layer neuron
Battle array;bjFor hidden layer bias vector;ν is visual layers neuron vector;
5th step, by following various update parameters:
a←a+p(ν0-ν1)
b←b+p(P(h0=1 | ν0)-p(h1=1 | ν1));
In formula: a is visual layers bias vector;B is hidden layer bias vector;W is connection visual layers neuron and hidden layer
The weight matrix of neuron;H is hidden layer neuron vector;ν is visual layers neuron vector;
Step 6: repeat second step to the 5th step, and it is sufficiently small until reaching maximum number of iterations or reconstructed error, terminate this
The training of layer RBM.
Wherein, described in step 4 " by artificial fish school optimization method to the connection weight and section in DBNC model
Point bias conditions optimize ", specific practice is as follows:
The first step loads training sample, initializes artificial fish-swarm and its parameter: population invariable number N, sensing range
(Visual), moving step length (Step), crowding, maximum times of looking for food, variable number, maximum number of iterations (try number)
Deng;
Second step generates the artificial fish-swarm of set scale at random in Visual, calculates the current food of each Artificial Fish
Object concentration is maximized into bulletin board, the weight and bias of the Artificial Fish individual of food concentration, that is, every;
The food concentration recorded in itself food concentration and billboard is compared by third step, if itself food concentration
It is then the Artificial Fish state by the information update in billboard greater than in bulletin board;Conversely, then continue select food concentration compared with
Big Artificial Fish knocked into the back, is bunched and foraging behavior;
4th step is first carried out one and judges sentence, and whether the number of iterations which is used to judgment method has reached the upper limit,
Or whether resulting optimal solution meets required precision;If then entering the 5th step, third step is otherwise returned;
5th step will the weight in Artificial Fish individual information existing in bulletin board and partially after this method end of run
It sets output to come out, which is the best initial weights and bias of neuron node.
Wherein, the specific practice of " to the DBNC model training outcome evaluation " described in step 5 are as follows: first with defeated
Value rebuilds the input of training out, is then compared this result with actual input value, obtains accuracy, finally relatively more accurate
Relationship degree and the threshold value set in advance between obtains assessment result.
Wherein, " mode is carried out to avionics system operation data using trained DBNC model described in step 6
The specific practice of identification " is: the fault data that avionics system is collected into is obtained as network inputs data by DBNC model
Corresponding output data after analysis to output data obtains the state of avionics system, to carry out mode to avionics system
Identification.
A kind of avionics system mode identification method of deepness belief network of the present invention, i.e., it is a kind of to be based on artificial fish school optimization side
The avionics system mode identification method of method deepness belief network, has the advantage that
1, deepness belief network can obtain higher identification accuracy, Neng Gougai compared to traditional method with faster rate
Kind conventional method learning efficiency is low, convergence rate is slow, easily falls into local minimum state, is difficult to converge to asking for optimal weight solution
Topic.
2, artificial fish-swarm method early period has stronger search and convergence capabilities, can substitute traditional BP method and come to depth
It spends belief network and carries out tuning.Connection weight and each neuron section of this method to each layer in the DBNC model in learning process
The biasing of point optimizes, and can effectively improve the accuracy rate of avionics system pattern-recognition, shortens Diagnostic Time, for improving me
Economic Growth of Civil Aviation Transportation reliability plays an important role, and can effectively improve Mission Success rate and Combat readiness, before wide application
Scape and value.
Detailed description of the invention:
Fig. 1 is the method for the invention flow chart.
Fig. 2 is to sdpecific dispersion method flow diagram.
Fig. 3 is artificial fish-swarm method flow diagram.
Fig. 4 is that avionics system different faults mode is compared based on the diagnosis situation of both of which identification model.
Serial number, symbol, code name are described as follows in figure:
AFSA: one kind being based on artificial fish school optimization method (Artificial Fish Swarm Algorithm, AFSA)
DBN: deepness belief network (Deep Belief Network, DBN)
DBNC model: deepness belief network classification (Deep Belief Network Classifier, DBNC) model
RBM: limited Boltzmann machine (Restricted Boltzmann Machines, RBM)
CD: to sdpecific dispersion (Contrastive Divergence, CD)
BP: backpropagation (Back Propagation, BP)
Visual: sensing range
Step: moving step length
Try number: maximum number of iterations
Logistic function:
Specific embodiment:
The invention proposes a kind of avionics system mode identification methods of deepness belief network, i.e., a kind of to be based on artificial fish-swarm
The avionics system mode identification method of optimization algorithm deepness belief network with reference to the accompanying drawing specifically describes invention:
See Fig. 1, a kind of avionics system pattern-recognition side based on artificial fish school optimization method deepness belief network of the present invention
The specific implementation step of method, this method is as follows:
Step 1, the building of DBNC model, setting model parameter initialize.
The building bottom RBM number of plies is N first0The pattern-recognition mould based on artificial fish-swarm method deepness belief network of layer
Type.The input layer and output layer neuron number of nodes of the model are set to m and n, the neuron node number point of hidden layer
It Wei not x1, x2, x3And x4, training set number is set as X, and test set number is set as C, and every layer of the number of iterations is set as D times, each layer
Between connection weight W be initialized as the random number of Gaussian distributed, top layer is the output of Softmax classification function composition
Layer initializes each neuron node bias, initial learning rate ρ and study momentum m.It is inputted when pattern recognition model training
Data are the 2-D data of X*R, export the one-dimensional data for X*L;The input data selected when model measurement for C*R two-dimemsional number
According to exporting the one-dimensional data for C*L.
Step 2 chooses typical fault mode, and the historical failure for being used for DBNC model training is obtained from avionics system
Data.
Ten kinds of typical fault modes are therefrom chosen, are respectively: 74 failure of GDC, GDC74 configuration module failure, GDC 74
Route or power failure, 74 dynamic and static pressure interface of GDC and pipeline blockage, atmospheric air temperature probe failure, atmospheric air temperature probe route
With joint failure, GPS signal failure, GRS77 failure, GRS77 configuration module failure, GMU44 failure.Collect these fault modes
Relevant information, form data set, the parameter of each data has R in the data set, shows respectively avionics system
State in the process of running, including runing time, air pressure, ground level, cabin temperature, delivery temperature, communication/navigation frequency
Rate, amount of fuel, oil temperature, the roll angle of aircraft flight, pitch angle and course angle etc..
Step 3 carries out DBNC model using to sdpecific dispersion (Contrastive Divergence, CD) method
Pre-training.The bias of the connection weight of each node and each node layer between layers is initialized.
The pre-training process of deepness belief network actually carries out the process of network parameter initialization, sees Fig. 2, used
Specific step is as follows for CD method:
The initial state value ν of the first step, initialization network parameter θ and visual layer unit0=x0, setting RBM maximum is trained to change
Generation number.
Second step calculates all hiding layer unitsFrom condition distribution p
(h0j|ν0) in extract h0j∈{0,1};
In formula:For activation primitive;h0jFor hidden layer neuron vector;WijTo connect visual layers
The weight matrix of neuron and hidden layer neuron;bjFor hidden layer bias vector;ν is visual layers neuron vector.
Third step calculates all visible elementsFrom p (ν1i|h0) in take out
Take ν1i∈{0,1};
In formula: wherein ajFor visual layers bias vector.
4th step calculates all hiding layer units
In formula: h0jFor hidden layer neuron vector;WijFor the weight square for connecting visual layers neuron and hidden layer neuron
Battle array;bjFor hidden layer bias vector;ν is visual layers neuron vector.
5th step, by following various update parameters:
a←a+p(ν0-ν1)
b←b+p(P(h0=1 | ν0)-p(h1=1 | ν1));
In formula: a is visual layers bias vector;B is hidden layer bias vector;W is connection visual layers neuron and hidden layer
The weight matrix of neuron;H is hidden layer neuron vector;ν is visual layers neuron vector;
Step 6: repeat second step to the 5th step, and it is sufficiently small until reaching maximum number of iterations or error, terminate the layer
The training of RBM.
Step 4 trains the reversed fine tuning of DBNC model, i.e., by artificial fish school optimization method in DBNC model
Connection weight and node bias situation optimize;
See Fig. 3, the reversed fine tuning training of DBNC model is as follows using the specific practice of AFSA:
The first step loads training sample, initializes artificial fish-swarm and its parameter: population invariable number N, sensing range, mobile step
Length, crowding, maximum times of looking for food, variable number, maximum number of iterations etc..
Second step generates the artificial fish-swarm of set scale at random in sensing range, and it is current to calculate each Artificial Fish
Food concentration is maximized into bulletin board, the weight and bias of the Artificial Fish individual of food concentration, that is, every.
The food concentration recorded in itself food concentration and billboard is compared by third step, if itself food concentration
It is then the Artificial Fish state by the information update in billboard greater than in bulletin board.Conversely, then continue select food concentration compared with
Big Artificial Fish knocked into the back, is bunched and foraging behavior.
4th step is first carried out one and judges sentence, and whether the number of iterations which is used to judgment method has reached the upper limit,
Or whether resulting optimal solution meets required precision.If then entering the 5th step, third step is otherwise returned.
5th step, after method end of run, by Artificial Fish individual information existing in bulletin board weight and biasing
Output comes out, which is the best initial weights and bias of neuron node.
Step 5, to DBNC model training outcome evaluation, according to the study situation of error judgment model, if error reaches
It is required that step 6 is then carried out, and if not up to accuracy requirement, resume at step two.
According to the study situation of error judgment DBNC model.After to model training, using reserve in advance come C item
Data test the DBNC model of building, and test set includes ten kinds of typical fault modes altogether, respectively to every kind of failure mould
Formula is tested.
The training of the known DBNC model based on AFSA optimization there will necessarily be error, and the presence of this error makes one
Secondary calculated result can not absolutely it is credible, in order to reduce this error as far as possible, improve the confidence level of calculated result, this
Invention uses and repeatedly calculates the method being averaged.Avionics system different faults mode is examined based on both of which identification model
Disconnected situation comparison such as Fig. 4.
This it appears that the DBNC pattern recognition model based on AFSA optimization is relative to DBNC pattern-recognition mould from Fig. 4
Type has higher accuracy rate of diagnosis, is very suitable for the valid model as avionics system pattern-recognition.
Step 6 carries out pattern-recognition to avionics system operation data using trained DBNC model.
The fault data that avionics system is collected into passes through AFSA-DBNC pattern recognition model as network inputs data
Corresponding output data is obtained, after analysis to output data, judges avionics system state in which, avionics system is carried out
Pattern-recognition.
Claims (7)
1. a kind of avionics system mode identification method of deepness belief network, i.e., a kind of to be believed based on artificial fish school optimization method depth
Read the avionics system mode identification method of network, it is characterised in that: the specific steps of which are as follows:
Step 1, the building of DBNC model, setting model parameter initialize;
Step 2 chooses typical fault mode, and the historical failure data for being used for DBNC model training is obtained from avionics system;
Step 3 carries out pre-training to DBNC model using to sdpecific dispersion method, that is, CD method;I.e. to each between layers
The bias of the connection weight of node and each node layer is initialized;
Step 4 trains the reversed fine tuning of DBNC model, i.e., by artificial fish school optimization method to the connection in DBNC model
Weight and node bias situation optimize;
Step 5 wants DBNC model training outcome evaluation according to the study situation of error judgment model if error reaches
It asks, then carries out step 6, if not up to accuracy requirement, resume at step two;
Step 6 carries out pattern-recognition to avionics system operation data using trained DBNC model;
By above step, the error of DBNC model is constantly reduced, and is goed deep into the number of plies, and accuracy obviously rises, table
The training process of the bright model is rationally effective;The model is used for the pattern-recognition of avionics system, has higher diagnosis quasi-
True rate is very suitable for the valid model as avionics system pattern-recognition.
2. a kind of avionics system mode based on artificial fish school optimization method deepness belief network according to claim 1 is known
Other method, it is characterised in that:
In " building of DBNC model " described in step 1, specific practice is as follows: the bottom of building DBNC model is by all layers
RBM composition, top layer are the output layer of Softmax classification function composition;Training data is by all layer of RBM after training by classification layer
Output category result;The specific practice of described " setting model parameter initializes " are as follows: setting hidden layer number, input layer and defeated
Layer neuron node number out, the neuron node number of hidden layer, the connection weight W between each layer, each neuron node are biasings
It is worth initial value, initial learning rate ρ size and study momentum m.
3. a kind of avionics system mode based on artificial fish school optimization method deepness belief network according to claim 1 is known
Other method, it is characterised in that:
" choosing typical fault mode " described in step 2, specific practice is as follows: therefrom choosing ten kinds of typical events
Barrier mode is respectively: 74 failure of GDC, GDC74 configuration module failure, the route of GDC 74 and power failure, 74 sound of GDC
Crimp mouth and pipeline blockage, atmospheric air temperature probe failure, atmospheric air temperature probe route and joint failure, GPS signal failure,
GRS77 failure, GRS77 configuration module failure, GMU44 failure;
The tool of " historical failure data that operation is used for DBNC model training is obtained from avionics system " described in step 2
Body way has: collecting the data of avionics system in the process of running, these data reflect the runing time of avionics system, gas
The roll of pressure, ground level, cabin temperature, delivery temperature, communication/navigation frequency, amount of fuel, oil temperature, aircraft flight
Angle, pitch angle and course angle.
4. a kind of avionics system mode based on artificial fish school optimization method deepness belief network according to claim 1 is known
Other method, it is characterised in that:
Described in step 3 " using to sdpecific dispersion method, that is, CD method to DBNC model carry out pre-training " it is specific
Way are as follows:
The initial state value ν of the first step, initialization network parameter θ and visual layer unit0=x0, setting RBM maximum training iteration time
Number is 400;
Second step calculates all hiding layer unitsFrom condition distribution p (h0j|
ν0) in extract h0j∈{0,1};
In formula:For activation primitive;h0jFor hidden layer neuron vector;WijFor connection visual layers nerve
The weight matrix of member and hidden layer neuron;bjFor hidden layer bias vector;ν is visual layers neuron vector;
Third step calculates all visible elementsFrom p (ν1i|h0) in extract ν1i
∈{0,1};
In formula: wherein ajFor visual layers bias vector;
4th step calculates all hiding layer units
In formula: h0jFor hidden layer neuron vector;WijFor the weight matrix for connecting visual layers neuron and hidden layer neuron;bj
For hidden layer bias vector;ν is visual layers neuron vector;
5th step, by following various update parameters:
In formula: a is visual layers bias vector;B is hidden layer bias vector;W is connection visual layers neuron and hidden layer nerve
The weight matrix of member;H is hidden layer neuron vector;ν is visual layers neuron vector;
Step 6: repeat second step to the 5th step, and it is sufficiently small until reaching maximum number of iterations and reconstructed error, terminate the layer
The training of RBM.
5. a kind of avionics system mode based on artificial fish school optimization method deepness belief network according to claim 1 is known
Other method, it is characterised in that:
Described in step 4 " by artificial fish school optimization method to the connection weight and node bias situation in DBNC model
Optimize ", specific practice is as follows:
The first step loads training sample, initializes artificial fish-swarm and its parameter: population invariable number N, sensing range, that is, Visual, shifting
Dynamic step-length, that is, Step, crowding, maximum times of looking for food, variable number, maximum number of iterations i.e. try number;
Second step generates the artificial fish-swarm of set scale at random in Visual, it is dense to calculate the current food of each Artificial Fish
Degree is maximized into bulletin board, the weight and bias of the Artificial Fish individual of food concentration, that is, every;
The food concentration recorded in itself food concentration and billboard is compared by third step, if itself food concentration is greater than
It is then the Artificial Fish state by the information update in billboard in bulletin board;Conversely, then continuing the people for selecting food concentration big
Work fish knocked into the back, is bunched and foraging behavior;
4th step is first carried out one and judges sentence, and whether the number of iterations which is used to judgment method has reached the upper limit and institute
Whether the optimal solution obtained meets required precision;If then entering the 5th step, third step is otherwise returned;
5th step after this method end of run, the weight in Artificial Fish individual information existing in bulletin board and will bias defeated
It comes out, which is the best initial weights and bias of neuron node.
6. a kind of avionics system mode based on artificial fish school optimization method deepness belief network according to claim 1 is known
Other method, it is characterised in that:
The specific practice of " to DBNC model training outcome evaluation " described in step 5 are as follows: rebuild and instruct first with output valve
This result is then compared with actual input value, obtains accuracy by experienced input, finally compares accuracy and sets in advance
Relationship between fixed threshold value obtains assessment result.
7. a kind of avionics system mode based on artificial fish school optimization method deepness belief network according to claim 1 is known
Other method, it is characterised in that:
The tool of " pattern-recognition is carried out to avionics system operation data using trained DBNC model " described in step 6
Body way is: the fault data that avionics system is collected into is obtained corresponding defeated as network inputs data by DBNC model
Data out after analysis to output data obtain the state of avionics system, to carry out pattern-recognition to avionics system.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810693984.4A CN109034202A (en) | 2018-06-29 | 2018-06-29 | A kind of avionics system mode identification method of deepness belief network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810693984.4A CN109034202A (en) | 2018-06-29 | 2018-06-29 | A kind of avionics system mode identification method of deepness belief network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109034202A true CN109034202A (en) | 2018-12-18 |
Family
ID=65520874
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810693984.4A Pending CN109034202A (en) | 2018-06-29 | 2018-06-29 | A kind of avionics system mode identification method of deepness belief network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109034202A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109884465A (en) * | 2019-03-01 | 2019-06-14 | 辽宁工业大学 | A kind of one-way earth fault localization method based on signal injection method |
CN113189963A (en) * | 2021-04-26 | 2021-07-30 | 东北大学 | Rolling process fault diagnosis method based on unbalanced data |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103793718A (en) * | 2013-12-11 | 2014-05-14 | 台州学院 | Deep study-based facial expression recognition method |
-
2018
- 2018-06-29 CN CN201810693984.4A patent/CN109034202A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103793718A (en) * | 2013-12-11 | 2014-05-14 | 台州学院 | Deep study-based facial expression recognition method |
Non-Patent Citations (8)
Title |
---|
周昕奇: "基于鱼群神经网络的抽油机故障诊断研究", 《中国优秀硕士学位论文全文数据库工程科技I辑》 * |
曾向阳: "《智能水中目标识别》", 31 March 2016 * |
李本威等: "深度置信网络在发动机气路部件性能衰退故障诊断中的应用研究", 《推进技术》 * |
石鑫: "基于深度学习的变压器故障诊断技术研究", 《万方数据知识服务平台》 * |
石鑫等: "基于深度信念网络的电力变压器故障分类建模", 《电力系统保护与控制》 * |
葛强强: "基于深度置信网路的数据驱动故障诊断方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
董海鹰: "《智能控制理论及应用》", 30 September 2016 * |
车畅畅: "基于深度信念网络的航空发动机维修等级决策", 《航空动力学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109884465A (en) * | 2019-03-01 | 2019-06-14 | 辽宁工业大学 | A kind of one-way earth fault localization method based on signal injection method |
CN109884465B (en) * | 2019-03-01 | 2023-09-29 | 辽宁工业大学 | Unidirectional ground fault positioning method based on signal injection method |
CN113189963A (en) * | 2021-04-26 | 2021-07-30 | 东北大学 | Rolling process fault diagnosis method based on unbalanced data |
CN113189963B (en) * | 2021-04-26 | 2024-03-19 | 东北大学 | Rolling process fault diagnosis method based on unbalanced data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112131760B (en) | CBAM model-based prediction method for residual life of aircraft engine | |
US20200402220A1 (en) | On-line real-time diagnosis system and method for wind turbine blade (wtb) damage | |
CN106355151B (en) | A kind of three-dimensional S AR images steganalysis method based on depth confidence network | |
CN101697006B (en) | Target identification method based on data fusion of airborne radar and infrared imaging sensor | |
CN111680875B (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
CN110163258A (en) | A kind of zero sample learning method and system reassigning mechanism based on semantic attribute attention | |
CN110245709A (en) | Based on deep learning and from the 3D point cloud data semantic dividing method of attention | |
CN109829236A (en) | A kind of Compressor Fault Diagnosis method based on XGBoost feature extraction | |
CN106156401A (en) | Data-driven system state model on-line identification methods based on many assembled classifiers | |
CN108256173A (en) | A kind of Gas path fault diagnosis method and system of aero-engine dynamic process | |
CN105608004A (en) | CS-ANN-based software failure prediction method | |
CN110059439A (en) | A kind of spacecraft orbit based on data-driven determines method | |
CN109034202A (en) | A kind of avionics system mode identification method of deepness belief network | |
CN105654136A (en) | Deep learning based automatic target identification method for large-scale remote sensing images | |
CN103020733B (en) | Method and system for predicting single flight noise of airport based on weight | |
CN102968553B (en) | A kind of aircraft lands risk evaluation method | |
CN115877483A (en) | Typhoon path forecasting method based on random forest and GRU | |
Calderano et al. | An enhanced aircraft engine gas path diagnostic method based on upper and lower singleton type-2 fuzzy logic system | |
CN114357372A (en) | Aircraft fault diagnosis model generation method based on multi-sensor data driving | |
CN109685136A (en) | A kind of high-pressure heater status data analysis method | |
CN109143355B (en) | Semi-supervised global optimization seismic facies quantitative analysis method based on SOM | |
CN114637305B (en) | Unmanned aerial vehicle shortest path planning method and device | |
CN112347917B (en) | Gas turbine fault diagnosis method, system, equipment and storage medium | |
CN115017797A (en) | Spacecraft inertial attitude sensor fault diagnosis method based on deep belief network | |
Fan et al. | Construction of a neural network and its application on target classification |
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: 20181218 |
|
RJ01 | Rejection of invention patent application after publication |