CN106628097A - Ship equipment fault diagnosis method based on improved radial basis function neutral network - Google Patents
Ship equipment fault diagnosis method based on improved radial basis function neutral network Download PDFInfo
- Publication number
- CN106628097A CN106628097A CN201611060650.0A CN201611060650A CN106628097A CN 106628097 A CN106628097 A CN 106628097A CN 201611060650 A CN201611060650 A CN 201611060650A CN 106628097 A CN106628097 A CN 106628097A
- Authority
- CN
- China
- Prior art keywords
- solution
- nectar source
- ship equipment
- layer
- hidden 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.)
- Granted
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B63—SHIPS OR OTHER WATERBORNE VESSELS; RELATED EQUIPMENT
- B63J—AUXILIARIES ON VESSELS
- B63J99/00—Subject matter not provided for in other groups of this subclass
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B63—SHIPS OR OTHER WATERBORNE VESSELS; RELATED EQUIPMENT
- B63B—SHIPS OR OTHER WATERBORNE VESSELS; EQUIPMENT FOR SHIPPING
- B63B79/00—Monitoring properties or operating parameters of vessels in operation
-
- 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]
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Mechanical Engineering (AREA)
- Ocean & Marine Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a ship equipment fault diagnosis method based on an improved radial basis function neutral network. The method is characterized by comprising the steps of optimizing parameters of a radial basis function neutral network by improving an artificial bee colony algorithm to construct a neutral network classifier; collecting training sample data on ship equipment, importing the training sample data into the constructed neutral network classifier to conduct classification training, and obtaining a trained neutral network classifier; collecting test data on the ship equipment, importing the collected test data into the trained neutral network classifier to conduct classification of faults and judging whether a fault exists or not. According to the ship equipment fault diagnosis method based on the improved radial basis function neutral network, veracity of the fault diagnosis is improved and practicability of the fault diagnosis is increased; meanwhile, the instantaneity requirement of the fault diagnosis of the ship equipment can be met.
Description
Technical field
The present invention relates to it is a kind of based on the ship equipment method for diagnosing faults for improving radial base neural net, belong to LY07 letters
Breath is perceived and technology of identification.
Background technology
Fault diagnosis is using the technologies such as failure data acquisition, fault detect, fault location, fault alarm, integrated use
Various inspections and method of testing, discovery system and equipment produce with the presence or absence of failure and in time alarm signal for fault message
Process.Because ship equipment undertakes particularity, the importance of task, need to carry out many key equipment state data in ship
Uninterrupted sampling, and constantly analyzing, to can be with the most short time when key equipment breaks down, minimum cost finds
Failure, positioning failure simultaneously send in time warning.How low latency, the high inline diagnosis system handling up and continue reliability service are built
System, is current problem demanding prompt solution.
For this purpose, needing a kind of method for diagnosing faults efficiently and accurately former to the equipment of generation during ship's navigation
Barrier is diagnosed.
The content of the invention
Goal of the invention:In order to overcome the deficiencies in the prior art, the present invention to provide a kind of based on improvement radial direction base god
The ship equipment method for diagnosing faults of Jing networks, the method can improve the accuracy of fault diagnosis, increase the suitable of fault diagnosis
With property, while disclosure satisfy that the requirement of real-time of ship fault diagnosis.
Technical scheme:For achieving the above object, the technical solution used in the present invention is:
A kind of ship equipment method for diagnosing faults based on improvement radial base neural net, by improving artificial bee colony algorithm
Neural network classifier is constructed to the parameter optimization of radial base neural net.Training sample data are gathered to ship equipment, will
Training sample data import the neural network classifier of construction and carry out classification based training, obtain the neural network classifier for training.
To ship equipment collecting test data, the test data of collection is imported the neural network classifier for training carries out failure point
Class, judges whether to break down.
Neural network classifier is constructed to the parameter optimization of radial base neural net by improving artificial bee colony algorithm
Method, comprises the following steps:
Step 1, it is determined that improving the |input paramete of artificial bee colony algorithm, the |input paramete includes nectar source quantity SN, maximum
Iterations T, current iteration number of times t=0, subsequent iteration number of times updates upper limit L and nectar source X, wherein RBF neural is joined
Structural parameters of the number as nectar source X.RBF neural parameter includes the number at RBF neural hidden layer excitation function center
C, central value xcWith excitation function width r.
Step 2, using backward learning method to nectar source X={ X1,X2,X3,…,XSNIt is every one-dimensional initialized, its
In, SN is nectar source quantity.It is first random to generate solution search space.Its reverse solution is sought each nectar source again.
Step 3, calculates each solution xijFitness value and be ranked up, fitness value it is worst be chosen as investigate honeybee, remain
Under the first half to lead honeybee, later half records optimal value to keep watch honeybee.
Step 4, leads honeybee according to the adaptive local search approach of the state adjust automatically step-length that can currently solve in neighbour
Scan in domain, if the new explanation for searching replaces current solution v better than current solution with new explanationij, otherwise, retain current solution.
The adaptive local search approach of the state adjust automatically step-length that can currently solve:
Wherein, v 'ijThe jth dimension component in the new nectar source to lead honeybee search to obtain after improvement, vijHoneybee is led to search
The jth dimension component in current nectar source, vbestjFor the jth dimension component in optimum nectar source in epicycle iteration, vkjIt is from m before every wheel iteration
The jth dimension component in the nectar source for preferably randomly selecting in nectar source, m is the random number on [0, SN/2t];
Step 5, solution optimum in epicycle iteration is selected as globally optimal solution X using following formulabest, for following honeybee to make
With.
Wherein, XbestFor globally optimal solution, t is current iteration number of times, and T is the iterations upper limit, xmin jRepresent solution in jth
Minimum of a value in dimension, xmax jRepresent maximum of the solution in jth dimension.
Step 6, according to vectorial XiThe selected Probability p in each nectar source of fitness value calculationi, follow honeybee select probability most
Big nectar source and using the state adjust automatically step-length that can currently solve in step 4 adaptive local search approach per one-dimensional
On scan for.
Step 7:If nectar source is not updated yet after subsequent iteration number of times L is reached, being used by search bee can work as in step 4
One new nectar source of adaptive local search approach random search of the state adjust automatically step-length of front solution is replaced, and remembers
Record optimal solution X so farbest.If having reached maximum iteration time T or having met minimal error precision, optimal solution is exported
XbestAnd corresponding fitness value, otherwise go to step 2.
The solution procedure of the reverse solution in the step 2 is:
oxij=xmin j+xmax j-xij
Wherein, oxijRepresent the reverse solution in i-th solution jth dimension, xmin jThe minimum of a value in initial solution jth dimension is represented,
xmax jRepresent the maximum in initial solution jth dimension, xijRepresent the data in i-th initial solution jth dimension.
Probability p is chosen in the step 6iComputational methods be:
Wherein, fitjRepresent solution vector XiFitness value, fitmaxWith fitminFitness is most in respectively all nectar sources
Big value and minimum of a value.
The radial base neural net is three layer feedforward neural networks, and ground floor is input layer, is responsible for reception processing input
Data, the second layer is hidden layer, is responsible for for the vector of input carrying out Function Mapping, and third layer is output layer, is responsible for classification
As a result export:
The neuron excitation function of radial base neural net:
In formula,For the excitation function of i-th neuron node of hidden layer, i=1,2 ..., m, m are hidden layer nerve
First nodes, x is input vector, ciFor the center of the excitation function of i-th neuron node of hidden layer, σiFor i-th of hidden layer
The center width of the excitation function of neuron node.
The output valve of each hidden layer neuron is multiplied by hidden layer with the connection weight of output layer then as output node layer
Input:
In formula, ykFor the output valve of k-th output node, k=1 ..., o, k is output layer nodes, and m is and the output
The connected hidden neuron number of nodes of node;For the excitation function of i-th neuron node of hidden layer, ωikIt is hiding
The connection weight of i-th neuron node of layer and k-th output node.
Preferably:Connection weight ω of i-th neuron node of hidden layer and k-th output nodeikIt is general to use a most young waiter in a wineshop or an inn
Multiplication or gradient descent method are tried to achieve.
Compared with existing fault diagnosis technology, the ship equipment method for diagnosing faults based on radial base neural net have with
Lower benefit effect:
(1) there can be preferably structure by improving the radial base neural net of artificial bee colony algorithm optimization, improve
Fault diagnosis accuracy and efficiency;
(2) by using neural net method, by learning outcome of the network memory to case, it is not necessary to search for case library,
Increased the judgement speed of failure mode;
(3) if there is new failure to occur, neutral net can with improve artificial bee colony algorithm continue to optimize self structure with
The new fault signature feature of study, improves the flexibility of identification.
Description of the drawings
Fig. 1 is the schematic diagram of radial basis function network structure;
Fig. 2 is based on the ship equipment failure diagnostic process of IABC-RBFNN;
Fig. 3 is the flow chart of the present invention.
Specific embodiment
Below in conjunction with the accompanying drawings and specific embodiment, the present invention is further elucidated, it should be understood that these examples are merely to illustrate this
Invention rather than restriction the scope of the present invention, after the present invention has been read, those skilled in the art are various to the present invention's
The modification of the equivalent form of value falls within the application claims limited range.
A kind of ship equipment method for diagnosing faults based on improvement radial base neural net, as Figure 1-3, by improving
Artificial bee colony algorithm constructs neural network classifier to the parameter optimization of radial base neural net.Ship equipment collection is trained
Sample data, by the neural network classifier that training sample data import construction classification based training is carried out, and obtains the nerve for training
Network classifier.To ship equipment collecting test data, the test data of collection is imported the neural network classifier for training
Failure modes are carried out, judges whether to break down.The present invention is carried out by improving artificial group algorithm to the parameter of neutral net
Optimizing, constructs the neutral net of well-formed, improves the training speed of neutral net, improves the performance of neutral net, solves
Certainly various problems in neural network training process.
(1) improved radial base neural net
Radial basis function neural network (RBFNN) is a kind of three layer feedforward neural networks of good performance, and it is with radially
Basic function passes through Nonlinear Mapping as the activation primitive of hidden layer node, and the data of low-dimensional linearly inseparable are mapped to into higher-dimension
Space, so as to become linear separability, thus it can approach arbitrary nonlinear function, can be parsed with being difficult in processing system
Regularity, not only topological structure is simple, None-linear approximation ability is strong, fast convergence rate, and existence anduniquess most preferably approaches value,
Be successfully applied to fault diagnosis, nonlinear function approach, time series analysis, data classification, pattern-recognition, information processing,
The fields such as image procossing, system modelling and control.As shown in figure 1, the radial base neural net is three layer feedforward neural networks,
Ground floor is input layer, is responsible for reception processing input data, and the second layer is hidden layer, is responsible for for the vector of input entering line function and reflects
Penetrate, be to prepare in higher dimensional space classification, third layer is output layer, be responsible for the result output of classification, wherein, x1,x2,…,xn
To be input into,For the excitation function of hidden layer neuron node, ωij(i=1,2 ..., m;J=1 ..., it is o) each for hidden layer
The connection weight of neuron node and each node of output layer, y1,…,yoFor output.
In RBF neural, input layer to hidden layer is transformed to nonlinear transformation, and from hidden layer to output layer then
For linear transformation.The quantity of hidden layer neuron is determined that how many input node is individual with regard to how many by the quantity of input node
Hidden layer neuron.The input vector of input layer is multiplied by input layer and is multiplied by threshold value b again with the connection weight of hidden layer, constitutes implicit
The input of layer neuron.Radial basis kernel function learning ability compared with other excitation functions is good, and convergence domain is wider, and parameter is less,
Simple structure, it is easy to construct, therefore hidden layer neuron typically chooses Radial basis kernel function as the neuron of RBF neural
Excitation function:
In formula,For the excitation function of i-th neuron node of hidden layer, i=1,2 ..., m, m are hidden layer nerve
First nodes, x is input vector, ciFor the center of the excitation function of i-th neuron node of hidden layer, σiFor i-th of hidden layer
The center width of the excitation function of neuron node, the size of its value illustrates the size of excitation function central role scope, with
And the overlapping degree between each excitation function scope.
The output valve of each hidden layer neuron is multiplied by hidden layer with the connection weight of output layer then as output node layer
Input:
In formula, ykFor the output valve of k-th output node, k=1 ..., o, k is output layer nodes, and m is and the output
The connected hidden neuron number of nodes of node;For the excitation function of i-th neuron node of hidden layer, ωikIt is hiding
The connection weight of i-th neuron node of layer and k-th output node, is typically tried to achieve with least square method or gradient descent method.
Artificial neural network is applied to into the fault diagnosis of ship equipment, is exactly by ship equipment fault diagnosis example and specially
The training study of the diagnostic experiences such as family, learnt fault diagnosis knowledge is expressed with the connection weight for being distributed in network internal
The features such as process, the associative memory to fault mode, pattern match and similar inducing ability that it possesses, can be anti-well
Mirror Nonlinear Mapping relation complicated between failure and sign.At present, artificial neural network is in ship equipment fault diagnosis
Application essentially consist in:Key parameter is selected as input layer, fault parameter is distributed as output layer using artificial neural network
Formula information Store and parallel processing, avoid the trouble of modeling and feature extraction in pattern-recognition, are not inconsistent and special so as to eliminate pattern
Improper the brought impact of extraction is levied, learning gained weights using typical sample carries out pattern-recognition, as shown in Fig. 2 ship sets
The general process of standby fault diagnosis.
(2) improve artificial bee colony algorithm and be applied to radial base neural net parameter optimization
By improving parameter optimization of the artificial bee colony algorithm to radial base neural net, to construct the nerve of function admirable
Network classifier, comprises the following steps:
Step 1, it is determined that improving the |input paramete of artificial bee colony algorithm, the |input paramete includes nectar source quantity SN, maximum
Iterations T, current iteration number of times t=0, subsequent iteration number of times updates upper limit L and nectar source X, wherein RBF neural is joined
Structural parameters of the number as nectar source X.RBF neural parameter includes the number at RBF neural hidden layer excitation function center
C, central value xcWith excitation function width r.
Step 2, using backward learning method to nectar source X={ X1,X2,X3,…,XSNIt is every one-dimensional initialized, its
In, SN is nectar source quantity.It is first random to generate solution search space.Its reverse solution is sought each nectar source again.
By vectorial XiIt is every one-dimensional individually carry out optimizing, can speed up convergence of algorithm.To have more initialized nectar source
Diversity and tend to optimum as far as possible, make full use of search space information, backward learning method to nectar source per it is one-dimensional carry out it is initial
Change:It is first random to generate solution search space;Its reverse solution is sought each nectar source again;Two class solutions are ranked up by fitness, are selected
Fitness preferably nectar source is so favorably improved the search efficiency in optimum nectar source and improves honey as initial nectar source search space
Source quality.
Reversely the solution procedure of solution is:
oxij=xmin j+xmax j-xij (3)
Wherein, oxijRepresent the reverse solution in i-th solution jth dimension, xmin jThe minimum of a value in initial solution jth dimension is represented,
xmax jRepresent the maximum in initial solution jth dimension, xijRepresent the data in i-th initial solution jth dimension.
Step 3, calculates each solution xijFitness value and be ranked up, fitness value it is worst be chosen as investigate honeybee, remain
Under the first half to lead honeybee, later half records optimal value to keep watch honeybee.
Step 4, leads honeybee to pass through the adaptive local search approach according to the state adjust automatically step-length that can currently solve
Scan in neighborhood, if the new explanation for searching replaces current solution v better than current solution with new explanationij, otherwise, retain current
Solution.
The artificial neural network structure for being generally used for fault diagnosis is complex, and interstitial content and the number of plies are all larger, cause
The population quantity in nectar source is larger, and to the training of fault diagnosis grader difficulty is increased.In order to lift local optimal searching ability, accelerate
Convergence of algorithm speed, this paper presents the adaptive Local Search of state adjust automatically step-length that a kind of basis can be solved currently
Method:
Wherein, v 'ijThe jth dimension component in the new nectar source to lead honeybee search to obtain after improvement, vijHoneybee is led to search
The jth dimension component in current nectar source, vbestjFor the jth dimension component in optimum nectar source in epicycle iteration, vkjIt is from m before every wheel iteration
The jth dimension component in the nectar source for preferably randomly selecting in nectar source, m is the random number on [0, SN/2t];
Step 5, solution optimum in epicycle iteration is selected as globally optimal solution X using formula (6)best, for following honeybee to make
With.
In order to improve the local space development ability of artificial bee colony algorithm, search bee uses when new nectar source is found and is based on
The adaptive step strategy of current optimal solution is scanned for, and iteration early stage chooses larger step size, increases hunting zone, lifts convergence
Speed, and less step-length is chosen in the iteration later stage, lift search precision.
Wherein, XbestFor globally optimal solution, t is current iteration number of times, and T is the iterations upper limit, xmin jRepresent solution in jth
Minimum of a value in dimension, xmax jRepresent maximum of the solution in jth dimension.From formula (6), iteration initial stage, the value of t is less, algorithm
Search radius it is larger, be conducive to algorithm to find global more excellent nectar source in a wider context;With the increase of iterations, algorithm
Search radius reduce, contribute to algorithm accelerate convergence rate, value is optimal as early as possible.
Step 6, according to vectorial XiThe selected Probability p in each nectar source of fitness value calculationi, follow honeybee select probability most
Big nectar source and using the state adjust automatically step-length that can currently solve in step 4 adaptive local search approach per one-dimensional
On scan for.
Selected Probability piComputational methods be:
Wherein, fitjRepresent vector XiFitness value, fitmaxWith fitminThe maximum of fitness in respectively all nectar sources
Value and minimum of a value.The selected probability in nectar source with more excellent fitness can so be increased, so as to accelerate convergence of algorithm speed
Degree.
Step 7:If nectar source is not updated yet after subsequent iteration number of times L is reached, being used by search bee can work as in step 4
One new nectar source of adaptive local search approach random search of the state adjust automatically step-length of front solution is replaced, and remembers
Record optimal solution X so farbest.If having reached maximum iteration time T or having met minimal error precision, optimal solution is exported
XbestAnd corresponding fitness value, otherwise go to step 2.
The above is only the preferred embodiment of the present invention, it should be pointed out that:For the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (6)
1. a kind of based on the ship equipment method for diagnosing faults for improving radial base neural net, it is characterised in that:By improving people
Work ant colony algorithm constructs neural network classifier to the parameter optimization of radial base neural net;To ship equipment collection training sample
Notebook data, by the neural network classifier that training sample data import construction classification based training is carried out, and obtains the nerve net for training
Network grader;To ship equipment collecting test data, the test data of collection is imported the neural network classifier for training and is entered
Row failure modes, judge whether to break down.
2. according to claim 1 based on the ship equipment method for diagnosing faults for improving radial base neural net, its feature
It is:The side of neural network classifier is constructed to the parameter optimization of radial base neural net by improving artificial bee colony algorithm
Method, comprises the following steps:
Step 1, it is determined that improving the |input paramete of artificial bee colony algorithm, the |input paramete includes nectar source quantity SN, greatest iteration
Number of times T, current iteration number of times t=0, subsequent iteration number of times updates upper limit L and nectar source X, wherein RBF neural parameter is made
For the structural parameters of nectar source X;The number C of RBF neural parameter including RBF neural hidden layer excitation function center, in
Center value xcWith excitation function width r;
Step 2, using backward learning method to nectar source X={ X1,X2,X3,…,XSNIt is every one-dimensional initialized, wherein, SN
For nectar source quantity;It is first random to generate solution search space;Its reverse solution is sought each nectar source again;
Step 3, calculates each solution XiFitness value and be ranked up, fitness value it is worst be chosen as investigate honeybee, it is remaining before
Half records optimal value to lead honeybee, later half to keep watch honeybee;
Step 4, leads honeybee according to the adaptive local search approach of the state adjust automatically step-length that can currently solve in neighborhood
Scan for, if the new explanation for searching replaces current solution v better than current solution with new explanationij, otherwise, retain current solution;
The adaptive local search approach of the state adjust automatically step-length that can currently solve:
Wherein, v 'ijThe jth dimension component in the new nectar source to lead honeybee search to obtain after improvement, vijIt is current for lead honeybee to search
The jth dimension component in nectar source, vbestjFor the jth dimension component in optimum nectar source in epicycle iteration, vkjIt is more excellent from m before every wheel iteration
Nectar source in randomly select a nectar source jth dimension component, m be one [0, SN/2t] on random number;
Step 5, solution optimum in epicycle iteration is selected as globally optimal solution X using following formulabest, for following honeybee to use;
Wherein, XbestFor globally optimal solution, t is current iteration number of times, and T is the iterations upper limit, xmin jRepresent all solution jth dimensions
On minimum of a value, xmax jRepresent the maximum in all solution jth dimensions;
Step 6, according to vectorial XiThe selected Probability p in jth dimension component fitness value calculation each nectar sourcei, follow honeybee to select general
The maximum nectar source of rate simultaneously utilizes the adaptive local search approach of the state adjust automatically step-length that can currently solve in step 4 every
Scan on one-dimensional;
Step 7:If nectar source is not updated yet after subsequent iteration number of times L is reached, can currently be solved using in step 4 by search bee
One new nectar source of adaptive local search approach random search of state adjust automatically step-length be replaced, and record mesh
Before till optimal solution Xbest;If having reached maximum iteration time T or having met minimal error precision, optimal solution X is exportedbest
And corresponding fitness value, otherwise go to step 2.
3. according to claim 2 based on the ship equipment method for diagnosing faults for improving radial base neural net, its feature
It is:The solution procedure of the reverse solution in the step 2 is:
oxij=xminj+xmaxj-xij
Wherein, oxijRepresent the reverse solution in i-th solution jth dimension, xmin jRepresent the minimum of a value in initial solution jth dimension, xmax jTable
Show the maximum in initial solution jth dimension, xijRepresent the data in i-th initial solution jth dimension.
4. according to claim 2 based on the ship equipment method for diagnosing faults for improving radial base neural net, its feature
It is:Probability p is chosen in the step 6iComputational methods be:
Wherein, fitjRepresent vector XiFitness value, fitmaxWith fitminJth dimension component fitness in respectively all nectar sources
Maxima and minima.
5. according to claim 1 based on the ship equipment method for diagnosing faults for improving radial base neural net, its feature
It is:The radial base neural net is three layer feedforward neural networks, and ground floor is input layer, is responsible for reception processing input number
According to the second layer is hidden layer, is responsible for for the vector of input carrying out Function Mapping, and third layer is output layer, is responsible for the knot of classification
Fruit exports:
The neuron excitation function of radial base neural net:
In formula,For the excitation function of i-th neuron node of hidden layer, i=1,2 ..., m, m are hidden layer neuron section
Points, x is input vector, ciFor the center of the excitation function of i-th neuron node of hidden layer, σiIt is neural for i-th of hidden layer
The center width of the excitation function of first node;
The output valve of each hidden layer neuron is multiplied by hidden layer with the connection weight of output layer then as the input of output node layer:
In formula, ykFor the output valve of k-th output node, k=1 ..., o, k is output layer nodes, and m is and the output node phase
Hidden neuron number of nodes even;For the excitation function of i-th neuron node of hidden layer, ωikFor hidden layer i-th
The connection weight of individual neuron node and k-th output node.
6. according to claim 6 based on the ship equipment method for diagnosing faults for improving radial base neural net, its feature
It is:Connection weight ω of i-th neuron node of hidden layer and k-th output nodeikTypically with least square method or gradient
Descent method is tried to achieve.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611060650.0A CN106628097B (en) | 2016-11-22 | 2016-11-22 | A kind of ship equipment method for diagnosing faults based on improvement radial base neural net |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611060650.0A CN106628097B (en) | 2016-11-22 | 2016-11-22 | A kind of ship equipment method for diagnosing faults based on improvement radial base neural net |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106628097A true CN106628097A (en) | 2017-05-10 |
CN106628097B CN106628097B (en) | 2018-08-28 |
Family
ID=58812259
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611060650.0A Expired - Fee Related CN106628097B (en) | 2016-11-22 | 2016-11-22 | A kind of ship equipment method for diagnosing faults based on improvement radial base neural net |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106628097B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107316059A (en) * | 2017-06-16 | 2017-11-03 | 陕西师范大学 | Learner's gesture recognition method |
CN107588937A (en) * | 2017-09-06 | 2018-01-16 | 成都大汇物联科技有限公司 | A kind of hydro-generating Unit equipment health analysis method |
CN107797149A (en) * | 2017-11-10 | 2018-03-13 | 广东工业大学 | A kind of ship classification method and device |
CN109062177A (en) * | 2018-06-29 | 2018-12-21 | 无锡易通精密机械股份有限公司 | A kind of Trouble Diagnostic Method of Machinery Equipment neural network based and system |
CN109116821A (en) * | 2018-09-04 | 2019-01-01 | 江苏远望仪器集团有限公司 | A kind of control and monitoring system for ships, method and apparatus |
CN109508470A (en) * | 2017-09-22 | 2019-03-22 | 广东工业大学 | The method for establishing ship weight computation model based on deep neural network study |
CN110119349A (en) * | 2018-02-05 | 2019-08-13 | 罗伯特·博世有限公司 | Method and apparatus for examining the function of neural network |
CN110399928A (en) * | 2019-07-29 | 2019-11-01 | 集美大学 | Voltage of solid oxide fuel cell prediction technique, terminal device and storage medium |
CN110705797A (en) * | 2019-10-09 | 2020-01-17 | 浙江海洋大学 | Ship oil consumption data prediction method based on ship sensor network |
CN112270270A (en) * | 2020-10-30 | 2021-01-26 | 湘潭大学 | Contact network fault type identification method based on ABC optimization neural network |
CN112990852A (en) * | 2021-02-23 | 2021-06-18 | 杭州安恒信息技术股份有限公司 | Junk mail filtering method and device, electronic device and storage medium |
CN113688559A (en) * | 2021-06-24 | 2021-11-23 | 上海大学 | Seawater desalination system fault diagnosis method based on improved selective evolution random network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102063109A (en) * | 2010-11-29 | 2011-05-18 | 株洲南车时代电气股份有限公司 | Neural network-based subway train fault diagnosis device and method |
CN102707713A (en) * | 2012-06-15 | 2012-10-03 | 长春工业大学 | Fault diagnosis system and method for automobile safety air bag assembly working procedure |
CN103808509A (en) * | 2014-02-19 | 2014-05-21 | 华北电力大学(保定) | Fan gear box fault diagnosis method based on artificial intelligence algorithm |
CN104965941A (en) * | 2015-06-02 | 2015-10-07 | 上海电力学院 | Magnetic flux leakage testing defect reconstruction method based on improved artificial bee colony algorithm |
-
2016
- 2016-11-22 CN CN201611060650.0A patent/CN106628097B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102063109A (en) * | 2010-11-29 | 2011-05-18 | 株洲南车时代电气股份有限公司 | Neural network-based subway train fault diagnosis device and method |
CN102707713A (en) * | 2012-06-15 | 2012-10-03 | 长春工业大学 | Fault diagnosis system and method for automobile safety air bag assembly working procedure |
CN103808509A (en) * | 2014-02-19 | 2014-05-21 | 华北电力大学(保定) | Fan gear box fault diagnosis method based on artificial intelligence algorithm |
CN104965941A (en) * | 2015-06-02 | 2015-10-07 | 上海电力学院 | Magnetic flux leakage testing defect reconstruction method based on improved artificial bee colony algorithm |
Non-Patent Citations (2)
Title |
---|
印洪浩: "基于人工蜂群优化支持向量机的船舶管网泄漏识别", 《机床与液压》 * |
尹川: "基于人工蜂群优化机理的船舶电力***故障诊断", 《CNKI优秀硕士学位论文全文库》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107316059B (en) * | 2017-06-16 | 2020-07-28 | 陕西师范大学 | Learner gesture recognition method |
CN107316059A (en) * | 2017-06-16 | 2017-11-03 | 陕西师范大学 | Learner's gesture recognition method |
CN107588937A (en) * | 2017-09-06 | 2018-01-16 | 成都大汇物联科技有限公司 | A kind of hydro-generating Unit equipment health analysis method |
CN107588937B (en) * | 2017-09-06 | 2019-11-05 | 成都大汇物联科技有限公司 | A kind of hydro-generating Unit equipment health analysis method |
CN109508470A (en) * | 2017-09-22 | 2019-03-22 | 广东工业大学 | The method for establishing ship weight computation model based on deep neural network study |
CN109508470B (en) * | 2017-09-22 | 2022-06-14 | 广东工业大学 | Method for establishing ship weight calculation model based on deep neural network learning |
CN107797149A (en) * | 2017-11-10 | 2018-03-13 | 广东工业大学 | A kind of ship classification method and device |
CN110119349A (en) * | 2018-02-05 | 2019-08-13 | 罗伯特·博世有限公司 | Method and apparatus for examining the function of neural network |
CN109062177A (en) * | 2018-06-29 | 2018-12-21 | 无锡易通精密机械股份有限公司 | A kind of Trouble Diagnostic Method of Machinery Equipment neural network based and system |
CN109116821A (en) * | 2018-09-04 | 2019-01-01 | 江苏远望仪器集团有限公司 | A kind of control and monitoring system for ships, method and apparatus |
CN110399928A (en) * | 2019-07-29 | 2019-11-01 | 集美大学 | Voltage of solid oxide fuel cell prediction technique, terminal device and storage medium |
CN110705797A (en) * | 2019-10-09 | 2020-01-17 | 浙江海洋大学 | Ship oil consumption data prediction method based on ship sensor network |
CN110705797B (en) * | 2019-10-09 | 2023-09-22 | 浙江海洋大学 | Ship fuel consumption data prediction method based on ship sensing network |
CN112270270A (en) * | 2020-10-30 | 2021-01-26 | 湘潭大学 | Contact network fault type identification method based on ABC optimization neural network |
CN112990852A (en) * | 2021-02-23 | 2021-06-18 | 杭州安恒信息技术股份有限公司 | Junk mail filtering method and device, electronic device and storage medium |
CN112990852B (en) * | 2021-02-23 | 2024-03-29 | 杭州安恒信息技术股份有限公司 | Junk mail filtering method, device, electronic device and storage medium |
CN113688559A (en) * | 2021-06-24 | 2021-11-23 | 上海大学 | Seawater desalination system fault diagnosis method based on improved selective evolution random network |
CN113688559B (en) * | 2021-06-24 | 2023-11-17 | 上海大学 | Sea water desalination system fault diagnosis method based on improved selective evolution random network |
Also Published As
Publication number | Publication date |
---|---|
CN106628097B (en) | 2018-08-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106628097A (en) | Ship equipment fault diagnosis method based on improved radial basis function neutral network | |
CN101464964B (en) | Pattern recognition method capable of holding vectorial machine for equipment fault diagnosis | |
CN108228716A (en) | SMOTE_Bagging integrated sewage handling failure diagnostic methods based on weighting extreme learning machine | |
CN110213244A (en) | A kind of network inbreak detection method based on space-time characteristic fusion | |
CN108062572A (en) | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on DdAE deep learning models | |
CN103728551B (en) | A kind of analog-circuit fault diagnosis method based on cascade integrated classifier | |
CN109800875A (en) | Chemical industry fault detection method based on particle group optimizing and noise reduction sparse coding machine | |
CN106570513A (en) | Fault diagnosis method and apparatus for big data network system | |
CN102651088B (en) | Classification method for malicious code based on A_Kohonen neural network | |
CN110533070A (en) | A kind of mechanical breakdown intelligent diagnosing method based on migration original shape network under small sample | |
CN105975504A (en) | Recurrent neural network-based social network message burst detection method and system | |
CN108958217A (en) | A kind of CAN bus message method for detecting abnormality based on deep learning | |
CN109298993A (en) | A kind of method, apparatus and computer readable storage medium detecting failure | |
CN110516305A (en) | Intelligent fault diagnosis method under small sample based on attention mechanism meta-learning model | |
CN110009030A (en) | Sewage treatment method for diagnosing faults based on stacking meta learning strategy | |
CN106874963B (en) | A kind of Fault Diagnosis Method for Distribution Networks and system based on big data technology | |
CN110363230A (en) | Stacking integrated sewage handling failure diagnostic method based on weighting base classifier | |
CN101871994B (en) | Method for diagnosing faults of analog circuit of multi-fractional order information fusion | |
CN110247910A (en) | A kind of detection method of abnormal flow, system and associated component | |
CN111931505A (en) | Cross-language entity alignment method based on subgraph embedding | |
CN105487009A (en) | Motor fault diagnosis method based on k-means RBF neural network algorithm | |
CN103092975A (en) | Detection and filter method of network community garbage information based on topic consensus coverage rate | |
CN105893876A (en) | Chip hardware Trojan horse detection method and system | |
CN109858503A (en) | The traction converter failure diagnostic method of decision tree is promoted based on gradient | |
CN117034143B (en) | Distributed system fault diagnosis method and device based on machine learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180828 Termination date: 20211122 |