CN108594793A - A kind of improved RBF flight control systems fault diagnosis network training method - Google Patents
A kind of improved RBF flight control systems fault diagnosis network training method Download PDFInfo
- Publication number
- CN108594793A CN108594793A CN201810341117.4A CN201810341117A CN108594793A CN 108594793 A CN108594793 A CN 108594793A CN 201810341117 A CN201810341117 A CN 201810341117A CN 108594793 A CN108594793 A CN 108594793A
- Authority
- CN
- China
- Prior art keywords
- gene
- population
- crossover
- rbf
- formula
- 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
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0259—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterized by the response to fault detection
- G05B23/0275—Fault isolation and identification, e.g. classify fault; estimate cause or root of failure
- G05B23/0281—Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
-
- 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
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention belongs to aircraft fault diagnosis technology fields, and in particular to a kind of improved RBF flight control systems fault diagnosis network training method.The present invention uses based on training sample initialization population, carries out crossover operation to network parameter gene based on differential evolution operator and arithmetic crossover operator：Based on training sample initialization population：In initialization of population, the neuronal center of RBF networks is initialized using training sample, network hidden neuron number and radial base extend the method that constant uses random initializtion.Crossover operation based on differential evolution operator and arithmetic crossover operator：In crossover operation, to the crossover operation of network parameter Gene Partial on the basis of the arithmetic crossover operator of generally use, differential evolution operator is used with certain probability, the parameter gene of offspring individual is made to obtain more rich diversity.
Description
Technical field
The invention belongs to aircraft fault diagnosis technology fields, and in particular to a kind of improved RBF flight control systems failure is examined
Circuit network training method.
Background technology
The failure of core system of the flight control system as aircraft, system unit not only affects the property of flight control system
Can, can also be that the flight safety of aircraft brings great threat, intelligent Fault Diagnosis Technique, which is applied to aircraft, flies control system
In the fault diagnosis of system, auxiliary flight crew excludes flight control system failure in time, improves craft preservation efficiency, ensures aircraft
Safe flight, be currently have there is an urgent need to research contents.
The common method of fault diagnosis of flight control system has the fault diagnosis based on model, and the failure based on signal processing is examined
Disconnected and Knowledge based engineering fault diagnosis.
Fault diagnosis based on model is a kind of mathematical model diagnosing object by foundation, is output in by model practical defeated
Go out to calculate system residual error, by analyzing residual error to make a kind of method for diagnosing faults of level diagnosis to failure.But fly
The row device dynamical system complicated as one, many relevant mathematical models of component are difficult often to establish, and residual error amount is also
It can be interfered by many noises so that the application range of such method is limited by very large.
Method based on signal processing is that a kind of signal model by analyzing measurand completes the side of fault diagnosis
Method, commonly the method based on signal model have wavelet analysis method, principle component analysis etc..But the failure letter that such methods utilize
It ceases relatively simple, the uncertain problem in fault diagnosis can not be solved, cause to the diagnosis capability of complex fault relatively
It is weak.
Knowledge based engineering method for diagnosing faults is a kind of independent of mathematical model, and fault diagnosis knowledge is used by study
The method for completing fault diagnosis.Common Knowledge based engineering method for diagnosing faults has：Method based on expert system, based on fuzzy
The method of logic, the method based on grey relational grade, the method for case-based reasioning, is based on information at the method based on rough set
The method of fusion, the method based on artificial neural network, the method based on Bayesian network, the method based on fault tree and base
In the method etc. of support vector machines.Knowledge based engineering method for diagnosing faults since it does not depend on accurate mathematical model, so its
The scope of application is wider, but the acquisition of knowledge often influences the key factor of its performance of fault diagnosis.
Method for diagnosing faults based on RBF neural is a kind of common Knowledge based engineering method for diagnosing faults,
Solve that there is preferable effect in complex fault diagnosis.The performance of fault diagnosis of RBF neural is mainly by its training algorithm
It influences, common training algorithm has the training algorithm based on the center that randomly selects, the training algorithm based on clustering algorithm, based on ladder
Spend the training algorithm of descent method, the training algorithm based on Orthogonal Least Square, the training algorithm based on evolution algorithm.Pass rank something lost
Propagation algorithm is one kind in evolution algorithm, because the hierarchical structure of its chromosome can indicate simultaneously RBF networks network structure and
Network parameter and while training requires relatively low and is widely used in the training of RBF networks to the priori of data, but because
Genetic algorithm belongs to heuristic search algorithm, and the algorithm training time is caused to be lengthened significantly compared to other non-evolutionary training algorithms,
Its application in RBF network trainings is constrained, traditional hierarchy genetic algorithm generally comprises parameter coding, initial population is set
Six elements, the initial population such as fixed, fitness function design, genetic manipulation design, control parameter setting, constraints affect
Iterations, that is, run time of algorithm, traditional passs rank training algorithm generally use random initializtion in initialization of population
Mode；Selection, intersection and variation constitute genetic manipulation, and wherein crossover operation is hierarchy genetic algorithm enhancing population diversity
Primary operational, optimizing ability, iterations important to algorithm.
Invention content
The purpose of the present invention proposes a kind of improved RBF flight control systems fault diagnosis network instruction aiming at the above problem
Practice method.
The technical solution adopted in the present invention is：
A kind of improved RBF flight control systems fault diagnosis network training method, as shown in Figure 1, including the following steps：
S1, initialization：
In initialization of population, the neuronal center of RBF networks, network hidden nodes are initialized using training sample
Mesh and radial base extend constant using random initializtion, and specific method is：
S11, it is encoded to passing rank chromosome, passs rank chromosome structure by control gene and corresponding parameter gene
It constitutes, as shown in Fig. 2, it is table to RBF network parameters that the control gene, which is the expression to RBF network structures, parameter gene,
It reaches；Binary coding is used to control gene, real coding is used to parameter gene, wherein the corresponding base in 1 position of control gene
Because of the neuron of current network, the parameter gene of control is the parameter of corresponding neuron；
S12, initialization population use the extension constant component of control gene and parameter gene the side of random initializtion
Formula initializes the radial base neuronal center part of parameter gene using training sample；
S2, RBF net structure：
S21, chromosome decoding construction hidden layer, by passing the decoding of rank chromosome, obtaining the hidden neuron of RBF networks
Number, neuronal center and extension constant；
S22, hidden layer output is calculated by decoding the hidden layer obtained, and then RBF networks is calculated using least square method
Hidden layer and output layer connection weight, obtain complete RBF networks；
S3, the fitness that RBF networks are calculated by following formula：
Wherein, N indicates that sample number, n indicate that input dimension, M are hidden layer neuron number, and a, b, d are that constant controls
The balance of neural network accuracy and complicated network structure degree,For the corresponding network output of i-th of sample, yiFor desired output；
Determined whether to stop genetic evolution according to the maximum iteration of setting, fitness function desired value, if so, knot
Beam simultaneously exports RBF network parameters, if it is not, then entering step S4；
S4, selection：
Selection operation is carried out to population using roulette operator, obtains population of new generation, specific method is：
S41, the adaptive value summation F for seeking all chromosomes in populationsum；
S42, one 0 is randomly generated to FsumBetween random number k；
S43, sequentially add up each chromosome adaptive value since No. 1 chromosome, until adding up and being greater than or equal to random number
K, it is the individual chosen to be eventually used for cumulative chromosome；
S5, intersection：
The crossover probability of population is calculated according to formula 2：
In formula 2, fmaxFor the maximum adaptation angle value of population, favgFor the average fitness value of population, f' expressions are handed over
Larger adaptive value, P in fork two individuals of operationc1、Pc2For constant, it is generally respectively set to 0.9,0.6, it can be according to particular problem
Do further modification；
Crossover operation is carried out to parameter gene using differential evolution operator shown in formula 3 with probability value ρ：
In formula 3, A represents difference and intersects the new individual generated, and B, C, D indicate to carry out three of difference crossover operation
Body, F are that difference intersects the factor, the constant that β intervals are 0.1 to 0.6, gmaxFor maximum iteration, g is current iteration time
Number；
Remaining individual carries out arithmetic crossover operation by formula 4：
In formula 4, E, F are two individuals that crossover operation is participated in arithmetic crossover operator, and E', F' are newly generated
Body；
Crossover operation is carried out using single-point crossover operator to control Gene Partial, i.e., randomly chooses one in controlling gene and hands over
Vent is set, and the gene after two individual crossover locations to participating in crossover operation carries out；
S6, variation：
The mutation probability that population is calculated according to formula 5 uses basic bit mutation to control Gene Partial, i.e., chooses one at random
Position, inversion operation is carried out to it：
In formula 5, fmaxFor the maximum adaptation angle value of population, favgFor the average fitness value of population, f is indicated into row variation
The fitness value of operator, Pm1、Pm2For constant, it is generally respectively set to 0.1,0.001, can be done according to particular problem into one
The modification of step；
Real number mutation operator is used to parameter gene, as shown in formula 6：
X'=X ± △ (6)
In formula 6, X is the object of mutation operation, and X' is object after variation, and △ is random variation step-length, value range
It is related with the value range of variable；
Return to step S2.
Beneficial effects of the present invention, which are that method of the invention uses, is based on training sample initialization population, improves initialization
The average fitness of population shortens the training time of algorithm, and differential evolution operator enhancing kind is introduced in the crossover operation of algorithm
The diversity of group improves the optimizing ability of algorithm to inhibit the precocity of population.
Description of the drawings
Fig. 1 is the flow diagram of the present invention；
Fig. 2 is to pass rank chromosome structure schematic diagram.
Specific implementation mode
Technical scheme of the present invention is described in detail in Summary, generally speaking, the present invention adopts
Network parameter gene is intersected with based on training sample initialization population, based on differential evolution operator and arithmetic crossover operator
Operation：
Based on training sample initialization population：In initialization of population, the nerve of RBF networks is initialized using training sample
First center, network hidden neuron number and radial base extend the method that constant uses random initializtion；
Crossover operation based on differential evolution operator and arithmetic crossover operator：In crossover operation, to network parameter gene
Partial crossover operation uses differential evolution operator on the basis of the arithmetic crossover operator of generally use, with certain probability, makes
The parameter gene of offspring individual obtains more rich diversity.
Claims (1)
1. a kind of improved RBF flight control systems fault diagnosis network training method, which is characterized in that include the following steps：
S1, initialization：
In initialization of population, using training sample initialize RBF networks neuronal center, network hidden neuron number and
Using random initializtion, specific method is radial base extension constant：
S11, it is encoded to passing rank chromosome, passs rank chromosome structure by control gene and corresponding parameter gene structure
At, the control gene be the expression to RBF network structures, parameter gene it is expression to RBF network parameters；To controlling gene
Using binary coding, real coding is used to parameter gene, wherein the corresponding gene in 1 position of control gene is current network
Neuron, the parameter gene of control are the parameter of corresponding neuron；
S12, initialization population, to control gene and parameter gene extension constant component by the way of random initializtion,
The radial base neuronal center part of parameter gene is initialized using training sample；
S2, RBF net structure：
S21, chromosome decoding construction hidden layer, by passing the decoding of rank chromosome, obtaining the hidden nodes of RBF networks
Mesh, neuronal center and extension constant；
S22, hidden layer output is calculated by decoding the hidden layer obtained, and then the hidden of RBF networks is calculated using least square method
Connection weight containing layer and output layer obtains complete RBF networks；
S3, the fitness that RBF networks are calculated by following formula：
Wherein, N indicates that sample number, n indicate that input dimension, M are hidden layer neuron number, and a, b, d are that constant controls network
The balance of precision and complicated network structure degree,For the corresponding network output of i-th of sample, yiFor desired output；
Determined whether to stop genetic evolution according to the maximum iteration of setting, fitness function desired value, if so, terminating simultaneously
RBF network parameters are exported, if it is not, then entering step S4；
S4, selection：
Selection operation is carried out to population using roulette operator, obtains population of new generation, specific method is：
S41, the adaptive value summation F for seeking all chromosomes in populationsum；
S42, one 0 is randomly generated to FsumBetween random number k；
S43, sequentially add up each chromosome adaptive value since No. 1 chromosome, until adding up and being greater than or equal to random number k, most
It is the individual chosen to be used for cumulative chromosome afterwards；
S5, intersection：
The crossover probability of population is calculated according to formula 2：
In formula 2, fmaxFor the maximum adaptation angle value of population, favgFor the average fitness value of population, f' expressions carry out intersection behaviour
Make adaptive value larger in two individuals, Pc1、Pc2For constant；
Crossover operation is carried out to parameter gene using differential evolution operator shown in formula 3 with probability value ρ：
In formula 3, A represents difference and intersects the new individual generated, and B, C, D indicate to carry out three individuals of difference crossover operation, F
Intersect the factor, the constant that β intervals are 0.1 to 0.6, g for differencemaxFor maximum iteration, g is current iteration number；
Remaining individual carries out arithmetic crossover operation by formula 4：
In formula 4, E, F are two individuals that crossover operation is participated in arithmetic crossover operator, and E', F' are newly generated individual；
Crossover operation is carried out using single-point crossover operator to control Gene Partial, i.e., randomly chooses one in controlling gene and intersects position
It sets, the gene after two individual crossover locations to participating in crossover operation carries out；
S6, variation：
The mutation probability that population is calculated according to formula 5 uses basic bit mutation to control Gene Partial, i.e., chooses one at random,
Inversion operation is carried out to it：
In formula 5, fmaxFor the maximum adaptation angle value of population, favgFor the average fitness value of population, f indicates to carry out mutation operation
The fitness value of individual, Pm1、Pm2For constant；
Real number mutation operator is used to parameter gene, as shown in formula 6：
X'=X ± △ (6)
In formula 6, X is the object of mutation operation, and X' is object after variation, and △ is random variation step-length, value range and change
The value range of amount is related；
Return to step S2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810341117.4A CN108594793A (en) | 2018-04-17 | 2018-04-17 | A kind of improved RBF flight control systems fault diagnosis network training method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810341117.4A CN108594793A (en) | 2018-04-17 | 2018-04-17 | A kind of improved RBF flight control systems fault diagnosis network training method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108594793A true CN108594793A (en) | 2018-09-28 |
Family
ID=63622750
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810341117.4A Pending CN108594793A (en) | 2018-04-17 | 2018-04-17 | A kind of improved RBF flight control systems fault diagnosis network training method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108594793A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109298706A (en) * | 2018-11-02 | 2019-02-01 | 中国航空工业集团公司西安飞机设计研究所 | A kind of flight control system method for diagnosing faults based on Bayesian network |
| CN110060692A (en) * | 2019-04-19 | 2019-07-26 | 山东优化信息科技有限公司 | A kind of Voiceprint Recognition System and its recognition methods |
| CN110262251A (en) * | 2019-07-11 | 2019-09-20 | 电子科技大学 | The prediction of flight control system data and aided diagnosis method based on LSTM neural network |
| CN110458039A (en) * | 2019-07-19 | 2019-11-15 | 华中科技大学 | A kind of construction method of industrial process fault diagnosis model and its application |
| CN112418411A (en) * | 2020-12-16 | 2021-02-26 | 安徽三禾一信息科技有限公司 | Air brake system fault diagnosis based on RBF neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945553A (en) * | 2012-11-27 | 2013-02-27 | 西安电子科技大学 | Remote sensing image partition method based on automatic difference clustering algorithm |
| CN104270281A (en) * | 2014-09-03 | 2015-01-07 | 河海大学 | Web Service QoS (Quality of Service) combined forecasting method |
| CN106411896A (en) * | 2016-09-30 | 2017-02-15 | 重庆邮电大学 | APDE-RBF neural network based network security situation prediction method |
| CN107194524A (en) * | 2017-07-28 | 2017-09-22 | 合肥工业大学 | A kind of Coal and Gas Outbursts Prediction method based on RBF neural |
-
2018
- 2018-04-17 CN CN201810341117.4A patent/CN108594793A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945553A (en) * | 2012-11-27 | 2013-02-27 | 西安电子科技大学 | Remote sensing image partition method based on automatic difference clustering algorithm |
| CN104270281A (en) * | 2014-09-03 | 2015-01-07 | 河海大学 | Web Service QoS (Quality of Service) combined forecasting method |
| CN106411896A (en) * | 2016-09-30 | 2017-02-15 | 重庆邮电大学 | APDE-RBF neural network based network security situation prediction method |
| CN107194524A (en) * | 2017-07-28 | 2017-09-22 | 合肥工业大学 | A kind of Coal and Gas Outbursts Prediction method based on RBF neural |
Non-Patent Citations (2)
| Title |
|---|
| 肖长冬: "RBF网络的改进及其在焊膏印刷厚度预测中的应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 葛珉昊: "基于混合递阶遗传算法的RBF网络优化及在BOD软测量中的应用", 《中国优秀硕士学位论文全文数据库 工程科技I辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109298706A (en) * | 2018-11-02 | 2019-02-01 | 中国航空工业集团公司西安飞机设计研究所 | A kind of flight control system method for diagnosing faults based on Bayesian network |
| CN110060692A (en) * | 2019-04-19 | 2019-07-26 | 山东优化信息科技有限公司 | A kind of Voiceprint Recognition System and its recognition methods |
| CN110262251A (en) * | 2019-07-11 | 2019-09-20 | 电子科技大学 | The prediction of flight control system data and aided diagnosis method based on LSTM neural network |
| CN110458039A (en) * | 2019-07-19 | 2019-11-15 | 华中科技大学 | A kind of construction method of industrial process fault diagnosis model and its application |
| CN112418411A (en) * | 2020-12-16 | 2021-02-26 | 安徽三禾一信息科技有限公司 | Air brake system fault diagnosis based on RBF neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108594793A (en) | A kind of improved RBF flight control systems fault diagnosis network training method | |
| CN108090658A (en) | Arc fault diagnostic method based on time domain charactreristic parameter fusion | |
| CN112329348A (en) | Intelligent decision-making method for military countermeasure game under incomplete information condition | |
| CN108062572A (en) | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on DdAE deep learning models | |
| CN110046710A (en) | A kind of the nonlinear function Extremal optimization method and system of neural network | |
| CN110544011B (en) | Intelligent system combat effectiveness evaluation and optimization method | |
| CN102622515B (en) | A kind of weather prediction method | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| Vaidhehi | The role of Dataset in training ANFIS System for Course Advisor | |
| CN108460461A (en) | Mars earth shear parameters prediction technique based on GA-BP neural networks | |
| Sanz et al. | IIVFDT: Ignorance functions based interval-valued fuzzy decision tree with genetic tuning | |
| CN110414718A (en) | A kind of distribution network reliability index optimization method under deep learning | |
| CN114757351A (en) | Defense method for resisting attack by deep reinforcement learning model | |
| CN105427241A (en) | Distortion correction method for large-field-of-view display device | |
| CN112581264A (en) | Grasshopper algorithm-based credit risk prediction method for optimizing MLP neural network | |
| CN112634019A (en) | Default probability prediction method for optimizing grey neural network based on bacterial foraging algorithm | |
| CN109800517B (en) | Improved reverse modeling method for magnetorheological damper | |
| CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
| Liang et al. | Recent advances in particle swarm optimization via population structuring and individual behavior control | |
| CN105976029A (en) | Cuckoo-behavior-RNA-GA-based bridge crane neural network modeling method | |
| CN103473599A (en) | Genetic algorithm and Kalman filtering based RBFN (Radial Basis Function Networks) combined training method | |
| CN112200208B (en) | Cloud workflow task execution time prediction method based on multi-dimensional feature fusion | |
| Lizhe et al. | BP network model optimized by adaptive genetic algorithms and the application on quality evaluation for class teaching | |
| Ying et al. | A novel optimization algorithm for BP neural network based on RS-MEA | |
| Wang et al. | Research on computer network security evaluation method based on Levenberg-Marquardt algorithms |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180928 |