CN106411896B - Network security situation prediction method based on APDE-RBF neural network - Google Patents
Network security situation prediction method based on APDE-RBF neural network Download PDFInfo
- Publication number
- CN106411896B CN106411896B CN201610871705.XA CN201610871705A CN106411896B CN 106411896 B CN106411896 B CN 106411896B CN 201610871705 A CN201610871705 A CN 201610871705A CN 106411896 B CN106411896 B CN 106411896B
- Authority
- CN
- China
- Prior art keywords
- individual
- degree
- population
- indicate
- matrix
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0418—Architecture, e.g. interconnection topology using chaos or fractal principles
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/20—Network architectures or network communication protocols for network security for managing network security; network security policies in general
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to technical field of network security, in particular to the network security situation prediction method of a kind of radial basis function APDE-RBF neural network that differential evolution algorithm is propagated based on attractor, partition clustering is carried out to sample data including the use of AP clustering algorithm, to obtain the center of radial basis function RBF and the node in hidden layer of network;Group difference degree is obtained using AP cluster, adaptively changes the zoom factor and crossover probability of DE algorithm, the width and connection weight of RBF are optimized;Simultaneously in order to avoid falling into local optimum and jumping out Local Extremum, Chaos Search is carried out to the elite individual of every generation population and group difference degree center;Determine that final RBF network model, input test data set export situation predicted value;While the present invention is directed to enhance generalization ability, the precision of prediction to network safety situation is improved.
Description
Technical field
The invention belongs to technical field of network security, in particular to a kind of diameter that differential evolution algorithm is propagated based on attractor
To basic function (Affinity Propagation Differential evolution-Radial Basis Function, letter
Claim APDE-RBF) neural network network security situation prediction method.
Background technique
It is aobvious according to " the 35th China Internet state of development report " of China Internet Network Information Center's in January, 2015 publication
Show there is 46.3% netizen to meet with network security problem in China totality netizen in by the end of December, 2014 within cut-off, show China
The safe condition that people internet uses allows of no optimist.With network security problem become increasingly conspicuous with seriously, some traditional safety
Defense technique is unable to do what one wishes, to solve the above problems, the research of network security situation awareness is come into being.
The network safety situation prediction mainly network administrator before network loss under attack takes corresponding measure,
According to current and previous network safety situation value, reasonable mathematical model is established to the network safe state of following a period of time
It is predicted.Since network attack is random and uncertain, so being a complicated non-linear mistake to the prediction of situation value
Journey.
Researcher proposes the method much predicted at present, such as statistical method, gray prediction method, neural network side
Method, Markov model, support vector machines etc., but all there is respective limitation and deficiency in the above method.
The more common model of statistical method has: autoregression model, moving average model and autoregressive moving-average model,
However the following limitations exist for these models: the data demand of time series is steady, if it is multiple regression, also require variable it
Between be independent;Gray prediction method is applicable in the time series being monotonically changed, for fluctuate biggish time series it is difficult to predict;
1987, Lapdes et al. was first by Application of Neural Network in the study of the time series emulation data generated by computer and pre-
Survey, but neural network there are convergence rates it is slow, structure choice is difficult and is easily trapped into local minimum the problems such as, simultaneously because the party
Method is affected by complicated network structure degree and sample complex, thus will appear overfitting or the low phenomenon of generalization ability;
Markov model needs the mathematical formulae of large amount of complex to derive, it is difficult to establish accurate prediction model;Support vector machines
(Support Vector Machine, abbreviation SVM) is difficult to carry out large-scale training sample, and convergence rate is slow.
Summary of the invention
In view of the above shortcomings of the prior art, the present invention proposes a kind of network security state based on APDE-RBF neural network
Gesture prediction technique, it is intended to while enhancing generalization ability, improve the precision of prediction to network safety situation.
To achieve the above object, a kind of network safety situation prediction based on APDE-RBF neural network proposed by the present invention
Method, which is characterized in that the prediction technique the following steps are included:
Step 1: using attraction propagate (Affinity Propagation, abbreviation AP) clustering algorithm to sample data into
Row partition clustering, to obtain the center of RBF and the node in hidden layer of network;
Step 2: obtaining group difference degree using AP cluster, adaptively change differential evolution (Differential
Evolution, vehicle economy) algorithm zoom factor and crossover probability, the width and connection weight of RBF are optimized;
Step 3: in order to avoid falling into local optimum and jumping out Local Extremum, to the elite individual of every generation population and
Group difference degree center carries out Chaos Search;
Step 4: determining that final RBF network model, input test data set export situation predicted value.
Preferably, the step 1 further includes steps of
Step 11: calculating the similarity matrix S between input node using Euclidean distance are as follows: S (i, k)=- | | xi-xk|
|2, wherein xiAnd xkIndicate RBF neural any two input node, S (i, k) indicates point xkAs point xiCluster centre
Similarity;
Step 12: initializing Attraction Degree matrix R and degree of membership matrix A is R (i, k)=0, A (i, k)=0, wherein R (i, k)
Indicate point xkIt is suitable as data point xiCluster centre degree, A (i, k) indicate point xiSelected element xkAs its cluster centre
Appropriateness;
Step 13: determining and be biased to parameterWherein N indicates the quantity of input node,
Intermediate numerical value is occupy in one group of numerical value of median function representation;
Step 14: Attraction Degree matrix R and degree of membership matrix A are calculated according to following formula:
Wherein p (k) indicates data point xkAs the point of reference of cluster centre, R (k, k) indicates data point xkIt is suitable as certainly
The degree of oneself cluster centre, A (k, k') indicate data point xkSelect data point xk'As the degree of its cluster centre, S (k,
K' data point x) is indicatedkWith data point xk'Similarity degree;From above-mentioned formula can be seen that when p (k) is larger its corresponding R (k,
K) also can be larger, and then A (i, k) value also can be larger, so as to represent a possibility that k is as final cluster centre larger for class;Phase
Ying Di, when more p (k) is larger, more class representatives tends to final cluster centre, therefore, increases or subtracts
Small p (k) can increase or decrease the clusters number of AP output;
Step 15: updating the formula of Attraction Degree matrix R and degree of membership matrix A are as follows:
R (i, k)=λ * R (i, k)old+(1-λ)*R(i,k)new
A (i, k)=λ * A (i, k)old+(1-λ)*A(i,k)new
When above-mentioned more new formula indicates each iteration, new Attraction Degree matrix R (i, k)newWith degree of membership matrix A (i, k)new
Will respectively with last R (i, k)oldWith A (i, k)oldIt is weighted update, obtains the Attraction Degree matrix and ownership of the secondary iteration
Matrix is spent, wherein λ indicates updating factor;
Step 16: if meeting one of the following conditions: the class center 1. selected keeps stablizing, and is 2. more than greatest iteration time
Number, then go to step 17, otherwise go to step 14;
Step 17, cluster result is exported.
Preferably, the step 2 further includes steps of
Step 21: initialization is executed, process is as follows:
σi=σmin+rand(0,1)*(σmax-σmin)
wi=rand (0,1)
Wherein σiFor RBF neural basic function width, σmaxIndicate in all sample number strong points two farthest data points
Apart from width, its calculation formula is:σminIndicate in all sample number strong points two most
Nearly data point apart from width, its calculation formula is:wiIndicate hidden layer to output
Layer connection weight, rand (0,1) indicate equally distributed random number between (0,1);
Step 22: mutation process is executed, by g+1 for variation individual V in populationi(g+1) g is modeled as in population three
The function of individual:
Vi(g+1)=Xr1(g)+F*(Xr2(g)-Xr3(g))
i≠r1≠r2≠r3
Wherein XiIt (g) is g for i-th of individual, i.e. X in populationr1(g)、Xr2(g) and Xr3(g) g generation kind is respectively indicated
R1, r2 and the r3 individuals in group, F is zoom factor;
Step 23: executing crossover process, generate i-th of jth reform individual u of g+1 generationij(g+1) formula are as follows:
Wherein vij(g+1) indicate that g carries out the individual after mutation operation, x for i-th of jth dimension individual of populationij(g) it indicates
For g for i-th of jth dimension individual of population, rand is equally distributed random number between (0,1), jrandIt is random whole between [1, n]
Number, CR indicate crossover probability;Above-mentioned formula meaning is when stochastic variable rand is less than element pair in crossover probability CR or individual
Ordinal number j is answered to be equal to stochastic variable jrand, i.e., using the element in variation individual as new individual, it is intended to which that improves individual variation can
It can property;Otherwise, target individual x is still keptij(g) constant;
Step 24: selection course is executed, as follows:
Wherein UiIt (g+1) is candidate individual, XiIt (g) is corresponding individual, f () is the fitness function of individual, is used herein as
Mean square error (mean square error, abbreviation MSE) is used as fitness function.
Further, the formula of dynamic adjustment is carried out in the step 22 to zoom factor F are as follows:
Wherein FmaxAnd FminThe bound of zoom factor is respectively indicated, PD (g) is the group difference degree in g generation, and is planted
The expression of constellation variance degree carries out individuals all in population space to cluster obtained cluster number, when group difference degree is bigger,
Individual is distributed more uniform in population space, and it is bigger to acquire globally optimal solution possibility;τ1For iteration threshold, gmaxIt changes for maximum
Generation number.
Further, the formula that adjust crossover probability CR can adaptively in the step 23 are as follows:
Wherein CRminAnd CRmaxRespectively indicate the bound of crossover probability, τ2For the iteration threshold of setting.
Preferably, Chaos Search implements in the step 3 are as follows: modeling 1 dimensional Logistic Map chaos mould first
Type, expression formula are as follows: Zt+1=μ Zt(1-Zt), wherein μ is control parameter,It is one to generate at random
D dimensional vector, t indicate chaos iteration number;
Secondly, optimum individual and diversity factor center iteration more new formula in modeling population:
Wherein XiIndicate optimum individual or the diversity factor center of population,New individual after indicating Chaos Search, α table
Show chaos adjustment parameter, r is the random number between [0,1].
The beneficial effects of the present invention are: group difference degree is obtained using AP clustering algorithm proposed by the present invention, and adaptive
Change DE algorithm zoom factor and crossover probability with answering not only optimizes the width and connection weight of RBF, but also to every generation
The elite individual of population and group difference degree center carry out Chaos Search, avoid and fall into local optimum, are not only able to enhance
Generalization ability, and can be improved precision of prediction.
Detailed description of the invention
Fig. 1 is the preferred embodiment provided by the invention based on APDE-RBF neural network network security situation prediction method
Flow chart;
Fig. 2 is algorithms of different situation value prediction contrast simulation figure;
Fig. 3 is the different error contrast simulation figures of algorithms of different;
Fig. 4 is different improvement DE algorithm prediction contrast simulation figures;
Fig. 5 is the different different error contrast simulation figures for improving DE algorithm.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawing to tool of the invention
Body embodiment is described in further detail.
Fig. 1 is the preferred implementation of the network security situation prediction method provided by the invention based on APDE-RBF neural network
Example flow chart, this method specifically includes the following steps:
Step 1: partition clustering being carried out to sample data using AP clustering algorithm, to obtain center and the network of RBF
Node in hidden layer;
Step 2: obtaining group difference degree using AP cluster, adaptively change the zoom factor of DE algorithm and intersect general
Rate optimizes the width and connection weight of radial basis function RBF;
Step 3: in order to avoid falling into local optimum and jumping out Local Extremum, to the elite individual of every generation population and
Group difference degree center carries out Chaos Search;
Step 4: determining that final RBF network model, input test data set export situation predicted value.
According to the present invention, it is further included steps of in the step 1
Step 11: calculating the similarity matrix S between input node using Euclidean distance are as follows: S (i, k)=- | | xi-xk|
|2, wherein xiAnd xkIndicate RBF neural any two input node, S (i, k) indicates point xkAs point xiCluster centre
Similarity;
Step 12: initializing Attraction Degree matrix R and degree of membership matrix A is R (i, k)=0, A (i, k)=0, wherein R (i, k)
Indicate point xkIt is suitable as data point xiCluster centre degree, A (i, k) indicate point xiSelected element xkAs its cluster centre
Appropriateness;
Step 13: determining and be biased to parameterpkIndicate that each sample data point is selected as clustering
A possibility that center, be the value of element on similar matrix S diagonal line, k=1 ..., N, and N indicates the quantity of input node,
Intermediate numerical value is occupy in one group of numerical value of median function representation;
Step 14: Attraction Degree matrix R and degree of membership matrix A are calculated according to following formula:
Wherein p (k) indicates data point xkAs the point of reference of cluster centre, R (k, k) indicates data point xkIt is suitable as certainly
The degree of oneself cluster centre, A (k, k') indicate data point xkSelect data point xk' degree as its cluster centre, S (k,
K' data point x) is indicatedkWith data point xk' similarity degree;From above-mentioned formula can be seen that when p (k) is larger its corresponding R (k,
K) also can be larger, and then A (i, k) value also can be larger, so as to represent a possibility that k is as final cluster centre larger for class;Phase
Ying Di, when more p (k) is larger, more class representatives tends to final cluster centre, therefore, increases or subtracts
Small p (k) can increase or decrease the clusters number of AP output;
Step 15: updating the formula of Attraction Degree matrix R and degree of membership matrix A are as follows:
R (i, k)=λ * R (i, k)old+(1-λ)*R(i,k)new
A (i, k)=λ * A (i, k)old+(1-λ)*A(i,k)new
When above-mentioned more new formula indicates each iteration, new Attraction Degree matrix R (i, k)newWith degree of membership matrix A (i, k)new
Will respectively with last R (i, k)oldWith A (i, k)oldIt is weighted update, obtains the Attraction Degree matrix and ownership of the secondary iteration
Matrix is spent, wherein λ indicates updating factor;
Step 16: if meeting one of the following conditions: the class center 1. selected keeps stablizing, and is 2. more than greatest iteration time
Number, then go to step 17, otherwise go to step 14;
Step 17, cluster result is exported.
According to the present invention, it is further included steps of in the step 2
Step 21: initialization is executed, process is as follows:
σi=σmin+rand(0,1)*(σmax-σmin)
wi=rand (0,1)
Wherein σiFor RBF neural basic function width, σmaxIndicate in all sample number strong points two farthest data points
Apart from width, its calculation formula is:σminIndicate in all sample number strong points two most
Nearly data point apart from width, its calculation formula is:ci、cjIndicate any two not
Same hidden layer node, wiIndicate hidden layer arrive output layer connection weight, rand (0,1) expression (0,1) between it is equally distributed with
Machine number;
Step 22: mutation process is executed, by g+1 for variation individual V in populationi(g+1) g is modeled as in population three
The function of individual:
Vi(g+1)=Xr1(g)+F*(Xr2(g)-Xr3(g))i≠r1≠r2≠r3
Wherein Xr1(g)、Xr2(g) and Xr3(g) g is respectively indicated for r1 in population, r2 and r3
Body, F are zoom factor;
Step 23: executing crossover process, generate new individual uij(g+1) formula is as follows:
Wherein vij(g+1) indicate that g carries out the individual after mutation operation, x for i-th of jth dimension individual of populationij(g) it indicates
For g for i-th of jth dimension individual of population, rand is equally distributed random number between (0,1), jrandIt is random whole between [1, n]
Number, CR indicate crossover probability;Above-mentioned formula meaning is when stochastic variable rand is less than element pair in crossover probability CR or individual
Ordinal number j is answered to be equal to stochastic variable jrand, i.e., using the element in variation individual as new individual, it is intended to which that improves individual variation can
It can property;Otherwise, target individual x is still keptij(g) constant;
Step 24: selection course is executed, specific as follows:
Wherein UiIt (g+1) is candidate individual, XiIt (g) is corresponding individual, f () is the fitness function of individual, is used herein as
Mean square error (mean square error, abbreviation MSE) is used as fitness function.
According to the present invention, the formula that dynamic adjustment is carried out to F in the step 22 are as follows:
Wherein FmaxAnd FminThe bound of zoom factor is respectively indicated, PD (g) is the group difference degree in g generation, and is planted
The expression of constellation variance degree carries out individuals all in population space to cluster obtained cluster number, when group difference degree is bigger,
Individual is distributed more uniform in population space, and it is bigger to acquire globally optimal solution possibility;τ1For the iteration threshold of setting, gmaxFor
Maximum number of iterations.
According to the present invention, further, the formula that adjust crossover probability CR can adaptively in the step 23 are as follows:
Wherein CRminAnd CRmaxIt is the bound of crossover probability, τ2It is the iteration threshold of setting.
According to the present invention, further, Chaos Search implements in the step 3 are as follows: modeling dimensional Logistic first
Map chaotic model, expression formula are as follows: Zt+1=μ Zt(1-Zt) formula is iterative formula in mathematical meaning, valueIt is the D dimensional vector generated at random, wherein μ is control parameter, and t indicates chaos iteration number;
Secondly, optimum individual and diversity factor center iteration more new formula in modeling population:
Wherein XiIndicate optimum individual or the diversity factor center of population,New individual after indicating Chaos Search, α table
Show chaos adjustment parameter, r is the random number between [0,1].
Beneficial effect to illustrate the invention, is further analyzed according to simulation result.
Fig. 2 describes the network safety situation value obtained based on algorithms of different.Figure it is seen that autoregression sliding is flat
Equal model (Auto-Regressive and Moving Average Model, abbreviation ARMA) mainly for it is random stable when
Between sequence, but because network attack randomness and complexity, network safety situation sequence be it is non-smoothly;Gray model
(Grey Model, abbreviation GM) is good for the time series forecasting effect of monotone variation, otherwise error is big;Least square support to
The supporting vector of amount machine (Least Squares Support Vector Machines, abbreviation LSSVM) becomes all data
Point loses the sparsity feature of SVM;Kmeans-RBF needs to preset hidden layer node, has ignored the spy of data itself
Point weakens the generalization ability of RBF, however, using true value as measurement standard, it is different degrees of compared to what the above method occurred
Error and defect, APDE-RBF Neural Network model predictive precision highest proposed by the present invention.
Fig. 3 shows the different errors comparison of algorithms of different, as can be seen from Figure 3 whether average relative error, square
Root error or relative mean square error, APDE-RBF neural network model are held at lesser error level, embody higher
Precision of prediction.
Fig. 4 shows the network safety situation value of different improvement DE algorithms in different time points.DE algorithm be fixed F and
CR easily falls into local optimum;Simplified differential evolution version (Simplified Differential Evolution
Version, abbreviation SDE) algorithm F use simple random number;Differential evolution algorithm based on all parameters and Mutation Strategy
(Differential evolution algorithm with ensemble of parameters and mutation
Strategies, abbreviation EPSDE) algorithm using Mutation Strategy pond and parameter pond random combine is iterated evolution;It is adaptive poor
Divide the F and CR of (Self-Adaptive Differential Evolution, abbreviation jDE) algorithm of evolving to rely on random number to differentiate
To obtain different results;Differential evolution (Evolution based on composite testing vector generation strategy and control parameter
With Composite Trial Vector Generation Strategies and Control Parameters, referred to as
CoDE) algorithm is to be iterated evolution using three kinds of different Mutation Strategies and parameter setting competition coupling.Although the above method
Mutation Strategy and parameter setting to DE algorithm carry out adaptive impovement, but are all that random number or dependence random number carry out mostly
Differentiate and choose, causes to evolve unstable.Further, it can be seen from the figure that using true value as measurement standard, compared to above-mentioned
Algorithm, APDE algorithm generally maintain lower absolute error, and reason is that APDE-RBF neural network model relies on kind
Constellation variance degree and iterative evolution degree adaptively adjust F and CR, so that population is evolved to beneficial direction, accelerate algorithm
Convergence rate.
Fig. 5 is the different errors comparison of algorithms of different, and whether average relative error, root mean square miss as can be seen from Figure 5
Difference or relative mean square error, APDE-RBF neural network model proposed by the present invention are held at lesser error level, body
Higher precision of prediction is showed.
The lifted embodiment of the present invention or embodiment have carried out further the object, technical solutions and advantages of the present invention
Detailed description, it should be understood that embodiment provided above or embodiment be only the preferred embodiment of the present invention and
, be not intended to limit the invention, all within the spirits and principles of the present invention it is made for the present invention it is any modification, equally replace
It changes, improve, should all be included in the protection scope of the present invention.
Claims (5)
1. a kind of network security state for the radial basis function APDE-RBF neural network for propagating differential evolution algorithm based on attractor
Gesture prediction technique, which is characterized in that method includes the following steps:
Step 1: propagating AP clustering algorithm using attraction and partition clustering is carried out to sample data, to obtain radial basis function
The center of RBF and the node in hidden layer of network;
Step 2: obtaining group difference degree using AP cluster, adaptively change the zoom factor and intersection of differential evolution DE algorithm
Probability optimizes the width and connection weight of RBF;
Step 3: in order to avoid falling into local optimum and jumping out Local Extremum, to the elite individual of every generation population and population
Diversity factor center carries out Chaos Search;
Step 4: determining that final RBF network model, input test data set export situation predicted value;
It is further included steps of in the step 2
Step 21: initialization is executed, process is as follows:
σi=σmin+rand(0,1)*(σmax-σmin)
wi=rand (0,1)
Wherein σiFor RBF neural basic function width, σmaxIndicate the distance of two farthest data points in all sample number strong points
Width, its calculation formula is:σminIndicate that two count recently in all sample number strong points
Strong point apart from width, its calculation formula is:ci、cjIndicate that any two are different
Hidden layer node, wiExpression hidden layer is to output layer connection weight, equally distributed random number between rand (0,1) expression (0,1);
Step 22: mutation process is executed, by g+1 for variation individual V in populationi(g+1) g is modeled as three in population
The function of body:
Vi(g+1)=Xr1(g)+F*(Xr2(g)-Xr3(g))
i≠r1≠r2≠r3
Wherein XiIt (g) is g for i-th of individual, i.e. X in populationr1(g)、Xr2(g) and Xr3(g) g is respectively indicated in population
R1, r2 and the r3 individuals, F is zoom factor;
Step 23: executing crossover process, generate i-th of jth reform individual u of g+1 generationij(g+1) formula are as follows:
Wherein vij(g+1) indicate that g carries out the individual after mutation operation, x for i-th of jth dimension individual of populationij(g) g generation is indicated
I-th of jth dimension individual of population, rand is equally distributed random number between (0,1), jrandIt is the random integers between [1, n], CR table
Show crossover probability;Above-mentioned formula meaning are as follows: correspond to ordinal number j when stochastic variable rand is less than element in crossover probability CR or individual
Equal to stochastic variable jrand, then using the element in variation individual as new individual, it is intended to a possibility that improving individual variation;It is no
Then, target individual x is still keptij(g) constant;
Step 24: selection course is executed, as follows:
Wherein UiIt (g+1) is candidate individual, XiIt (g) is corresponding individual, f () is the fitness function of individual, is used herein as square
Error is as fitness function.
2. the network security situation prediction method according to claim 1 based on APDE-RBF neural network, feature exist
In the step 1 further includes steps of
Step 11: calculating the similarity matrix S between input node using Euclidean distance are as follows: S (i, k)=- | | xi-xk||2,
Middle xiAnd xkIndicate RBF neural any two input node, S (i, k) indicates point xkAs point xiCluster centre it is similar
Degree, value are stored in similar matrix S;
Step 12: initializing Attraction Degree matrix R and degree of membership matrix A is R (i, k)=0, A (i, k)=0, wherein R (i, k) is indicated
Point xkIt is suitable as data point xiCluster centre degree, A (i, k) indicate point xiSelected element xkAs the suitable of its cluster centre
Conjunction degree;
Step 13: determining and be biased to parameterpkIndicate that each sample data point is selected as cluster centre
A possibility that, it is the value of element on similar matrix S diagonal line, k=1 ..., N, the quantity of N expression input node, median letter
Number indicates to take the numerical value for occuping intermediate in one group of numerical value;
Step 14: Attraction Degree matrix R and degree of membership matrix A are calculated according to following formula:
Wherein p (k) indicates data point xkAs the point of reference of cluster centre, R (k, k) indicates data point xkIt is suitable as oneself
The degree of cluster centre, A (k, k') indicate data point xkSelect data point xk'As the degree of its cluster centre, S (k, k') table
Registration strong point xkWith data point xk'Similarity degree;
Step 15: updating the formula of Attraction Degree matrix R and degree of membership matrix A are as follows:
R (i, k)=λ * R (i, k)old+(1-λ)*R(i,k)new
A (i, k)=λ * A (i, k)old+(1-λ)*A(i,k)new
When above-mentioned more new formula indicates each iteration, new Attraction Degree matrix R (i, k)newWith degree of membership matrix A (i, k)newDivide
Not with last Attraction Degree matrix R (i, k)oldWith degree of membership matrix A (i, k)oldIt is weighted update, obtains the secondary iteration
Attraction Degree matrix and degree of membership matrix, wherein λ indicates updating factor;
Step 16: if meeting one of the following conditions: the class center 1. selected keeps stablizing, and is 2. more than maximum number of iterations, then
Step 17 is gone to, step 14 is otherwise gone to;
Step 17, cluster result is exported.
3. the network security situation prediction method according to claim 1 based on APDE-RBF neural network, feature exist
In to the formula of zoom factor F progress dynamic adjustment in the step 22 are as follows:
Wherein FmaxAnd FminThe bound of zoom factor is respectively indicated, PD (g) is the group difference degree in g generation, and population is poor
Different degree expression carries out individuals all in population space to cluster obtained cluster number, when group difference degree is bigger, individual
It is distributed in population space more uniform, it is bigger to acquire globally optimal solution possibility;τ1For the iteration threshold of setting, gmaxFor maximum
The number of iterations.
4. the network security situation prediction method according to claim 1 based on APDE-RBF neural network, feature exist
In the formula that adjust crossover probability CR can adaptively in the step 23 are as follows:
Wherein CRminAnd CRmaxRespectively indicate the bound of crossover probability, τ2For the iteration threshold of setting, gmaxFor greatest iteration time
Number.
5. the network security situation prediction method according to claim 1 based on APDE-RBF neural network, feature exist
In Chaos Search in the step 3 specifically: modeling 1 dimensional Logistic Map chaotic model first, expression formula are as follows: Zt+1
=μ Zt(1-Zt), wherein ZtIt is a D dimensional vector, μ is control parameter, and t indicates chaos iteration number;Secondly, in modeling population
Optimum individual and diversity factor center iteration more new formula:
Wherein XiIndicate optimum individual or the diversity factor center of population,New individual after indicating Chaos Search, α indicate mixed
Ignorant adjustment parameter, r are the random numbers between [0,1].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610871705.XA CN106411896B (en) | 2016-09-30 | 2016-09-30 | Network security situation prediction method based on APDE-RBF neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610871705.XA CN106411896B (en) | 2016-09-30 | 2016-09-30 | Network security situation prediction method based on APDE-RBF neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106411896A CN106411896A (en) | 2017-02-15 |
CN106411896B true CN106411896B (en) | 2019-04-23 |
Family
ID=59229451
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610871705.XA Active CN106411896B (en) | 2016-09-30 | 2016-09-30 | Network security situation prediction method based on APDE-RBF neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106411896B (en) |
Families Citing this family (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107222333A (en) * | 2017-05-11 | 2017-09-29 | 中国民航大学 | A kind of network node safety situation evaluation method based on BP neural network |
CN107194524B (en) * | 2017-07-28 | 2020-05-22 | 合肥工业大学 | RBF neural network-based coal and gas outburst prediction method |
CN108009241B (en) * | 2017-11-30 | 2021-12-07 | 瞳见科技有限公司 | Industrial Internet of things security aggregation association method based on PSO parameter optimization |
CN108594793A (en) * | 2018-04-17 | 2018-09-28 | 电子科技大学 | A kind of improved RBF flight control systems fault diagnosis network training method |
CN108548846A (en) * | 2018-06-21 | 2018-09-18 | 电子科技大学 | Bearing device subsurface defect extracting method based on thermal image AP clusters |
CN108881250B (en) * | 2018-06-28 | 2020-07-07 | 广东电网有限责任公司 | Power communication network security situation prediction method, device, equipment and storage medium |
CN109040027B (en) * | 2018-07-12 | 2020-08-18 | 北京理工大学 | Active prediction method of network vulnerability node based on gray model |
CN109117872A (en) * | 2018-07-24 | 2019-01-01 | 贵州电网有限责任公司信息中心 | A kind of user power utilization behavior analysis method based on automatic Optimal Clustering |
CN109102062B (en) * | 2018-08-15 | 2020-03-13 | 桂林电子科技大学 | 3D NoC test planning method based on Petri network and chaotic differential firefly algorithm |
CN109543914B (en) * | 2018-11-29 | 2023-02-28 | 北京化工大学 | Method for predicting RBF building operation energy consumption based on AP clustering algorithm |
CN110362558B (en) * | 2019-06-12 | 2022-12-16 | 广东工业大学 | Energy consumption data cleaning method based on neighborhood propagation clustering |
CN110474904B (en) * | 2019-08-16 | 2022-04-12 | 武汉思普崚技术有限公司 | Situation awareness method and system for improving prediction |
CN111783988B (en) * | 2020-06-16 | 2024-03-26 | 长江大学 | Multi-target gray prediction evolution method and system for solving environmental economic scheduling problem |
CN113191917B (en) * | 2021-03-09 | 2023-04-07 | 中国大唐集团科学技术研究院有限公司 | Power plant industrial control system network security threat classification method based on radial basis function algorithm |
CN113141272B (en) * | 2021-04-20 | 2022-03-15 | 浙江大学 | Network security situation analysis method based on iteration optimization RBF neural network |
CN114155390B (en) * | 2021-12-02 | 2022-05-06 | 北京中科智易科技有限公司 | Clustering device for classifying metadata of classified objects |
CN116843368B (en) * | 2023-07-17 | 2024-01-26 | 杭州火奴数据科技有限公司 | Marketing data processing method based on ARMA model |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101975092A (en) * | 2010-11-05 | 2011-02-16 | 中北大学 | Real-time prediction method of mine gas concentration in short and medium terms based on radial basis function neural network integration |
CN104978857A (en) * | 2015-05-26 | 2015-10-14 | 重庆邮电大学 | Traffic state prediction method based on chaos theory and device thereof |
CN105678380A (en) * | 2016-01-08 | 2016-06-15 | 浙江工业大学 | Ecological niche and adaptive negative correlation learning-based evolutionary neural network integration method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9813343B2 (en) * | 2013-12-03 | 2017-11-07 | Akamai Technologies, Inc. | Virtual private network (VPN)-as-a-service with load-balanced tunnel endpoints |
-
2016
- 2016-09-30 CN CN201610871705.XA patent/CN106411896B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101975092A (en) * | 2010-11-05 | 2011-02-16 | 中北大学 | Real-time prediction method of mine gas concentration in short and medium terms based on radial basis function neural network integration |
CN104978857A (en) * | 2015-05-26 | 2015-10-14 | 重庆邮电大学 | Traffic state prediction method based on chaos theory and device thereof |
CN105678380A (en) * | 2016-01-08 | 2016-06-15 | 浙江工业大学 | Ecological niche and adaptive negative correlation learning-based evolutionary neural network integration method |
Also Published As
Publication number | Publication date |
---|---|
CN106411896A (en) | 2017-02-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106411896B (en) | Network security situation prediction method based on APDE-RBF neural network | |
Ergezer et al. | Oppositional biogeography-based optimization | |
CN106327240A (en) | Recommendation method and recommendation system based on GRU neural network | |
CN106228245B (en) | Infer the knowledge base complementing method with tensor neural network based on variation | |
CN111416797B (en) | Intrusion detection method for optimizing regularization extreme learning machine by improving longicorn herd algorithm | |
CN109523021A (en) | A kind of dynamic network Structure Prediction Methods based on long memory network in short-term | |
CN113190688A (en) | Complex network link prediction method and system based on logical reasoning and graph convolution | |
CN108322541A (en) | A kind of adaptive Distributed architecture | |
Donate et al. | Evolutionary optimization of sparsely connected and time-lagged neural networks for time series forecasting | |
CN110619082A (en) | Project recommendation method based on repeated search mechanism | |
CN110751183A (en) | Image data classification model generation method, image data classification method and device | |
Ma et al. | Discrete and continuous optimization based on hierarchical artificial bee colony optimizer | |
Rad et al. | GP-RVM: Genetic programing-based symbolic regression using relevance vector machine | |
CN115169555A (en) | Edge attack network disruption method based on deep reinforcement learning | |
CN110298506A (en) | A kind of urban construction horizontal forecast system | |
CN101893852A (en) | Multi-target modeling method for complex industrial process | |
CN107704944B (en) | Construction method of stock market fluctuation interval prediction model based on information theory learning | |
CN111412795B (en) | Test point setting scheme generation method and device | |
CN117217287A (en) | Training method of multi-element sub-strategy generation model for hierarchical reinforcement learning | |
Singhal et al. | Greedy Shapley Client Selection for Communication-Efficient Federated Learning | |
CN112132259B (en) | Neural network model input parameter dimension reduction method and computer readable storage medium | |
CN109241134A (en) | A kind of grouping of commodities mode multiple target method for digging based on agent model | |
CN107480768A (en) | Bayesian network structure adaptive learning method and device, storage device and terminal device | |
Drugan et al. | Generalized adaptive pursuit algorithm for genetic Pareto local search algorithms | |
Putri et al. | A deep-genetic algorithm (deep-GA) approach for high-dimensional nonlinear parabolic partial differential equations |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |