CN101655911A - Mode identification method based on immune antibody network - Google Patents
Mode identification method based on immune antibody network Download PDFInfo
- Publication number
- CN101655911A CN101655911A CN200910164886A CN200910164886A CN101655911A CN 101655911 A CN101655911 A CN 101655911A CN 200910164886 A CN200910164886 A CN 200910164886A CN 200910164886 A CN200910164886 A CN 200910164886A CN 101655911 A CN101655911 A CN 101655911A
- Authority
- CN
- China
- Prior art keywords
- antibody
- network
- immune
- antigen
- training
- 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
Images
Abstract
The invention relates to a mode identification method based on an immune antibody network, which is used for mode identification. The technical scheme of the invention comprises the following steps: generating an original antibody in the immune antibody network by extracting all classes of training samples with a certain number stochastically and completing the initialization of the immune antibody network; taking all the training samples as an input antigen, training the immune antibody network according to an antibody formation algorithm and extracting the mode characteristics of all classesof training samples effectively; and training the immune antibody network repeatedly, stopping training when the continuous training results of two times are accordant, preserving the obtained immuneantibody network and carrying out the mode identification. The invention has simple and convenient calculation and very high correct identification rate without manually setting parameters and thresholds and is suitable for the computer application fields of character identification, fault diagnosis and the like.
Description
Technical field
The present invention relates to that a kind of energy fast, is accurately classified to things or the method for pattern-recognition, belong to the recognition technology field.
Background technology
Pattern-recognition has obtained paying attention to widely in a lot of Science and Technology fields, has promoted the development of artificial intelligence system, has enlarged the possibility of computer utility.Generally calling pattern by concrete indivedual things being carried out the concrete time that observation station gets and the information of space distribution, the classification under the pattern or similar in pattern totally call mode class.
Artificial immune system is a kind ofly to inspire and the calculating normal form of coming based on theoretical biology, and it has used for reference the solution that some functions of immune system, principle and model are used for challenge.Artificial immune system has obtained many researchs and application at aspects such as information security, pattern-recognition, data mining and fault diagnosises, has obtained effect preferably.But, at present Artificial Immune Algorithm exist mostly the algorithm complexity, manually be provided with parameter and threshold values more, problem such as need constantly adjust.
Summary of the invention
The objective of the invention is to overcome the defective of prior art and the mode identification method based on immune antibody network that a kind of algorithm is easy, accuracy rate is high is provided.
Problem of the present invention realizes with following technical proposals:
A kind of mode identification method based on immune antibody network, before immune antibody network carries out pattern-recognition, need immune antibody network to be trained according to user's training sample data, immune antibody network after the training promptly can be applicable to solve pattern recognition problem, and the training process of immune antibody network is:
Randomly draw all kinds of training samples of some, generate the initial antibodies in the immune antibody network, finish the initialization of immune antibody network;
In order or at random with all training samples as input antigen, be input to immune antibody network, the antibody in the immune antibody network carries out learning and memory according to the antibody generating algorithm to input antigen, extracts the pattern feature of all kinds of training sample data;
The repetition training immune antibody network, when the result of the double training of immune antibody network is consistent (twice continuous training, the number of each antibody-like does not change in the immune antibody network), the immune antibody network training stops.
Above-mentioned mode identification method based on immune antibody network, the initialized method of described immune antibody network is:
The given training sample of user as antigen, is represented j antigen A g with the real number vector of n dimension
jBe expressed as follows:
Ag
j＝(Ag
j1，Ag
j2，...，Ag
jn)
Correspondingly, the antibody in the immune antibody network is represented i antibody A b in the immune antibody network with the real number vector of (n+2) dimension
iBe expressed as follows:
Ab
i＝(T
i，C
i，Ab
i1，Ab
i2，...，Ab
in)
Wherein, T
iBe antibody A b
iAffiliated classification information, i.e. antibody A b
iThe classification of identification antigen; C
iBe antibody A b
iConcentration, expression antibody A b
iThe quantity of identification antigen how much, Ab
1～Ab
nRepresent antibody A b respectively
iEach dimension attribute information, the pattern feature of expressing antibodies;
The classification number of antigen (both training samples) is m, every class antigen is randomly drawed k, constitute the initial antibodies in the immune antibody network, concrete grammar is: the classification of the antigen that each is randomly drawed and attribute information, determined the classification information and the attribute information of an initial antibodies, and the concentration value of all initial antibodies is zero.So initial immune antibody network includes m * k initial antibodies altogether, wherein m is the classification number (identical with antigen) of antibody, and k is the initial number of every antibody-like.The classification information T of each initial antibodies
i, consistent with corresponding classification that randomly draw, the antigen that generates this initial antibodies, its attribute information is also consistent with corresponding antigen that randomly draw, that generate this initial antibodies, the concentration initial value C of all initial antibodies
iAll get zero.Described numerical value k, determine according to the complexity of classification and identification (difference between the Different Individual of same classification between the fluctuation of eigenwert and the different classes of sample characteristics): difficulty is high more, and the numerical value value is big more, but the general k value is 10 to 30.
Above-mentioned mode identification method based on immune antibody network, the training method of described immune antibody network is: in order or training sample of picked at random be input to immune antibody network as antigen, antibody in the immune antibody network, according to the antibody generating algorithm, antigen to input carries out learning and memory, finish until all training samples (antigen) input, this process is called has finished once training.Immune antibody network is repeatedly trained, when double training result is consistent, twice promptly continuous training, when the number of each antibody-like no longer changed in the immune antibody network, then training stopped.
Above-mentioned mode identification method based on immune antibody network, described antibody generating algorithm is:
Definition:
1. similarity: the similar or close degree between two samples is described, often with the measurement standard of certain distance as similarity, the moral distance as Europe commonly used in several;
2. d
Ij: antibody A b in the immune antibody network
iWith antigen A g
jBetween similarity;
3. s
Ij: antibody A b in the immune antibody network
iWith antibody A b
jBetween similarity;
4. D: the similarity matrix of antibody in input antigen and the immune antibody network, its data element is d
Ij
5. S: the similarity matrix in the immune antibody network between the antibody, its data element are s
Ij
6. Ab
r: in the immune antibody network with input antigen A g
jThe antibody that similarity is the highest (if exist a plurality of, can choose one of them wantonly) promptly satisfies d
Rj=min (D), this antibody is called identification antibody, is designated as Ab
r
7. Ab
b: the identification antibody A b in the immune antibody network
rIf, its classification and input antigen A g
jUnanimity, then this antibody is called best identified antibody, is designated as Ab
b
8. antibody is evolved: the antibody A b in the immune antibody network
iTo input antigen A g
jFinish learning and memory, its computing formula is:
C
i′＝C
i+1
Ab
im′＝(C
i×Ab
im+Ag
jm)/C
i′，m＝1，2，...，n
Ab
i＝(T
i，C
i′，Ab
i1′，Ab
i2′，...，Ab
in′)
9. antibody merges: two similar antibody A b in the immune antibody network
iWith Ab
jMerging becomes a new antibodies Ab
k, original antibody (Ab in the immune antibody network
iAnd Ab
j) from immune antibody network, delete new antibodies Ab
kValue be calculated as follows:
C
k＝C
i+C
j
Ab
km＝(C
i×Ab
im+C
j×Ab
jm)/C
k，m＝1，2，...，n
Ab
k＝(T
i，C
k，Ab
k1，Ab
k2，...，Ab
m)
10. antibody new life: according to current input antigen A g
j, in immune antibody network, insert a new antibody A b
i, new antibodies Ab
iValue be:
Ab
i＝(T
i，1，Ag
j1，Ag
j2，...，Ag
jn)
Wherein, new antibodies Ab
iClassification information T
iBy input antigen A g
jClassification determine.
According to above definition, described antibody generating algorithm is: at first calculate input antigen A g
jAnd the similarity in the immune antibody network between all antibody obtains similarity matrix D; According to D, in immune antibody network, choose identification antibody A b
rIf identification antibody A b
rClassification and input antigen A g
jClassification is inconsistent, immune antibody network generation antibody new life then, and promptly immune antibody network is at input antigen A g
jProduce new antibodies Ab
i, algorithm finishes; If identification antibody A b
rClassification and input antigen A g
jThe classification unanimity then obtains best identified antibody A b
b, i.e. Ab
b=Ab
rCalculate Ab
bThe similarity of all antibody identical with classification in the immune antibody network obtains antibody similarity matrix S.According to S, select and best identified antibody A b
bThe same antibody-like that similarity is the highest, this antibody is designated as Ab
jIf, best identified antibody A b
bThe similar antibody A b the highest with its similarity
jSimilarity s
Ij, less than best identified antibody A b
bWith input antigen A g
jSimilarity d
Bj, promptly satisfy s
Ij＜d
Bj, then according to input antigen A g
j, immune antibody network generation antibody new life produces new antibodies Ab
i, while best identified antibody A b
bThe similar antibody A b the highest with similarity
jExecution antibody merges, and algorithm finishes; Otherwise, best identified antibody A b
bAccording to input antigen A g
jGeneration antibody is evolved, best identified antibody A b
bTo input antigen A g
jFinish learning and memory, algorithm finishes.
Antibody has proposed to have the immune antibody network of self-organization, self study and memory capability to the efficient identification and the memory mechanism of antigen in the bionical Immune System of the present invention, and has designed the antibody generating algorithm, can be applied to pattern-recognition.The inventive method is calculated easy, need not parameter and threshold values manually are set, and has very high correct recognition rata, is suitable for computer application fields such as Character Font Recognition, fault diagnosis.
Description of drawings
The invention will be further described below in conjunction with accompanying drawing.
Fig. 1 is the immune antibody network synoptic diagram.
Among the figure and literary composition in used symbol be: a, b, c, antibody, Ag
j, antigen or training sample, Ag
J1, Ag
J2..., Ag
Jn, Ag
jEach dimension attribute information, Ab
i, antibody, T
i, antibody A b
iAffiliated classification, C
i, antibody A b
iConcentration, Ab
I1～Ab
In, antibody A b
iAttribute information, d
Ij, antibody A b
iWith antigen A g
jBetween similarity, s
Ij, antibody A b in the immune antibody network
iWith antibody A b
jBetween similarity, D, Ag-Ab similarity matrix, S, antibody-antibody similarity matrix, Ab
r, identification antibody, Ab
b, best identified antibody.
Embodiment
Antibody is to the efficient identification and the memory mechanism of antigen in the bionical Immune System, and the present invention has designed a kind of immune antibody network and antibody generating algorithm, is applied to area of pattern recognition.
Below the present invention is done further explanation.
1) immune antibody network
Immune antibody network is to be the figure of node formation with some different classes of antibody.Wherein, the antibody that classification is identical links together (antibody that links together has identical type), forms connected subgraph, and each antibody (node) has certain concentration, concentration is as the weights of this node, and immune antibody network is the incomplete connected graph of a node cum rights.As Fig. 1, three kinds of different classes of antibody of a, b and c are arranged in the immune antibody network, 2,3 and 4 antibody are arranged respectively successively, the antibody of identical category links together.The various to external world antigens of each antibody among the figure all have the function of identification, learning and memory, and all antibody in the immune antibody network interacts, and finishes identification, the learning and memory of various antigens to external world jointly.
2) antigen, antibody coding
In the immune antibody network, antigen and antibody coding adopt the real coding mode.
Antigen encoding: as antigen, an antigen (training sample) is with the real number vector Ag of a n dimension the given training sample data of user
jBe expressed as follows:
Ag
j＝(Ag
j1，Ag
j2，...，Ag
jn) (1)
Antibody coding: the coding of each antibody comprises two partial contents in the immune antibody network, is respectively antibody essential information and antibody attribute information.Wherein, the antibody essential information comprises the classification information of this antibody and the concentration information of this antibody, and the classification of antibody is promptly represented the classified information of this antibody, i.e. the schema category of antigen that this antibody is discerned; The concentration of antibody has then been represented the ability of such antigen of this antibody recognition, after the immune antibody network training finishes, has promptly represented the total quantity according to such antigen of this antibody recognition substantially, has characterized the power of this antibody recognition antigenic capacity.The antibody attribute information is then represented the data characteristics of antigen that antibody is discerned, i.e. the value of antibody each dimension in n dimension form space.Antibody A b
iCan be expressed as follows:
Ab
i＝(T
i，C
i，Ab
i1，Ab
i2，...，Ab
in) (2)
In formula (2) formula, T
iBe antibody A b
iAffiliated classification, i.e. antibody A b
iThe classification of identification antigen; C
iBe antibody A b
iConcentration, expression antibody A b
iThe quantity of identification antigen how much.Ab
1～Ab
nRepresent antibody A b respectively
iEach dimension attribute information, the pattern feature of expressing antibodies.
3) immune antibody network initialization
At first training sample is analyzed, determined other number of antibody class in the immune antibody network according to the number of training sample classification, other number of antibody class is designated as m; According to the complexity of classification and identification, be the fluctuation of eigenwert between the Different Individual of same classification and the difference between the different classes of sample characteristics, determine the number of every class initial antibodies: difficulty is high more, and the initial antibodies number value of setting is big more.The initial antibodies number of each classification is designated as k, and generally speaking, but the k value is 10 to 30.From the training sample of every class, extract k sample randomly, the attribute information of k the sample that extracts is respectively as the attribute information of such k initial antibodies in the immune antibody network, the classification information of antibody is determined according to the classification information of the training sample corresponding with it simultaneously, and the concentration initial value of all initial antibodies gets zero.
After the immune antibody network initialization was finished, immune antibody network was made of the different classes of initial antibodies of some (m * k), and the antibody of identical category links together, and at this moment, all antibody all is initial antibodies in the immune antibody network.
4) immune antibody network training
The training of immune antibody network is meant antibody in the immune antibody network to the process of learning and Memory of input antigen, and its basic process can be described below:
1. import antigen, promptly in order or training sample of picked at random be input to immune antibody network as antigen;
2. according to the antibody generating algorithm, the antibody in the immune antibody network is finished the learning and memory to input antigen;
3. repeat the above-mentioned 1-2 step, finish until all antigen inputs, this process is called has finished once training;
4. immune antibody network is repeatedly trained.When if the double training result of immune antibody network is consistent (twice continuous training, the number of each antibody-like does not change in the immune antibody network), the immune antibody network training stops, otherwise, train next time.
After above-mentioned method training end, the antibody in the immune antibody network can be described the characteristic information of all kinds of training sample data exactly through continuous learning and memory, becomes ripe memory antibody.
5) antibody generating algorithm
5.1) definition
1. similarity: the similar or close degree between two samples is described, often with the measurement standard of certain distance as similarity, the moral distance as Europe commonly used in several;
2. d
Ij: antibody A b in the immune antibody network
iWith antigen A g
jBetween similarity;
3. s
Ij: antibody A b in the immune antibody network
iWith antibody A b
jBetween similarity;
4. D: the similarity matrix of antibody in input antigen and the immune antibody network, its data element is d
Ij
5. S: the similarity matrix in the immune antibody network between the antibody, its data element are s
Ij
6. Ab
r: in the immune antibody network with input antigen A g
jThe antibody that similarity is the highest (if exist a plurality of, can choose one of them wantonly) promptly satisfies d
Rj=min (D), this antibody is called identification antibody, is designated as Ab
r
7. Ab
b: the identification antibody A b in the immune antibody network
rIf, its classification and input antigen A g
jUnanimity, then this antibody is called best identified antibody, is designated as Ab
b
8. antibody is evolved: the antibody A b in the immune antibody network
iTo input antigen A g
jFinish learning and memory, its computing formula is:
C
i′＝C
i+1
Ab
im′＝(C
i×Ab
im+Ag
jm)/C
i′，m＝1，2，...，n
Ab
i＝(T
i，C
i′，Ab
i1′，Ab
i2′，...，Ab
in′)
9. antibody merges: two similar antibody A b in the immune antibody network
iWith Ab
jMerging becomes a new antibodies Ab
k, original antibody (Ab in the immune antibody network
iAnd Ab
j) from immune antibody network, delete new antibodies Ab
kValue be calculated as follows:
C
k＝C
i+C
j
Ab
km＝(C
i×Ab
im+C
j×Ab
jm)/C
k，m＝1，2，...，n
Ab
k＝(T
i，C
k，Ab
k1，Ab
k2，...，Ab
in)
10. antibody new life: according to current input antigen A g
j, in immune antibody network, insert a new antibody A b
i, new antibodies Ab
iValue be:
Ab
i＝(T
i，1，Ag
j1，Ag
j2，...，Ag
jn)
Wherein, new antibodies Ab
iClassification information T
iBy input antigen A g
jClassification determine.
5.2) algorithm steps
The antibody generating algorithm has determined that antibody is to the learning and memory process of input antigen in the immune antibody network.According to different situations, immune antibody network generation antibody is evolved, antibody merges and antibody new life.In the algorithm implementation, in the immune antibody network each antibody-like number, the concentration and the attribute information thereof of each antibody regulate automatically according to the antibody generating algorithm, need not to set in advance parameter and threshold values.The concrete steps of antibody generating algorithm are described below:
Step1. calculate input antigen A g
jAnd the similarity in the immune antibody network between all antibody obtains similarity matrix D;
Step2. according to similarity matrix D, in immune antibody network, choose identification antibody A b
r
If Step3. discern antibody A b
rClassification and input antigen A g
jClassification is inconsistent, immune antibody network generation antibody new life then, and promptly immune antibody network is at input antigen A g
jProduce new antibodies Ab
i, algorithm finishes;
Step4. discern antibody A b
rClassification and input antigen A g
jThe classification unanimity then obtains best identified antibody A b
b, i.e. Ab
b=Ab
r
Step5. calculate Ab
bThe similarity of all antibody identical with classification in the immune antibody network obtains antibody similarity matrix S.According to antibody similarity matrix S, select and best identified antibody A b
bThe same antibody-like that similarity is the highest, this antibody is designated as Ab
j
If best identified antibody A b Step6.
bThe similar antibody A b the highest with its similarity
jSimilarity s
Ij, less than best identified antibody A b
bWith input antigen A g
jSimilarity d
Bj, promptly satisfy s
Ij＜d
Bj, then according to input antigen A g
j, immune antibody network generation antibody new life produces new antibodies Ab
i, while best identified antibody A b
bThe similar antibody A b the highest with similarity
jExecution antibody merges, and algorithm finishes;
Step7. best identified antibody A b
bAccording to input antigen A g
jGeneration antibody is evolved, best identified antibody A b
bTo input antigen A g
jFinish learning and memory, algorithm finishes.
The invention will be further described for example below.
In experimentation, according to every order that is recorded in the appearance of Letter Recognition data centralization, preceding 700 records that extract each letter respectively are as the experimental data collection.The experimental data collection is divided into training sample data and two parts of test sample book data.The training sample data concentrate preceding 630 records of each class-letter to form (according to its order) by experimental data, comprise 16380 records altogether; All the other experimental datas are concentrated 70 records of every class-letter, amount to 1820 records as the test sample book data.
Set the antibody initial number k=25 of immune antibody network, import the concentrated data of training sample one by one, immune antibody network is trained as input antigen.After training finishes, use the immune antibody network that obtains, all test sample books are carried out letter identification according to the most contiguous criterion.Experimental result shows, the correct recognition rata of the inventive method is 95.54%, on same data set, document 1[P.W.Frey and D.J.Slate.Letter Recognition Using Holland-styleAdaptive Classifiers.Machine Learning.1991,6 (2): 161～182.] use Holland StyleAdaptive Classifiers method and analyze, its highest correct recognition rata is a little more than 80%; Document 2[opens happy cutting edge of a knife or a sword, Yu Hua, and Xia Shengping waits .RSOM algorithm and applied research thereof. Fudan Journal (natural science edition), 2004,43 (5): 704～709] based on two kinds of RSOM algorithms of self-organized mapping network, its correct recognition rata is about 90%.Immune antibody network of the present invention demonstrates optimum pattern classification ability, and its correct recognition rata has significant raising than other two kinds of methods.
Claims (5)
1, a kind of mode identification method based on immune antibody network is characterized in that, to randomly draw all kinds of training samples of some, generates the initial antibodies of respective classes in the immune antibody network, finishes the initialization of immune antibody network; With all training samples as the input antigen, use the antibody generating algorithm, the training immune antibody network, antibody in the immune antibody network effectively extracts the pattern feature of all kinds of training samples, when double training result is consistent, training stops and preserves the immune antibody network that trains, and the immune antibody network after the training is used for pattern-recognition.
According to the described mode identification method of claim 1, it is characterized in that 2, the initialized method of described immune antibody network is based on immune antibody network:
The given training sample of user as antigen, with the real number vector representation of n dimension, j antigen A g
jBe expressed as:
Ag
j＝(Ag
j1，Ag
j2，...，Ag
jn) (1)
Corresponding with it, the antibody in the immune antibody network is with the real number vector representation of one (n+2) dimension, I antibody A b in the immune antibody network
iBe expressed as:
Ab
i＝(T
i，C
i，Ab
i1，Ab
i2，....，Ab
in)?(2)
Wherein, T
iBe antibody A b
iAffiliated classification information, i.e. antibody A b
iThe classification of the antigen that can discern; C
iBe antibody A b
iConcentration, expression antibody A b
iWhat of identification antigen quantity, Ab
i～Ab
nRepresent antibody A b respectively
iEach dimension attribute information, the pattern feature of expressing antibodies;
In the immune antibody network after the initialization, comprise m * k antibody altogether, the antibody in the immune antibody network is defined as initial antibodies at this moment, and wherein m is the classification number of initial antibodies, and k is the number of every class initial antibodies, the classification information T of each initial antibodies
i, determine that according to the classification of the training sample of randomly drawing of correspondence its attribute information is determined by the training sample of randomly drawing of correspondence that also the concentration value of all initial antibodies all gets zero.
3, according to claim 1 or 2 described mode identification methods based on immune antibody network, it is characterized in that, the method of described immune antibody network training is: in order or training sample of picked at random as input antigen, be input to immune antibody network, antibody in the immune antibody network is according to the antibody generating algorithm, finish learning and memory, finish, then be called and finished once training until all antigen inputs to input antigen; The repetition training immune antibody network, if double training result unanimity, twice promptly continuous training, the number of each antibody-like does not change in the immune antibody network, and the immune antibody network training finishes.
According to the described mode identification method of claim 3, it is characterized in that 4, described antibody generating algorithm is based on immune antibody network:
Definition:
1. similarity: describe the similar or close degree between two samples, often with the measurement standard of certain distance, as the distance of Euclid commonly used as similarity;
2. d
Ij: antibody A b in the immune antibody network
iWith antigen A g
iBetween similarity;
3. s
Ij: antibody A b in the immune antibody network
iWith antibody A b
jBetween similarity;
4. D: the similarity matrix of antibody in input antigen and the immune antibody network, its data element is d
Ij
5. S: the similarity matrix in the immune antibody network between the antibody, its data element are s
Ij
6. Ab
r: in the immune antibody network with input antigen A g
jThe antibody that similarity is the highest (if exist a plurality of, can choose one of them wantonly) promptly satisfies d
Rj=min (D), this antibody is called identification antibody, is designated as Ab
r
7. Ab
b: the identification antibody A b in the immune antibody network
rIf, its classification and input antigen A g
jUnanimity, then this antibody is called best identified antibody, is designated as Ab
b
8. antibody is evolved: the antibody A b in the immune antibody network
iTo input antigen A g
jFinish learning and memory, its computing formula is:
C
i′＝C
i+1
Ab
im′＝(C
i×Ab
im+Ag
jm)/C
i′，m＝1，2，...，n
Ab
i＝(T
i，C
i′，Ab
i1′，Ab
i2′，...，Ab
in′)
9. antibody merges: two similar antibody A b in the immune antibody network
iWith Ab
jMerging becomes a new antibodies Ab
k, original antibody (Ab in the immune antibody network
iAnd Ab
j) from immune antibody network, delete new antibodies Ab
kValue be calculated as follows:
C
k＝C
i+C
j
Ab
km＝(C
i×Ab
im+C
j×Ab
jm)/C
k，m＝1，2，...，n
Ab
k＝(T
i，C
k，Ab
k1，Ab
k2，...，Ab
in)
10. antibody new life: according to current input antigen A g
j, in immune antibody network, insert a new antibody A b
i, new antibodies Ab
iValue be:
Ab
i＝(T
i，1，Ag
j1，Ag
j2，...，Ag
jn)
Wherein, new antibodies Ab
iClassification information T
iBy input antigen A g
jClassification determine.
The antibody generating algorithm has determined that antibody is to the continuous learning and memory process of input antigen in the immune antibody network, and according to different situations, the antibody generation antibody in the immune antibody network is evolved, antibody merges and antibody new life; In the algorithm implementation, in the immune antibody network each antibody-like number, the concentration and the attribute information thereof of each antibody regulate automatically according to the antibody generating algorithm, need not to set in advance parameter and threshold values; The concrete steps of antibody generating algorithm are described below:
Step1. calculate input antigen A g
jAnd the similarity in the immune antibody network between all antibody obtains similarity matrix D;
Step2. according to similarity matrix D, in immune antibody network, choose identification antibody A b
r
If Step3. discern antibody A b
rClassification and input antigen A g
jClassification is inconsistent, immune antibody network generation antibody new life then, and promptly immune antibody network is at input antigen A g
jProduce new antibodies Ab
i, algorithm finishes;
Step4. discern antibody A b
rClassification and input antigen A g
jThe classification unanimity, then obtain best identified antibody A b
b, i.e. Ab
b=Ab
r
Step5. calculate Ab
bThe similarity of all antibody identical with classification in the immune antibody network obtains similarity matrix S; According to S, select and best identified antibody A b
bThe same antibody-like that similarity is the highest, this antibody is designated as Ab
j
If best identified antibody A b Step6.
bThe similar antibody A b the highest with its similarity
jSimilarity s
Ij, less than best identified antibody A b
bWith input antigen A g
jSimilarity d
Bj, promptly satisfy s
Ij＜d
Bj, then according to input antigen A g
j, immune antibody network generation antibody new life produces new antibodies Ab
i, while best identified antibody A b
bThe similar antibody A b the highest with similarity
jExecution antibody merges, and algorithm finishes;
Step7. best identified antibody A b
bAccording to input antigen A g
jGeneration antibody is evolved, best identified antibody A b
bTo input antigen A g
jFinish learning and memory, algorithm finishes.
5, according to the described mode identification method of claim 2, it is characterized in that based on immune antibody network, described all kinds of initial antibodies number k, according to the complexity of classification problem, value is 10 to 30.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101648862A CN101655911B (en) | 2009-05-11 | 2009-08-07 | Mode identification method based on immune antibody network |
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910074328.7 | 2009-05-11 | ||
| CNA2009100743287A CN101561880A (en) | 2009-05-11 | 2009-05-11 | Pattern recognition method based on immune antibody network |
| CN2009101648862A CN101655911B (en) | 2009-05-11 | 2009-08-07 | Mode identification method based on immune antibody network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101655911A true CN101655911A (en) | 2010-02-24 |
| CN101655911B CN101655911B (en) | 2012-05-30 |
Family
ID=41220676
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2009100743287A Pending CN101561880A (en) | 2009-05-11 | 2009-05-11 | Pattern recognition method based on immune antibody network |
| CN2009101648862A Expired - Fee Related CN101655911B (en) | 2009-05-11 | 2009-08-07 | Mode identification method based on immune antibody network |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2009100743287A Pending CN101561880A (en) | 2009-05-11 | 2009-05-11 | Pattern recognition method based on immune antibody network |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN101561880A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279944A (en) * | 2013-04-22 | 2013-09-04 | 哈尔滨工程大学 | Image division method based on biogeography optimization |
| CN103745005A (en) * | 2014-01-24 | 2014-04-23 | 清华大学 | Dynamic artificial immunity fault diagnosis method based on vaccine transplantation |
| CN105160291A (en) * | 2015-07-02 | 2015-12-16 | 上海闻泰电子科技有限公司 | Age estimation system based on artificial immunization identification system and method |
| CN109376779A (en) * | 2018-10-19 | 2019-02-22 | 西安交通大学 | A kind of complex electromechanical systems military service mode automatic identifying method |
| CN110211465A (en) * | 2019-06-24 | 2019-09-06 | 漳州卫生职业学院 | A kind of immune antiboidy intelligent teaching device |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102436252B (en) * | 2010-09-29 | 2014-01-01 | 清华大学 | Process industry fault diagnosis method and system based on immune hazard theory |
| CN115296856B (en) * | 2022-07-12 | 2024-04-19 | 四川大学 | ResNet-AIS-based evolution learning method for encrypted traffic network threat detector |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100595780C (en) * | 2007-12-13 | 2010-03-24 | 中国科学院合肥物质科学研究院 | Handwriting digital automatic identification method based on module neural network SN9701 rectangular array |
-
2009
- 2009-05-11 CN CNA2009100743287A patent/CN101561880A/en active Pending
- 2009-08-07 CN CN2009101648862A patent/CN101655911B/en not_active Expired - Fee Related
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279944A (en) * | 2013-04-22 | 2013-09-04 | 哈尔滨工程大学 | Image division method based on biogeography optimization |
| CN103745005A (en) * | 2014-01-24 | 2014-04-23 | 清华大学 | Dynamic artificial immunity fault diagnosis method based on vaccine transplantation |
| CN105160291A (en) * | 2015-07-02 | 2015-12-16 | 上海闻泰电子科技有限公司 | Age estimation system based on artificial immunization identification system and method |
| CN109376779A (en) * | 2018-10-19 | 2019-02-22 | 西安交通大学 | A kind of complex electromechanical systems military service mode automatic identifying method |
| CN109376779B (en) * | 2018-10-19 | 2020-10-27 | 西安交通大学 | Complex electromechanical system service mode automatic identification method |
| CN110211465A (en) * | 2019-06-24 | 2019-09-06 | 漳州卫生职业学院 | A kind of immune antiboidy intelligent teaching device |
| CN110211465B (en) * | 2019-06-24 | 2024-02-06 | 漳州卫生职业学院 | Immune antibody teaching intelligent device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101655911B (en) | 2012-05-30 |
| CN101561880A (en) | 2009-10-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101655911B (en) | Mode identification method based on immune antibody network | |
| CN103559175B (en) | A kind of Spam Filtering System based on cluster and method | |
| CN110598800A (en) | Garbage classification and identification method based on artificial intelligence | |
| CN109977416A (en) | A kind of multi-level natural language anti-spam text method and system | |
| CN107563439A (en) | A kind of model for identifying cleaning food materials picture and identification food materials class method for distinguishing | |
| CN102289522B (en) | Method of intelligently classifying texts | |
| CN106776538A (en) | The information extracting method of enterprise's noncanonical format document | |
| CN107992764B (en) | Sensitive webpage identification and detection method and device | |
| CN101540017B (en) | Feature extracting method based on byte level n-gram and twit filter | |
| CN103984943A (en) | Scene text identification method based on Bayesian probability frame | |
| CN105139041A (en) | Method and device for recognizing languages based on image | |
| CN109934260A (en) | Image, text and data fusion sensibility classification method and device based on random forest | |
| CN105574063A (en) | Image retrieval method based on visual saliency | |
| CN110415071B (en) | Automobile competitive product comparison method based on viewpoint mining analysis | |
| CN108764302B (en) | Bill image classification method based on color features and bag-of-words features | |
| CN101604322A (en) | A kind of decision level text automatic classified fusion method | |
| CN105930792A (en) | Human action classification method based on video local feature dictionary | |
| CN102156871A (en) | Image classification method based on category correlated codebook and classifier voting strategy | |
| CN104951554A (en) | Method for matching landscape with verses according with artistic conception of landscape | |
| CN105117740A (en) | Font identification method and device | |
| Thokchom et al. | Recognition of Handwritten Character of Manipuri Script. | |
| CN105912525A (en) | Sentiment classification method for semi-supervised learning based on theme characteristics | |
| CN106203510A (en) | A kind of based on morphological feature with the hyperspectral image classification method of dictionary learning | |
| CN109871449A (en) | A kind of zero sample learning method end to end based on semantic description | |
| CN110263164A (en) | A kind of Sentiment orientation analysis method based on Model Fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| 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: 20120530 Termination date: 20170807 |