CN109901064A - Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ - Google Patents

Abstract

The invention discloses the Fault Diagnosis for HV Circuit Breakers methods based on ICA-LVQ, specifically implement according to the following steps: step 1: choosing typical data sample, be divided into training set and test set according to the ratio of 3:1；Step 2: extracting the input feature value through the resulting training sample of step 1, dimension-reduction treatment is carried out using LDA algorithm, obtains a new input feature value；Step 3: the input feature value that step 2 is obtained learns output failure modes as a result, setting up the Fault Diagnosis for HV Circuit Breakers model based on ICA-LVQ with this as the input of building ICA-LVQ neural network, by training；Step 4: the Fault Diagnosis for HV Circuit Breakers model that step 3 obtains classifying to the test set sample in step 1, counts its classification accuracy.Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ of the invention can accurately realize Fault Diagnosis for HV Circuit Breakers.

Description

Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ

Technical field

The invention belongs to high-voltage circuitbreaker On-line Fault monitoring technical fields, and in particular to a kind of height based on ICA-LVQ Voltage breaker method for diagnosing faults.

Background technique

Breaker is the important equipment in electric system.What the reliability of breaker performance was directly related to electric system can By operation, and the reliability of breaker depends greatly on the reliability of its operating mechanism, and wherein divide-shut brake coil is The critical component of its operating mechanism.On/off switch coil current can provide Mechanical Failure of HV Circuit Breaker diagnosis abundant letter used Breath；Therefore fault diagnosis can be carried out by extracting divide-shut brake coil current signal.

The method of existing high-voltage circuitbreaker fault detection has very much, is directed to various intelligent algorithms, such as: fuzzy Control can be with accurate mathematical tool by fuzzy concept or natural language sharpening, but its membership function and fuzzy rule are really Determining process, there are certain human factors；Radial base neural net provides a kind of relatively good for the troubleshooting issue of breaker Structural system, but there is the reasoning process of no method interpretation oneself and reasoning according to and data it is insufficient when neural network without The shortcomings that method works normally；LVQ has network structure simple for high-voltage circuitbreaker diagnosis, is not necessarily to data prediction, only needs to survey It calculates input vector and competition interfloor distance can be achieved with the advantages of effectively classifying.But LVQ classification is larger by initial value affecting and does not examine Consider the different feature of each dimension attribute importance of data, therefore, to solve the above-mentioned problems, this patent proposes a kind of based on ICA- The Fault Diagnosis for HV Circuit Breakers method of LVQ (ICA is immune clone algorithm, and LVQ is learning vector quantization neural network), can be While effectively solving the above problems, more quickly and accurately classify to failure.

Summary of the invention

The Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ that the object of the present invention is to provide a kind of, can be accurately real Existing Fault Diagnosis for HV Circuit Breakers.

The technical scheme adopted by the invention is that the Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ, it is specific by with Lower step is implemented:

Step 1: choosing typical data sample, be divided into training set and test set according to the ratio of 3:1；

Step 2: extracting the input feature value through the resulting training sample of step 1, carried out at dimensionality reduction using LDA algorithm Reason, obtains a new input feature value；

Step 3: the input feature value that step 2 is obtained is as the input of building ICA-LVQ neural network, by training Study output failure modes are as a result, set up the Fault Diagnosis for HV Circuit Breakers model based on ICA-LVQ with this；

Step 4: the Fault Diagnosis for HV Circuit Breakers model that step 3 obtains divides the test set sample in step 1 Class counts its classification accuracy.

The features of the present invention also characterized in that

Step 1 is specific to be implemented according to the following steps:

Step 1.1, I breaker monitored₁,I₂,I₃,t₁,t₂,t₃,t₄,t₅As Fault Diagnosis for HV Circuit Breakers mould The input parameter of type；

Step 1.2, it is divided into training set and test set according to the ratio of 3:1 to through the resulting sample data of step 1.1, training Collection is used to construct Fault Diagnosis for HV Circuit Breakers model, and test set is then used to the classifying quality of test model.

Step 2 is specific to be implemented according to the following steps:

Step 2.1, sample set D={ (x is constructed₁,y₁),(x₂,y₂),…(x_m,y_m), wherein x_iFor the input feature vector of n dimension Vector, y_iFor corresponding class label；

Step 2.2, dimension-reduction treatment is carried out to the input feature value that obtains through step 2.1, the input feature vector after dimensionality reduction to Amount is D', and the dimension after dimensionality reduction is d；

Specific step is as follows for dimension-reduction treatment in step 2.2:

(1) Scatter Matrix S in class is calculated_w,

Wherein, X_jFor the set of jth class sample, x ∈ X_j, j=1,2 ..., m；μ_jFor the mean vector of jth class sample, j= 1,2,…m；

(2) class scatter matrix S is calculated_b,

S_b=(μ₀-μ₁)(μ₀-μ₁)^T (2)

Wherein, μ_jFor the mean vector of jth class sample, j=1,2 ... m；

(3) calculating matrix S_W ^-1S_b, wherein S_w ^-1It is S_wInverse matrix, calculation formula is as follows,

(4) calculating matrix S_W ^-1S_bCharacteristic value, select maximum d characteristic value and its corresponding feature vector (w₁, w₂,…w_d), projection matrix W is obtained,

S_bW=λ S_ww (4)

Wherein λ is Lagrange multiplier, it is noted that S_bThe direction perseverance of w is μ₀-μ₁, can enable

S_bW=λ (μ₀-μ₁) (5)

Above formula is substituted into obtain

W=S_w ^-1(μ₀-μ₁) (6)

The closed solutions of W are then Sw^-1S_bThe corresponding matrix of d' maximum non-zero generalized eigenvalue, d'≤N-1, N-1 are non- The quantity of zero eigenvalue,

S_bW=λ S_wW (7)

W is by (w₁,w₂,…w_d) be a n × d composed by base vector matrix, the solution of w can by formula (4), formula (5), Formula (6) acquires；

(5) x is converted into each of sample set sample characteristics_iNew sample z_i=W^Tx_i；

(6) input feature value the D'={ (z after obtaining dimensionality reduction₁,y₁),(z₂,y₂),…(z_m,y_m)}。

Step 3 is specifically implemented according to the following steps:

Step 3.1, the input feature vector parameter in extraction step 2 in sample set is as the defeated of building ICA-LVQ neural network Enter；

Step 3.2, network is initialized, it is initial to set weight W¹, W²；Taking ICA algorithm is that the selection of LVQ neural network is best W¹；Take following policy setting W²: when the neuron weight vector of competition layer and the expectation classification pair of subclass linear layer neuron At once, the connection weight between the two neurons sets 1, otherwise sets 0, W²Once setting, no longer changes, learning rate is set as η, most Big frequency of training t_m；

Step 3.3, input sample vector X (t),

Calculating input sample vector X (t) indicates the W of each subclass cluster centre¹The Euclidean distance of each row vector:

Step 3.4, it finds out apart from nearest weight row vector and determines the competition layer neuron j of triumph^*:

The jth of the output vector Y of competition layer^*A element is set as 1, and other is 0；

Step 3.5, the value of linear convergent rate layer output vector O, O=W are calculated²Its sequence of a unique nonzero element in Y, O Number k* indicates that X (t) is divided into kth * class；

Step 3.6, comparing cell reality output O and target desired output d；

If classification correctly corrects weight vector in the following manner:

Otherwise weight is adjusted against input sample direction:

The weight of other non-winning neurons remains unchanged；

Step 3.7, renewal learning rate,

As t < t_mWhen, t=t+1 goes to step 3.3 and inputs next sample, repeats each step until t=t_m, t_mFor most Big frequency of training.

It is that LVQ neural network chooses optimal W that ICA algorithm is taken in step 3.2¹Specific step is as follows:

(1) ICA parameter is defined

M n-dimensional vector v is randomly generated_k,0(k=1,2 ..., m) is as initial antibody A (0), A (0)=[v_1,0, v_2,0,…,v_m,0], clone sizes q, immunogene mutation probability p_mAnd error parameter epsilon, t=0, t are that frequency of training is initially 0；

(2) affinity of initial population is calculated, affinity calculation formula is as follows:

D_k,tThe average distance of antibody is arrived for individual,x_jIndicate that input sample vector, N indicate Number of groups, D_k,tCorresponding with D in formula in step 3.3 (8), i.e., numerical value is equal to each other, and dimension is identical；Individual x_jCorresponding input Corresponding to sample vector X (t), antibody A (k) and cluster centre W1, wherein antibody A (k)=[v_1,k,v_2,k,…,v_m,k]；

(3) initial weight is determined by ICA；

1) judge whether to meet termination condition:

If | d | < ε stops iteration, obtains optimized individual in population, that is, can determine and work as The optimal initial weight W of preceding population¹, otherwise continue；

2) clone sizes q carries out clone operations, population Y (k), Y (k)={ Y after being cloned to antibody population A (k)₁ (k),Y₂(k),Y₃(k),…,Y_n(k) },And Y_ij(k)=A_i(k), j= 1,2,3,…,q_i；

3) with Probability p_mImmunogene operation is realized to Y (k)New antibody population Z (k) is obtained, to realize above-mentioned behaviour Make process, algorithm takes step-by-step to negate strategy, i.e., to the new antibodies Y in clonal population Y (k)_ij(k) each gene position is with probability p_mPosition is exchanged, i.e.,

Random < p in formula_mIndicate that above procedure is in Probability p_mIn the case where occur；

4) size according to antibody affinity after immune operation carries out Immune Clone Selection operation, i.e., n are selected from A'(k) The biggish antibody of affinity forms new antibody population A (k+1), k=k+1, returns 1)

A'(k)=Z (k) ∪ A (k) (15).

The beneficial effects of the present invention are:

(1) the present invention is based on the Fault Diagnosis for HV Circuit Breakers method of ICA-LVQ, LDA is used to carry out dimensionality reduction to data, protects It stays its data important attribute to ignore secondary attribute, its classification accuracy can be improved.

(2) the present invention is based on the Fault Diagnosis for HV Circuit Breakers methods of ICA-LVQ, using ICA to the first of LVQ neural network Beginning weight optimizes, its convergence rate can be improved, and improves its LVQ neural network initial value tender subject, improves failure and examine Disconnected accuracy rate；

(3) the present invention is based on the Fault Diagnosis for HV Circuit Breakers method of ICA-LVQ, LDA and ICA is combined, is constructed Fault Diagnosis for HV Circuit Breakers model improves the efficiency and accuracy rate of Fault Diagnosis for HV Circuit Breakers.

Detailed description of the invention

Fig. 1 is the flow chart of the Fault Diagnosis for HV Circuit Breakers method the present invention is based on ICA-LVQ；

Fig. 2 is mesohigh circuit-breaker switching on-off coil current waveform figure of the present invention.

Specific embodiment

The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.

The 1. Fault Diagnosis for HV Circuit Breakers methods based on ICA-LVQ of the invention specifically press following step as shown in Figs. 1-2 It is rapid to implement:

Step 1: choosing typical data sample, be divided into training set and test set according to the ratio of 3:1；

Step 1 is specific to be implemented according to the following steps:

Step 1.1, I breaker monitored₁,I₂,I₃,t₁,t₂,t₃,t₄,t₅As Fault Diagnosis for HV Circuit Breakers mould The input parameter of type；

Step 1.2, it is divided into training set and test set according to the ratio of 3:1 to through the resulting sample data of step 1.1, training Collection is used to construct Fault Diagnosis for HV Circuit Breakers model, and test set is then used to the classifying quality of test model；

Step 2: the input feature value through the resulting training sample of step 1 is extracted, using LDA algorithm (linear discriminant point Analysis) dimension-reduction treatment is carried out, obtain a new input feature value；

Step 2 is specific to be implemented according to the following steps:

Step 2.1, sample set D={ (x is constructed₁,y₁),(x₂,y₂),…(x_m,y_m), wherein x_iFor the input feature vector of n dimension Vector, y_iFor corresponding class label；

Step 2.2, dimension-reduction treatment is carried out to the input feature value that obtains through step 2.1, the input feature vector after dimensionality reduction to Amount is D', and the dimension after dimensionality reduction is d；

Specific step is as follows for dimension-reduction treatment in step 2.2:

(1) Scatter Matrix S in class is calculated_w,

Wherein, X_jFor the set of jth class sample, x ∈ X_j, j=1,2 ..., m；μ_jFor the mean vector of jth class sample, j= 1,2,…m；

(2) class scatter matrix S is calculated_b,

S_b=(μ₀-μ₁)(μ₀-μ₁)^T (2)

Wherein, μ_jFor the mean vector of jth class sample, j=1,2 ... m；

(3) calculating matrix S_W ^-1S_b, wherein S_w ^-1It is S_wInverse matrix, calculation formula is as follows,

(4) calculating matrix S_W ^-1S_bCharacteristic value, select maximum d characteristic value and its corresponding feature vector (w₁, w₂,…w_d), projection matrix W is obtained,

S_bW=λ S_ww (4)

Wherein λ is Lagrange multiplier, it is noted that S_bThe direction perseverance of w is μ₀-μ₁, can enable

S_bW=λ (μ₀-μ₁) (5)

Above formula is substituted into obtain

W=S_w ^-1(μ₀-μ₁) (6)

The closed solutions of W are then Sw^-1S_bThe corresponding matrix of d' maximum non-zero generalized eigenvalue, d'≤N-1, N-1 are non- The quantity of zero eigenvalue,

S_bW=λ S_wW (7)

W is by (w₁,w₂,…w_d) be a n × d composed by base vector matrix, the solution of w can by formula (4), formula (5), Formula (6) acquires；

(5) x is converted into each of sample set sample characteristics_iNew sample z_i=W^Tx_i；

(6) input feature value the D'={ (z after obtaining dimensionality reduction₁,y₁),(z₂,y₂),…(z_m,y_m)}。

Step 3: the input feature value that step 2 is obtained is as the input of building ICA-LVQ neural network, by training Study output failure modes are as a result, set up the Fault Diagnosis for HV Circuit Breakers model based on ICA-LVQ with this；

Step 3 is specifically implemented according to the following steps:

Step 3.1, the input feature vector parameter in extraction step 2 in sample set is as the defeated of building ICA-LVQ neural network Enter；

Step 3.2, network is initialized, it is initial to set weight W¹, W²；Taking ICA algorithm is that the selection of LVQ neural network is best W¹；Take following policy setting W²: when the neuron weight vector of competition layer and the expectation classification pair of subclass linear layer neuron At once, the connection weight between the two neurons sets 1, otherwise sets 0, W²Once setting, no longer changes, learning rate is set as η, most Big frequency of training t_m；

It is that LVQ neural network chooses optimal W that step 3.2, which takes ICA algorithm,¹Specific step is as follows:

(1) ICA parameter is defined

M n-dimensional vector v is randomly generated_k,0(k=1,2 ..., m) is as initial antibody A (0), A (0)=[v_1,0, v_2,0,…,v_m,0], clone sizes q, immunogene mutation probability p_mAnd error parameter epsilon, t=0, t are that frequency of training is initially 0；

(2) affinity of initial population is calculated, affinity calculation formula is as follows:

D_k,tThe average distance of antibody is arrived for individual,x_jIndicate that input sample vector, N indicate Number of groups, D_k,tCorresponding with D in formula in step 3.3 (8), i.e., numerical value is equal to each other, and dimension is identical；Individual x_jCorresponding input Corresponding to sample vector X (t), antibody A (k) and cluster centre W1, wherein antibody A (k)=[v_1,k,v_2,k,…,v_m,k]；

(3) initial weight is determined by ICA；

1) judge whether to meet termination condition:

If | d | < ε stops iteration, obtains optimized individual in population (antibody) Determine the optimal initial weight W of current population¹, otherwise continue；

2) clone sizes q carries out clone operations, population Y (k), Y (k)={ Y after being cloned to antibody population A (k)₁ (k),Y₂(k),Y₃(k),…,Y_n(k) },And Y_ij(k)=A_i(k), j= 1,2,3,…,q_i；

3) with Probability p_mImmunogene operation is realized to Y (k)New antibody population Z (k) is obtained, to realize above-mentioned behaviour Make process, algorithm takes step-by-step to negate strategy, i.e., to the new antibodies Y in clonal population Y (k)_ij(k) each gene position is with probability p_mPosition is exchanged, i.e.,

Random < p in formula_mIndicate that above procedure is in Probability p_mIn the case where occur；

4) size according to antibody affinity after immune operation carries out Immune Clone Selection operation, i.e., n are selected from A'(k) The biggish antibody of affinity forms new antibody population A (k+1), k=k+1, returns 1)

A'(k)=Z (k) ∪ A (k) (15).

Step 3.3, input sample vector X (t),

Calculating input sample vector X (t) indicates the W of each subclass cluster centre¹The Euclidean distance of each row vector:

Step 3.4, it finds out apart from nearest weight row vector and determines the competition layer neuron j of triumph^*:

The jth of the output vector Y of competition layer^*A element is set as 1, and other is 0；

Step 3.5, the value of linear convergent rate layer output vector O, O=W are calculated²Its sequence of a unique nonzero element in Y, O Number k* indicates that X (t) is divided into kth * class；

Step 3.6, comparing cell reality output O and target desired output d；

If classification correctly corrects weight vector in the following manner:

Otherwise weight is adjusted against input sample direction:

The weight of other non-winning neurons remains unchanged；

Step 3.7, renewal learning rate,

As t < t_mWhen, t=t+1 goes to step 3.3 and inputs next sample, repeats each step until t=t_m, t_mFor most Big frequency of training.

Step 4: the Fault Diagnosis for HV Circuit Breakers model that step 3 obtains divides the test set sample in step 1 Class counts its classification accuracy.

Claims (5)

1. the Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ, which is characterized in that specifically implement according to the following steps:

Step 1: choosing typical data sample, be divided into training set and test set according to the ratio of 3:1；

Step 2: extracting the input feature value through the resulting training sample of step 1, dimension-reduction treatment is carried out using LDA algorithm, is obtained The input feature value new to one；

Step 3: the input feature value that step 2 is obtained learns as the input of building ICA-LVQ neural network by training Failure modes are exported as a result, setting up the Fault Diagnosis for HV Circuit Breakers model based on ICA-LVQ with this；

Step 4: the Fault Diagnosis for HV Circuit Breakers model that step 3 obtains classifying to the test set sample in step 1, unites Count its classification accuracy.

2. the Fault Diagnosis for HV Circuit Breakers method according to claim 1 based on ICA-LVQ, which is characterized in that described Step 1 is specific to be implemented according to the following steps:

Step 1.1, I breaker monitored₁,I₂,I₃,t₁,t₂,t₃,t₄,t₅As Fault Diagnosis for HV Circuit Breakers model Input parameter；

Step 1.2, it is divided into training set and test set according to the ratio of 3:1 to through the resulting sample data of step 1.1, training set is used Fault Diagnosis for HV Circuit Breakers model is constructed, test set is then used to the classifying quality of test model.

3. the Fault Diagnosis for HV Circuit Breakers method according to claim 1 based on ICA-LVQ, which is characterized in that described Step 2 is specific to be implemented according to the following steps:

Step 2.1, sample set D={ (x is constructed₁,y₁),(x₂,y₂),…(x_m,y_m), wherein x_iFor n dimension input feature value, y_iFor corresponding class label；

Step 2.2, dimension-reduction treatment is carried out to the input feature value obtained through step 2.1, the input feature value after dimensionality reduction is D', the dimension after dimensionality reduction are d；

Specific step is as follows for dimension-reduction treatment in step 2.2:

(1) Scatter Matrix S in class is calculated_w,

Wherein, X_jFor the set of jth class sample, x ∈ X_j, j=1,2 ..., m；μ_jFor the mean vector of jth class sample, j=1, 2,…m；

(2) class scatter matrix S is calculated_b,

S_b=(μ₀-μ₁)(μ₀-μ₁)^T (2)

Wherein, μ_jFor the mean vector of jth class sample, j=1,2 ... m；

(3) calculating matrix S_W ^-1S_b, wherein S_w ^-1It is S_wInverse matrix, calculation formula is as follows,

(4) calculating matrix S_W ^-1S_bCharacteristic value, select maximum d characteristic value and its corresponding feature vector (w₁,w₂,…w_d), Projection matrix W is obtained,

S_bW=λ S_ww (4)

Wherein λ is Lagrange multiplier, it is noted that S_bThe direction perseverance of w is μ₀-μ₁, can enable

S_bW=λ (μ₀-μ₁) (5)

Above formula is substituted into obtain

W=S_w ^-1(μ₀-μ₁) (6)

The closed solutions of W are then Sw^-1S_bThe corresponding matrix of d' maximum non-zero generalized eigenvalue, d'≤N-1, N-1 are that non-zero is special The quantity of value indicative,

S_bW=λ S_wW (7)

W is by (w₁,w₂,…w_d) be base vector composed by a n × d matrix, the solution of w can be by formula (4), formula (5), formula (6) It acquires；

(5) x is converted into each of sample set sample characteristics_iNew sample z_i=W^Tx_i；

(6) input feature value the D'={ (z after obtaining dimensionality reduction₁,y₁),(z₂,y₂),…(z_m,y_m)}。

4. the Fault Diagnosis for HV Circuit Breakers method according to claim 1 based on ICA-LVQ, which is characterized in that described Step 3 is specifically implemented according to the following steps:

Step 3.1, input of the input feature vector parameter in extraction step 2 in sample set as building ICA-LVQ neural network；

Step 3.2, network is initialized, it is initial to set weight W¹, W²；Taking ICA algorithm is that LVQ neural network selects optimal W¹； Take following policy setting W²: when the neuron weight vector of competition layer is corresponding with the expectation classification of subclass linear layer neuron When, the connection weight between the two neurons sets 1, otherwise sets 0, W²Once setting, no longer changes, learning rate is set as η, maximum Frequency of training t_m；

Step 3.3, input sample vector X (t),

Calculating input sample vector X (t) indicates the W of each subclass cluster centre¹The Euclidean distance of each row vector:

Step 3.4, it finds out apart from nearest weight row vector and determines the competition layer neuron j of triumph^*:

The jth of the output vector Y of competition layer^*A element is set as 1, and other is 0；

Step 3.5, the value of linear convergent rate layer output vector O, O=W are calculated²Its serial number of unique nonzero element k* is just in Y, O Show that X (t) is divided into kth * class；

Step 3.6, comparing cell reality output O and target desired output d；

If classification correctly corrects weight vector in the following manner:

Otherwise weight is adjusted against input sample direction:

The weight of other non-winning neurons remains unchanged；

Step 3.7, renewal learning rate,

As t < t_mWhen, t=t+1 goes to step 3.3 and inputs next sample, repeats each step until t=t_m, t_mFor maximum instruction Practice number.

5. the Fault Diagnosis for HV Circuit Breakers method according to claim 4 based on ICA-LVQ, which is characterized in that step It is that LVQ neural network chooses optimal W that ICA algorithm is taken in 3.2¹Specific step is as follows:

(1) ICA parameter is defined

M n-dimensional vector v is randomly generated_k,0(k=1,2 ..., m) is as initial antibody A (0), A (0)=[v_1,0,v_2,0,…, v_m,0], clone sizes q, immunogene mutation probability p_mAnd error parameter epsilon, t=0, t are that frequency of training is initially 0；

(2) affinity of initial population is calculated, affinity calculation formula is as follows:

D_k,tThe average distance of antibody is arrived for individual,x_jIndicate that input sample vector, N indicate group Number, D_k,tCorresponding with D in formula in step 3.3 (8), i.e., numerical value is equal to each other, and dimension is identical；Individual x_jCorresponding input sample Corresponding to vector X (t), antibody A (k) and cluster centre W1, wherein antibody A (k)=[v_1,k,v_2,k,…,v_m,k]；

(3) initial weight is determined by ICA；

1) judge whether to meet termination condition:

If | d | < ε stops iteration, obtains optimized individual in population, that is, can determine current population Optimal initial weight W¹, otherwise continue；

2) clone sizes q carries out clone operations, population Y (k), Y (k)={ Y after being cloned to antibody population A (k)₁(k),Y₂ (k),Y₃(k),…,Y_n(k) },And Y_ij(k)=A_i(k), j=1,2, 3,…,q_i；

3) with Probability p_mImmunogene operation is realized to Y (k)Obtain new antibody population Z (k), Yao Shixian aforesaid operations mistake Journey, algorithm takes step-by-step to negate strategy, i.e., to the new antibodies Y in clonal population Y (k)_ij(k) each gene position is with Probability p_mIt hands over Change place, i.e.,

Random < p in formula_mIndicate that above procedure is in Probability p_mIn the case where occur；

4) size according to antibody affinity after immune operation carries out Immune Clone Selection operation, i.e., it is a affine that n is selected from A'(k) Biggish antibody is spent, new antibody population A (k+1), k=k+1 are formed, is returned 1)

A'(k)=Z (k) ∪ A (k) (15).