CN108470158A - A method of it finding error minimal network for dynamic ECG data and calculates structure - Google Patents
A method of it finding error minimal network for dynamic ECG data and calculates structure Download PDFInfo
- Publication number
- CN108470158A CN108470158A CN201810191992.9A CN201810191992A CN108470158A CN 108470158 A CN108470158 A CN 108470158A CN 201810191992 A CN201810191992 A CN 201810191992A CN 108470158 A CN108470158 A CN 108470158A
- Authority
- CN
- China
- Prior art keywords
- data
- gene
- individual
- new
- population
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The method that error minimal network calculates structure is found for dynamic ECG data the present invention provides a kind of, is included the following steps:Step 1, initialization:S11, each individual Xi of initialization;S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;S14, initialization algebraically gen, order gen=0;Step 2 calculates the error Fitness [i] of individual Xi:S21, life i=1;S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;S23, by gradient decline in the way of and current electrocardiogram (ECG) data in convolutional Neural network configuration CNN1 calculating formula carry out weights iteration convergence, obtain the best initial weights matrix W 1 of convolutional Neural network configuration CNN1.The when of leading to electrocardiosignal abnormality detection due to the method that error minimal network calculates structure solves the problems, such as Network Computing Architecture as the electrocardiogram (ECG) data for everyone is arranged in the prior art is found for dynamic ECG data to judge by accident.
Description
Technical field
The present invention relates to the processing of dynamic ECG data, and in particular to one kind is that dynamic ECG data finds error minimal network
The method for calculating structure.
Background technology
Since everyone electrocardiogram (ECG) data is different, being most suitable for everyone Network Computing Architecture of electrocardiogram (ECG) data is also
It is different.And in the prior art, in order to detect the appearance of electrocardiosignal abnormal signal, all it is being that certain network calculations are set
Structure, therefore can cause to occur phenomena such as erroneous judgement by accident.
Invention content
The present invention will provide a kind of method for finding error minimal network calculating structure for dynamic ECG data, solve existing
When leading to electrocardiosignal abnormality detection due to Network Computing Architecture as the electrocardiogram (ECG) data for everyone is arranged in technology
The problem of judging by accident.
To achieve the above object, present invention employs the following technical solutions:
Present invention firstly provides a kind of methods for finding error minimal network calculating structure for dynamic ECG data, including with
Lower step:
Step 1, initialization:
S11, each individual Xi being initialized, Xi is the interior array formed containing N number of gene, individual equipped with M Xi, 1≤i≤
M, each gene is interior there are three serial data, and a data string indicates that calculating formula, another two serial data indicate the node connected forward,
It is for indicating serial data expression or do not express 0 or 1 to have one-bit digital in each serial data of expression node;
S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;
S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;
S14, initialization algebraically gen, order gen=0;
Step 2 calculates the error Fitness [i] of individual Xi:
S21, life i=1;
S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;
S23, calculating formula in convolutional Neural network configuration CNN1 is carried out with current electrocardiogram (ECG) data in such a way that gradient declines
The iteration convergence of weights obtains the best initial weights matrix W 1 of convolutional Neural network configuration CNN1;
S24, current electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN1 with best initial weights matrix W 1, is counted
The error rate F obtainedi_new1;
S25, life Fitness [i] are equal to Fi_new1;
S26, i=i+1;
S27, judge whether i is more than M, if step S22 is otherwise carried out, if carrying out step 3;
Step 3: setting the first population A1Interior each individual is A1i, A1iFor length and XiEqual array, A1iXi is replicated,
It is A1i=Xi(1≤i≤M), G1 [i]=Fitness [i], obtains the first population A1Optimum individual the step of include:
S301, gen=gen+1 is calculated;
S302, life i=gen;
S303, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S304, intersection:First, it is the Probability p currently randomly generated to take arbitrary floating-point values in 0~11;Then, judge to work as
Before the Probability p that randomly generates1Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein
Intersecting step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range
An integer Z is randomly selected in (0, M)1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1It is a
Gene is to m-th gene;
S305, variation:First, it is the Probability p currently randomly generated to take arbitrary floating-point values in 0~12;Then, judge to work as
Before the Probability p that randomly generates2Whether mutagenic factor F is less than, if then taking DE/rand strategies to new individual ViInto row variation, if
It is not then without variation, wherein take DE/rand strategies to new individual ViInto row variation:S3051, from integer range (0, M)
In randomly select an integer Z2;S3052, M-Z is randomly selected2Each base is corresponded in the data group C of+1 mrna length, data group C
The serial data of cause meets the regulation of gene data string setting;S3053, new individual ViInterior Z2A character string to m-th gene
Become data in data group C successively;
S306, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient
Decline mode and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolution god
Best initial weights matrix W 2 through network configuration CNN2;
S307, verification assessment:Electrocardiogram (ECG) data is input to the convolutional Neural network configuration with best initial weights matrix W 2
In CNN2, the error rate Fi_new that is calculated;
S308, judge whether the error rate Fi_new of new individual Vi is less than the error rate G1 [i] of parent Ui, if so, life
G1 [i] is equal to Fi_new and A1iReplicate Vi(being so that the gene in i-th of individual Ai is completely as Vi);If it is not, then
G1 [i] is not changed;
S309, judge whether i is equal to N, step S310 is if it is carried out, if not then carrying out step S301;
S310, compare the corresponding error rate G1 [1] of all individuals in the first new population A1, G1 [2] ... G1
[i] ..., G1 [M-1], G1 [M] size select minimal error rate min G1 [Z3], and find minimal error rate min G1 [Z3]
Corresponding individualObtain the first population A1Optimum individual
The method that error minimal network calculates structure is found for dynamic ECG data the present invention also provides a kind of, including following
Step:
Step 1, initialization:
S11, each individual Xi being initialized, Xi is the interior array formed containing N number of gene, individual equipped with M Xi, 1≤i≤
M, each gene is interior there are three serial data, and a data string indicates that calculating formula, another two serial data indicate the node connected forward,
It is for indicating serial data expression or do not express 0 or 1 to have one-bit digital in each serial data of expression node;
S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;
S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;
S14, initialization algebraically gen, order gen=0;
Step 2 calculates the error Fitness [i] of first generation individual Xi:
S21, life i=1;
S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;
S23, calculating formula in convolutional Neural network configuration CNN1 is carried out with current electrocardiogram (ECG) data in such a way that gradient declines
The iteration convergence of weights obtains the best initial weights matrix W 1 of convolutional Neural network configuration CNN1;
S24, current electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN1 with best initial weights matrix W 1, is counted
The error rate F obtainedi_new1;
S25, life Fitness [i] are equal to Fi_new1;
S26, i=i+1;
S27, judge whether i is more than M, if step S22 is otherwise carried out, if carrying out step 3;
Step 3: setting second of population A2Interior each individual is A2i, A2iFor length and XiEqual array, A2iXi is replicated,
It is A2i=Xi(1≤i≤M), G2 [i]=Fitness [i] obtain second of population A2The step of include:
S411, gen=gen+1 is calculated;
S412, life i=gen;
S413, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S414, intersection:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~13;Then, judge to work as
Before the probability P that randomly generates3Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein
Intersecting step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range
An integer Z is randomly selected in (0, M)1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1It is a
Gene is to m-th gene;
S415, variation:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~14;Then, judge to work as
Before the probability P that randomly generates4Whether mutagenic factor F is less than, if then taking DE/best strategies to new individual ViInto row variation, if not
It is then without variation, wherein take DE/best strategies to new individual ViInto row variation:It is S4151, random from integer range (0, M)
Choose three integer R1、R2And R3;S4152, new individual ViIt is interior indicate gene each data withAndIt is interior
Indicate that each data of gene carry out the operation such as formula (1-1) successively,
In formula, bracket [] indicates the rounding by the way of rounding upResult of calculation, XimnTable
Show new individual ViNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualN-th in interior m-th of gene
Data, if the data volume of each gene is Y, then 1≤m≤M, 1≤n≤Y;
S416, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient
Decline mode and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolution god
Best initial weights matrix W 2 through network configuration CNN2;
S417, verification assessment:Electrocardiogram (ECG) data is input to the convolutional Neural network configuration with best initial weights matrix W 2
In CNN2, the error rate Fi_new that is calculated;
S418, judge whether the error rate Fi_new of new individual Vi is less than the error rate G2 [i] of parent Ui, if so, life
G2 [i] is equal to Fi_new and orders A2iReplicate Vi(it is so that i-th of individual A1iInterior gene is completely and ViEqually);If it is not,
G2 [i] is not changed then;
S419, judge whether i is equal to N, step S420 is if it is carried out, if not then carrying out step S411;
S420, compare the first new population A2The corresponding error rate G2 [1] of interior all individuals, G2 [2] ... G2
[i] ..., G2 [n-1], G2 [N] size select minimal error rate min G2 [Z4], and find minimal error rate min G2 [Z4]
Corresponding individualObtain second of population A2Global optimum individual
Compared with the prior art, the present invention has the advantages that:
The calculating step for replacing each calculating formula in convolutional neural networks structure at any time is realized, realizes use to current
The electrocardiogram (ECG) data of acquisition and electrocardiogram (ECG) data result judge a variety of convolutional neural networks structures, calculate each convolutional Neural
The error of the judging result of network structure and true electrocardiogram (ECG) data result passes through the error between each convolutional neural networks structure
Compare, obtained most adaptable convolutional neural networks structure so that more adapts to judge current personal electrocardiogram (ECG) data, avoid
There is the result judged by accident to occur.
Part is illustrated to embody by further advantage, target and the feature of the present invention by following, and part will also be by this
The research and practice of invention and be understood by the person skilled in the art.
Description of the drawings
Fig. 1 is a kind of structural schematic diagram of convolutional neural networks structure;
Fig. 2 is the individual Xi that convolutional neural networks structure is formed in Fig. 1.
Specific implementation mode
In order to which so that the present invention is realized technological means, creation characteristic, reached purpose more understand and are apparent to effect,
The present invention is further elaborated With reference to embodiment:
As shown in Figure 1 and Figure 2, it is that dynamic ECG data finds error minimal network calculating knot that the present invention, which proposes a kind of,
The method of structure, includes the following steps:
Step 1, initialization:
S11, each individual Xi of initialization (each individual Xi corresponds to a convolutional neural networks structure), Xi is interior contains
The array of N number of gene composition is equipped with M Xi individual, and 1≤i≤M, each gene is interior there are three serial data, and a data string indicates
Calculating formula, another two serial data indicate the node that connects forward, have in each serial data of expression node one-bit digital be for
Indicate serial data expression or do not express 0 or 1;Such as:
Being stored with main subcomponent (calculating formula 18, then each calculating formula can be indicated with 0 to 17) includes
ConvBlock, ResBlock, Max pooling, Average pooling, Summation, Concatenation, input,
output.Wherein ConvBlock and ResBlock all include three kinds output 32,64,128, convolution kernel all there are two types of, be 3* respectively
3 and 5*5.In figure, since input is first gene certainly, input is represented with 0 in gene, has 7 genes in Fig. 2,
Every three small frames represent a gene, and first small frame is intended to indicate that the code name of this calculating formula is (specific namely in database
Which calculating formula), whether the first digit represents data representation in second small frame, the second digit be connected to upper one as this calculating formula
The serial number (which gene in namely Xi) of calculating formula, whether the first digit represents data representation in the small frame of third, second
Number represents the serial number that this calculating formula is connected to a calculating formula;The code name of each calculating formula is as follows in the database:The meter of input
Formula code name is 0, the calculating formula code name of conv (64,3) (it is the ConvBlock that the convolution kernel that output is 64 is 3*3) is 1,
The calculating formula code name of pool (avg) (it is Average pooling) is 2, (it is the convolution kernel that output is 32 to conv (32,5)
For the ConvBlock of 5*5) calculating formula code name be 3, conv (32,3) (its be output be 32 convolution kernel be 3*3's
ConvBlock calculating formula code name) is 6, the calculating formula code name of sum is 4, the calculating formula code name of output is 5, then wherein one
Individual XiThere are the data group of 7 genes, Xi=(011,001 021,001 031,102 060,112 041,314 030113
051506), XiTo there is a data group of 7 genes, each gene is there are five data, and second digit is indicated with 0 in each gene
Fourth digit is not expressed, 1 indicates that fourth digit is expressed, and fourth digit indicates fourth digit not with 0 in each gene
Expression, 1 indicate fourth digit expression;So each individual Xi has 6*7 number;
S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;
S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;
S14, initialization algebraically gen, order gen=0;
Step 2 calculates the error Fitness [i] of individual Xi:
S21, life i=1;
S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;
S23, by gradient decline in the way of and current electrocardiogram (ECG) data (including testing result) to convolutional Neural network configuration
Calculating formula carries out the iteration convergence of weights in CNN1, obtains the best initial weights matrix W 1 of convolutional Neural network configuration CNN1;
S24, current electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN1 with best initial weights matrix W 1, is counted
The error rate F obtainedi_new1;
S25, life Fitness [i] are equal to Fi_new1;
S26, i=i+1;
S27, judge whether i is more than M, if step S22 is otherwise carried out, if carrying out step 3;
Step 3: setting the first population A1Interior each individual is A1i, A1iFor length and XiEqual array, A1iXi is replicated,
It is A1i=Xi(1≤i≤M, i take 1 to M to be intended to replicate), G1 [i]=Fitness [i], obtains the first population A1It is optimal
Individual step include:
S301, gen=gen+1 is calculated;
S302, life i=gen;
S303, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi(i takes 1 to M to be intended to again
System);
S304, intersection:First, it is the Probability p currently randomly generated to take arbitrary floating-point values in 0~11It (improves random
Property);Then, judge the Probability p currently randomly generated1Whether be less than intersect factor CR (for it is follow-up some do not intersect base be provided
Plinth), if then being intersected, if not then without intersecting, wherein intersecting step is:S3041, from integer range (1, M) with
Machine chooses an integer R, and R is not equal to i;S3042, an integer Z is randomly selected from integer range (0, M)1;S3043, by XR
In Z1A gene is all copied to V to m-th geneiZ1A gene to m-th gene (such as:It is V in Fig. 2i, one
Equal length designs X of each digit according to the same setting ruleRSo that data ViIt is just array XR) (such as Fig. 2 and Fig. 1,
For 7 genes, then choosing 1 to 7 any one integer, such as Z1Equal to 5, then just by ViThe 5th gene to the 7th base
Because replacing with XRThe 5th gene to the 7th gene);
S305, variation:First, it is the Probability p currently randomly generated to take arbitrary floating-point values in 0~12;Then, judge to work as
Before the Probability p that randomly generates2Whether mutagenic factor F is less than, if then taking DE/rand strategies to new individual ViInto row variation, if
It is not then without variation, wherein take DE/rand strategies to new individual ViInto row variation:S3051, from integer range (0, M)
In randomly select an integer Z2;S3052, M-Z is randomly selected2Each base is corresponded in the data group C of+1 mrna length, data group C
The serial data of cause meets the regulation of gene data string setting;S3053, new individual ViInterior Z2A character string to m-th gene
Become data in data group C successively;(such as Fig. 2 and Fig. 1, if 7 genes, then 1 to 7 any one integer is chosen, such as
Z2Equal to 6, then 10 bit array (4150450216) is just generated at random, by ViThe 6th gene and the 7th gene number
According to replacing with 10 bit arrays (41,504 50216))
S306, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient
Decline mode and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolution god
Best initial weights matrix W 2 through network configuration CNN2;
S307, verification assessment:Electrocardiogram (ECG) data is input to the convolutional Neural network configuration with best initial weights matrix W 2
In CNN2, the error rate Fi_new that is calculated;
S308, judge whether the error rate Fi_new of new individual Vi is less than the error rate G1 [i] of parent Ui, if so, life
G1 [i] is equal to Fi_new and A1iReplicate Vi(being so that the gene in i-th of individual Ai is completely as Vi);If it is not, then
G1 [i] is not changed;
S309, judge whether i is equal to N, step S310 is if it is carried out, if not then carrying out step S301;
S310, compare the corresponding error rate G1 [1] of all individuals in the first new population A1, G1 [2] ... G1
[i] ..., G1 [M-1], G1 [M] size select minimal error rate min G1 [Z3], and find minimal error rate min G1 [Z3]
Corresponding individualObtain the first population A1Optimum individual
In order to calculate a greater variety of convolutional neural networks structures so that according to not in each convolutional neural networks structure
Rule that can be same finds out the convolutional neural networks structure that the first population can not be found out, and further increases the convolutional Neural net searched out
The accuracy of network structure further includes:The first population A is obtained after having carried out step 31With second of population A2It is common optimal
The step of body;
Obtain the first population A1With second of population A2Common optimum individual the step of include:
S41, second of population A is set2Interior each individual is A2i, A2iFor length and XiEqual array, A2iXi is replicated, is
A2i=Xi(1≤i≤M), G2 [i]=Fitness [i] obtain second of population A2The step of include:
S411, gen=gen+1 is calculated;
S412, life i=gen;
S413, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S414, intersection:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~13;Then, judge to work as
Before the probability P that randomly generates3Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein
Intersecting step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range
An integer Z is randomly selected in (0, M)1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1It is a
Gene is to m-th gene;
S415, variation:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~14;Then, judge to work as
Before the probability P that randomly generates4Whether mutagenic factor F is less than, if then taking DE/best strategies to new individual ViInto row variation, if
It is not then without variation, wherein take DE/best strategies to new individual ViInto row variation:S4151, from integer range (0, M)
Randomly select three integer R1、R2And R3;S4152, new individual ViIt is interior indicate gene each data withAnd
It is interior to indicate that each data of gene carry out the operation such as formula (1-1) successively,
In formula, bracket [] indicates the rounding by the way of rounding upResult of calculation, XimnTable
Show new individual ViNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualN-th in interior m-th of gene
Data, if the data volume of each gene is Y, then 1≤m≤M, 1≤n≤Y;(such as Fig. 2 and Fig. 1, if 7 genes, then often
A ViAll it is the array of 6*7 digits, F=0.5, then Xi11=[0+0.5
× (0-0)]=0, Xi12=[3+0.5 × (2-4)]=1, Xi13=[1+0.5 × (1-1)]=1 ..., then Vi=
011……)
S416, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient
Decline mode and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolution god
Best initial weights matrix W 2 through network configuration CNN2;
S417, verification assessment:Electrocardiogram (ECG) data is input to the convolutional Neural network configuration with best initial weights matrix W 2
In CNN2, the error rate Fi_new that is calculated;
S418, judge whether the error rate Fi_new of new individual Vi is less than the error rate G2 [i] of parent Ui, if so, life
G2 [i] is equal to Fi_new and orders A2iReplicate Vi(it is so that i-th of individual A1iInterior gene is completely and ViEqually);If it is not,
G2 [i] is not changed then;
S419, judge whether i is equal to N, step S420 is if it is carried out, if not then carrying out step S411;
S420, compare the first new population A2The corresponding error rate G2 [1] of interior all individuals, G2 [2] ... G2
[i] ..., G2 [n-1], G2 [N] size select minimal error rate min G2 [Z4], and find minimal error rate min G2 [Z4]
Corresponding individual AZ4To get to second of population A2Global optimum individual AZ4;
S42, population exchange:The first population A1With second of population A2It is exchanged:Compare the first population A1Minimum
Error rate min G1 [Z3] and second of population A2Minimal error rate min G2 [Z4] size, small min G1 [Z3] or min
G2[Z4] corresponding individual for the first population A1With second of population A2Common optimum individual.
The present embodiment also provide it is a kind of for dynamic ECG data find error minimal network calculate structure method, including with
Lower step:
Step 1, initialization:
S11, each individual Xi being initialized, Xi is the interior array formed containing N number of gene, individual equipped with M Xi, 1≤i≤
M, each gene is interior there are three serial data, and a data string indicates that calculating formula, another two serial data indicate the node connected forward,
It is for indicating serial data expression or do not express 0 or 1 to have one-bit digital in each serial data of expression node;
S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;
S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;
S14, initialization algebraically gen, order gen=0;
Step 2 calculates the error Fitness [i] of first generation individual Xi:
S21, life i=1;
S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;
S23, calculating formula in convolutional Neural network configuration CNN1 is carried out with current electrocardiogram (ECG) data in such a way that gradient declines
The iteration convergence of weights obtains the best initial weights matrix W 1 of convolutional Neural network configuration CNN1;
S24, current electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN1 with best initial weights matrix W 1, is counted
The error rate F obtainedi_new1;
S25, life Fitness [i] are equal to Fi_new1;
S26, i=i+1;
S27, judge whether i is more than M, if step S22 is otherwise carried out, if carrying out step 3;
Step 3: setting second of population A2Interior each individual is A2i, A2iFor length and XiEqual array, A2iXi is replicated,
It is A2i=Xi(1≤i≤M), G2 [i]=Fitness [i] obtain second of population A2The step of include:
S411, gen=gen+1 is calculated;
S412, life i=gen;
S413, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S414, intersection:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~13;Then, judge to work as
Before the probability P that randomly generates3Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein
Intersecting step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range
An integer Z is randomly selected in (0, M)1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1It is a
Gene is to m-th gene;
S415, variation:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~14;Then, judge to work as
Before the probability P that randomly generates4Whether mutagenic factor F is less than, if then taking DE/best strategies to new individual ViInto row variation, if
It is not then without variation, wherein take DE/best strategies to new individual ViInto row variation:S4151, from integer range (0, M)
Randomly select three integer R1、R2And R3;S4152, new individual ViIt is interior indicate gene each data withAnd
It is interior to indicate that each data of gene carry out the operation such as formula (1-1) successively,
In formula, bracket [] indicates the rounding by the way of rounding upResult of calculation, XimnTable
Show new individual ViNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualN-th in interior m-th of gene
Data, if the data volume of each gene is Y, then 1≤m≤M, 1≤n≤Y;
S416, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient
Decline mode and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolution god
Best initial weights matrix W 2 through network configuration CNN2;
S417, verification assessment:Electrocardiogram (ECG) data is input to the convolutional Neural network configuration with best initial weights matrix W 2
In CNN2, the error rate Fi_new that is calculated;
S418, judge whether the error rate Fi_new of new individual Vi is less than the error rate G2 [i] of parent Ui, if so, life
G2 [i] is equal to Fi_new and orders A2iReplicate Vi(it is so that i-th of individual A1iInterior gene is completely and ViEqually);If it is not,
G2 [i] is not changed then;
S419, judge whether i is equal to N, step S420 is if it is carried out, if not then carrying out step S411;
S420, compare the first new population A2The corresponding error rate G2 [1] of interior all individuals, G2 [2] ... G2
[i] ..., G2 [n-1], G2 [N] size select minimal error rate min G2 [Z4], and find minimal error rate min G2 [Z4]
Corresponding individual AZ4To get to second of population A2Global optimum individual AZ4。
Finally illustrate, the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although with reference to compared with
Good embodiment describes the invention in detail, it will be understood by those of ordinary skill in the art that, it can be to the skill of the present invention
Art scheme is modified or replaced equivalently, and without departing from the objective and range of technical solution of the present invention, should all be covered at this
In the right of invention.
Claims (3)
1. a kind of finding the method that error minimal network calculates structure for dynamic ECG data, which is characterized in that including following step
Suddenly:
Step 1, initialization:
S11, each individual Xi is initialized, the array that Xi forms in containing N number of gene is individual equipped with M Xi, 1≤i≤M, often
There are three serial data in a gene, a data string indicates that calculating formula, another two serial data indicate the node connected forward, expression
It is for indicating serial data expression or do not express 0 or 1 to have one-bit digital in each serial data of node;
S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;
S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;
S14, initialization algebraically gen, order gen=0;
Step 2 calculates the error Fitness [i] of individual Xi:
S21, life i=1;
S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;
S23, by gradient decline in the way of and current electrocardiogram (ECG) data in convolutional Neural network configuration CNN1 calculating formula carry out weights
Iteration convergence, obtain the best initial weights matrix W 1 of convolutional Neural network configuration CNN1;
S24, current electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN1 with best initial weights matrix W 1, is calculated
The error rate F gone outi_new1;
S25, life Fitness [i] are equal to Fi_new1;
S26, i=i+1;
S27, judge whether i is more than M, if step S22 is otherwise carried out, if carrying out step 3;
Step 3: setting the first population A1Interior each individual is A1i, A1iFor length and XiEqual array, A1iXi is replicated, is
A1i=Xi(1≤i≤M), G1 [i]=Fitness [i], obtains the first population A1Optimum individual the step of include:
S301, gen=gen+1 is calculated;
S302, life i=gen;
S303, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S304, intersection:First, it is the Probability p currently randomly generated to take arbitrary floating-point values in 0~11;Then, judge currently with
The Probability p that machine generates1Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein intersect
Step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range (0, M)
In randomly select an integer Z1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1A gene
To m-th gene;
S305, variation:First, it is the Probability p currently randomly generated to take arbitrary floating-point values in 0~12;Then, judge currently with
The Probability p that machine generates2Whether mutagenic factor F is less than, if then taking DE/rand strategies to new individual ViInto row variation, if not
Then without variation, wherein take DE/rand strategies to new individual ViInto row variation:S3051, from integer range (0, M) with
Machine chooses an integer Z2;S3052, M-Z is randomly selected2Each gene is corresponded in the data group C of+1 mrna length, data group C
Serial data meets the regulation of gene data string setting;S3053, new individual ViInterior Z2A character string to m-th gene is successively
Become data in data group C;
S306, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient decline side
Formula and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolutional Neural network
The best initial weights matrix W 2 of structure C NN2;
S307, verification assessment:Electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN2 with best initial weights matrix W 2,
The error rate Fi_new being calculated;
S308, judge whether the error rate Fi_new of new individual Vi is less than the error rate G1 [i] of parent Ui, if so, life G1 [i]
Equal to Fi_new and A1iReplicate Vi(being so that the gene in i-th of individual Ai is completely as Vi);If it is not, not changing then
G1[i];
S309, judge whether i is equal to N, step S310 is if it is carried out, if not then carrying out step S301;
S310, compare the corresponding error rate G1 [1] of all individuals in the first new population A1, G1 [2] ... G1 [i] ..., G1
[M-1], G1 [M] size selects minimal error rate min G1 [Z3], and find minimal error rate min G1 [Z3] corresponding individualObtain the first population A1Optimum individual
2. the anomalous ecg method for early warning according to claim 1 based on dynamic ECG data, which is characterized in that further include:
The first population A is obtained after having carried out step 31With second of population A2Common optimum individual the step of;
Obtain the first population A1With second of population A2Common optimum individual the step of include:
S41, second of population A is set2Interior each individual is A2i, A2iFor length and XiEqual array, A2iXi is replicated, is A2i=
Xi(1≤i≤M), G2 [i]=Fitness [i] obtain second of population A2The step of include:
S411, gen=gen+1 is calculated;
S412, life i=gen;
S413, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S414, intersection:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~13;Then, judge currently with
The probability P that machine generates3Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein intersect
Step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range (0, M)
In randomly select an integer Z1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1A gene
To m-th gene;
S415, variation:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~14;Then, judge currently with
The probability P that machine generates4Whether mutagenic factor F is less than, if then taking DE/best strategies to new individual ViInto row variation, if not
Then without variation, wherein take DE/best strategies to new individual ViInto row variation:S4151, from integer range (0, M) with
Machine chooses three integer R1、R2And R3;S4152, new individual ViIt is interior indicate gene each data withAnd
It is interior to indicate that each data of gene carry out the operation such as formula (1-1) successively,
In formula, bracket [] indicates the rounding by the way of rounding upResult of calculation, XimnTable
Show new individual ViNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualN-th in interior m-th of gene
A data, if the data volume of each gene is Y, then 1≤m≤M, 1≤n≤Y;
S416, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient decline side
Formula and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolutional Neural network
The best initial weights matrix W 2 of structure C NN2;
S417, verification assessment:Electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN2 with best initial weights matrix W 2,
The error rate Fi_new being calculated;
S418, judge whether the error rate Fi_new of new individual Vi is less than the error rate G2 [i] of parent Ui, if so, life G2 [i]
Equal to Fi_new and order A2iReplicate Vi(it is so that i-th of individual A1iInterior gene is completely and ViEqually);If it is not, then not more
Change G2 [i];
S419, judge whether i is equal to N, step S420 is if it is carried out, if not then carrying out step S411;
S420, compare the first new population A2The corresponding error rate G2 [1] of interior all individuals, G2 [2] ... G2 [i] ..., G2
[n-1], G2 [N] size selects minimal error rate min G2 [Z4], and find minimal error rate min G2 [Z4] corresponding individualObtain second of population A2Global optimum individual
S42, population exchange:The first population A1With second of population A2It is exchanged:Compare the first population A1Minimal error
Rate min G1 [Z3] and second of population A2Minimal error rate min G2 [Z4] size, small min G1 [Z3] or min G2
[Z4] corresponding individual for the first population A1With second of population A2Common optimum individual.
3. a kind of finding the method that error minimal network calculates structure for dynamic ECG data, which is characterized in that including following step
Suddenly:
Step 1, initialization:
S11, each individual Xi is initialized, the array that Xi forms in containing N number of gene is individual equipped with M Xi, 1≤i≤M, often
There are three serial data in a gene, a data string indicates that calculating formula, another two serial data indicate the node connected forward, expression
It is for indicating serial data expression or do not express 0 or 1 to have one-bit digital in each serial data of node;
S12, initialization intersect factor CR:CR arbitrary floating-point values in taking 0~1;
S13, initialization mutagenic factor F:F arbitrary floating-point values in taking 0~1;
S14, initialization algebraically gen, order gen=0;
Step 2 calculates the error Fitness [i] of first generation individual Xi:
S21, life i=1;
S22, by new individual ViGene data be decoded into a convolutional neural networks structure C NN1;
S23, by gradient decline in the way of and current electrocardiogram (ECG) data in convolutional Neural network configuration CNN1 calculating formula carry out weights
Iteration convergence, obtain the best initial weights matrix W 1 of convolutional Neural network configuration CNN1;
S24, current electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN1 with best initial weights matrix W 1, is calculated
The error rate F gone outi_new1;
S25, life Fitness [i] are equal to Fi_new1;
S26, i=i+1;
S27, judge whether i is more than M, if step S22 is otherwise carried out, if carrying out step 3;
Step 3: setting second of population A2Interior each individual is A2i, A2iFor length and XiEqual array, A2iXi is replicated, is
A2i=Xi(1≤i≤M), G2 [i]=Fitness [i] obtain second of population A2The step of include:
S411, gen=gen+1 is calculated;
S412, life i=gen;
S413, new individual Vi is set as length and XiEqual array, Vi replicate Xi, are Vi=Xi;
S414, intersection:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~13;Then, judge currently with
The probability P that machine generates3Whether it is less than and intersects factor CR, if then being intersected, if not then without intersecting, wherein intersect
Step is:S3041, an integer R is randomly selected from integer range (1, M), and R is not equal to i;S3042, from integer range (0, M)
In randomly select an integer Z1;S3043, by XRIn Z1A gene is all copied to V to m-th geneiZ1A gene
To m-th gene;
S415, variation:First, it is the probability P currently randomly generated to take arbitrary floating-point values in 0~14;Then, judge currently with
The probability P that machine generates4Whether mutagenic factor F is less than, if then taking DE/best strategies to new individual ViInto row variation, if not
Then without variation, wherein take DE/best strategies to new individual ViInto row variation:S4151, from integer range (0, M) with
Machine chooses three integer R1、R2And R3;S4152, new individual ViIt is interior indicate gene each data withAnd
It is interior to indicate that each data of gene carry out the operation such as formula (1-1) successively,
In formula, bracket [] indicates the rounding by the way of rounding upResult of calculation, XimnTable
Show new individual ViNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualNth data in interior m-th of gene,Indicate new individualN-th in interior m-th of gene
Data, if the data volume of each gene is Y, then 1≤m≤M, 1≤n≤Y;
S416, by new individual ViGene data be decoded into a convolutional neural networks structure C NN2, then utilize gradient decline side
Formula and electrocardiogram (ECG) data carry out calculating formula in convolutional Neural network configuration CNN2 the iteration convergence of weights, obtain convolutional Neural network
The best initial weights matrix W 2 of structure C NN2;
S417, verification assessment:Electrocardiogram (ECG) data is input in the convolutional Neural network configuration CNN2 with best initial weights matrix W 2,
The error rate Fi_new being calculated;
S418, judge whether the error rate Fi_new of new individual Vi is less than the error rate G2 [i] of parent Ui, if so, life G2 [i]
Equal to Fi_new and order A2iReplicate Vi(it is so that i-th of individual A1iInterior gene is completely and ViEqually);If it is not, then not more
Change G2 [i];
S419, judge whether i is equal to N, step S420 is if it is carried out, if not then carrying out step S411;
S420, compare the first new population A2The corresponding error rate G2 [1] of interior all individuals, G2 [2] ... G2 [i] ..., G2
[n-1], G2 [N] size selects minimal error rate min G2 [Z4], and find minimal error rate min G2 [Z4] corresponding individualObtain second of population A2Global optimum individual
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810191992.9A CN108470158B (en) | 2018-03-08 | 2018-03-08 | Method for searching error minimum network computing structure for dynamic ECG data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810191992.9A CN108470158B (en) | 2018-03-08 | 2018-03-08 | Method for searching error minimum network computing structure for dynamic ECG data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108470158A true CN108470158A (en) | 2018-08-31 |
CN108470158B CN108470158B (en) | 2020-05-12 |
Family
ID=63265150
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810191992.9A Active CN108470158B (en) | 2018-03-08 | 2018-03-08 | Method for searching error minimum network computing structure for dynamic ECG data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108470158B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101972145A (en) * | 2010-10-12 | 2011-02-16 | 华南理工大学 | Fetus electrocardio blind separation method based on relative sparsity of time domain of source signal |
CN103473599A (en) * | 2013-10-11 | 2013-12-25 | 北京航空航天大学 | Genetic algorithm and Kalman filtering based RBFN (Radial Basis Function Networks) combined training method |
CN104463326A (en) * | 2014-12-09 | 2015-03-25 | 浙江大学 | Standard 12-lead ecg signal reconstruction method |
CN105595990A (en) * | 2016-01-27 | 2016-05-25 | 浙江铭众科技有限公司 | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal |
CN105726013A (en) * | 2016-01-27 | 2016-07-06 | 浙江铭众科技有限公司 | Electrocardiogram monitoring system with electrocardiosignal quality discrimination function |
CN105796091A (en) * | 2016-02-29 | 2016-07-27 | 浙江铭众科技有限公司 | Intelligent terminal for removing electrocardiosignal vehicle motion noise |
CN106344005A (en) * | 2016-10-28 | 2017-01-25 | 张珈绮 | Mobile ECG (electrocardiogram) monitoring system and monitoring method |
CN107122788A (en) * | 2017-03-09 | 2017-09-01 | 吉林大学 | A kind of personal identification method and device based on electrocardiosignal |
CN107203692A (en) * | 2017-05-09 | 2017-09-26 | 哈尔滨工业大学(威海) | The implementation method of atrial fibrillation detection based on depth convolutional neural networks |
CN107516075A (en) * | 2017-08-03 | 2017-12-26 | 安徽华米信息科技有限公司 | Detection method, device and the electronic equipment of electrocardiosignal |
-
2018
- 2018-03-08 CN CN201810191992.9A patent/CN108470158B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101972145A (en) * | 2010-10-12 | 2011-02-16 | 华南理工大学 | Fetus electrocardio blind separation method based on relative sparsity of time domain of source signal |
CN103473599A (en) * | 2013-10-11 | 2013-12-25 | 北京航空航天大学 | Genetic algorithm and Kalman filtering based RBFN (Radial Basis Function Networks) combined training method |
CN104463326A (en) * | 2014-12-09 | 2015-03-25 | 浙江大学 | Standard 12-lead ecg signal reconstruction method |
CN105595990A (en) * | 2016-01-27 | 2016-05-25 | 浙江铭众科技有限公司 | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal |
CN105726013A (en) * | 2016-01-27 | 2016-07-06 | 浙江铭众科技有限公司 | Electrocardiogram monitoring system with electrocardiosignal quality discrimination function |
CN105796091A (en) * | 2016-02-29 | 2016-07-27 | 浙江铭众科技有限公司 | Intelligent terminal for removing electrocardiosignal vehicle motion noise |
CN106344005A (en) * | 2016-10-28 | 2017-01-25 | 张珈绮 | Mobile ECG (electrocardiogram) monitoring system and monitoring method |
CN107122788A (en) * | 2017-03-09 | 2017-09-01 | 吉林大学 | A kind of personal identification method and device based on electrocardiosignal |
CN107203692A (en) * | 2017-05-09 | 2017-09-26 | 哈尔滨工业大学(威海) | The implementation method of atrial fibrillation detection based on depth convolutional neural networks |
CN107516075A (en) * | 2017-08-03 | 2017-12-26 | 安徽华米信息科技有限公司 | Detection method, device and the electronic equipment of electrocardiosignal |
Also Published As
Publication number | Publication date |
---|---|
CN108470158B (en) | 2020-05-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107368887B (en) | Deep memory convolutional neural network device and construction method thereof | |
Kung et al. | A power-aware digital feedforward neural network platform with backpropagation driven approximate synapses | |
Mielniczuk et al. | Stopping rules for mutual information-based feature selection | |
CN108920895A (en) | A kind of incidence relation prediction technique of circular rna and disease | |
Zhang et al. | Covariate adaptive false discovery rate control with applications to omics-wide multiple testing | |
CN111832725A (en) | Multi-robot multi-task allocation method and device based on improved genetic algorithm | |
Peres da Silva et al. | TUGDA: task uncertainty guided domain adaptation for robust generalization of cancer drug response prediction from in vitro to in vivo settings | |
Siewert-Rocks et al. | Leveraging gene co-regulation to identify gene sets enriched for disease heritability | |
CN110517728A (en) | A kind of gene order comparison method and device | |
CN108470158A (en) | A method of it finding error minimal network for dynamic ECG data and calculates structure | |
Ludl et al. | Comparison between instrumental variable and mediation-based methods for reconstructing causal gene networks in yeast | |
Choi et al. | New variable selection strategy for analysis of high-dimensional DNA methylation data | |
US7864577B2 (en) | Sharing physical memory locations in memory devices | |
Bao et al. | A novel genetic algorithm with cell crossover for circuit design optimization | |
CN115908909A (en) | Evolutionary neural architecture searching method and system based on Bayes convolutional neural network | |
CN109740804B (en) | Water quality prediction method, terminal device and computer-readable storage medium | |
Navamajiti et al. | McBel-Plnc: a deep learning model for multiclass multilabel classification of protein-lncRNA interactions | |
CN110147804B (en) | Unbalanced data processing method, terminal and computer readable storage medium | |
Karlebach | Inferring Boolean network states from partial information | |
US20180365350A1 (en) | Generating circuits | |
Choi et al. | Early termination of STDP learning with spike counts in spiking neural networks | |
US20220164637A1 (en) | Neural Processing Units (NPUs) and Computational Systems Employing the Same | |
KR102659917B1 (en) | Method for developing meta-gene based on non-negative matrix factorization and applications thereof | |
CN111985645A (en) | Neural network training method and device, electronic equipment and storage medium | |
CN116227239B (en) | Acceleration test data analysis method, device and equipment based on gray prediction model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |