CN108470158B - Method for searching error minimum network computing structure for dynamic ECG data - Google Patents

Method for searching error minimum network computing structure for dynamic ECG data Download PDF

Info

Publication number
CN108470158B
CN108470158B CN201810191992.9A CN201810191992A CN108470158B CN 108470158 B CN108470158 B CN 108470158B CN 201810191992 A CN201810191992 A CN 201810191992A CN 108470158 B CN108470158 B CN 108470158B
Authority
CN
China
Prior art keywords
data
gene
individual
new
equal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810191992.9A
Other languages
Chinese (zh)
Other versions
CN108470158A (en
Inventor
高英
罗彭婷
王禹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201810191992.9A priority Critical patent/CN108470158B/en
Publication of CN108470158A publication Critical patent/CN108470158A/en
Application granted granted Critical
Publication of CN108470158B publication Critical patent/CN108470158B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Physiology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Complex Calculations (AREA)

Abstract

The invention provides a method for searching an error minimum network computing structure for dynamic ECG data, which comprises the following steps: step one, initialization: s11, initializing each individual Xi; s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1; s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1; s14, initializing algebraic gen, wherein nominal gen is 0; step two, calculating the error Fitness [ i ] of the individual Xi]: s21, where i is 1; s22, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN 1; s23, iterative convergence of weights is carried out on the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data, and an optimal weight matrix W1 of the convolutional neural network structure CNN1 is obtained. The method for searching the network computing structure with the minimum error for the dynamic ECG data solves the problem that misjudgment occurs during the abnormal detection of the electrocardiosignals because the same network computing structure is set for the electrocardio data of each person in the prior art.

Description

Method for searching error minimum network computing structure for dynamic ECG data
Technical Field
The invention relates to processing of dynamic ECG data, in particular to a method for searching an error minimization network calculation structure for dynamic ECG data.
Background
Because the electrocardiogram data of each person is different, the network computing structure which is most suitable for the electrocardiogram data of each person is also different. In the prior art, a certain network computing structure is arranged for detecting the occurrence of abnormal electrocardiosignal signals, so that the phenomena of misjudgment and the like can occur.
Disclosure of Invention
The invention provides a method for searching a network computing structure with the minimum error for dynamic ECG data, which solves the problem of misjudgment during electrocardiosignal abnormal detection caused by setting the same network computing structure for electrocardio data of each person in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention firstly provides a method for searching a network computing structure with minimum error for dynamic ECG data, which comprises the following steps:
step one, initialization:
s11, initializing each individual Xi, wherein Xi is an array comprising N genes, M Xi individuals are arranged, i is more than or equal to 1 and is less than or equal to M, three data strings are arranged in each gene, one data string represents a calculation formula, the other two data strings represent nodes connected forwards, and one digit in each data string of an expression node is 0 or 1 for representing the expression or non-expression of the data string;
s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1;
s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1;
s14, initializing algebraic gen, wherein nominal gen is 0;
step two, calculating the error Fitness [ i ] of the individual Xi:
s21, where i is 1;
s22, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN 1;
s23, carrying out iterative convergence on the weight of the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data to obtain an optimal weight matrix W1 of the convolutional neural network structure CNN 1;
s24, inputting the current electrocardiogram data into the convolutional neural network structure CNN1 with the optimal weight matrix W1, and calculating the error rate Fi_new1；
S25, MinFitness [ i ]]Is equal to Fi_new1；
S26、i＝i+1；
S27, judging whether i is larger than M, if not, performing S22, and if so, performing a third step;
step three, setting a first population A1Each individual is A1i，A1iIs length and XiEqual array, A1iReplication Xi, i.e. A1i＝Xi(1≤i≤M)，G1[i]＝Fitness[i]Obtaining a first population A1The step of optimizing the individual of (a) comprises:
s301, calculating gen + 1;
s302, where, i ═ gen;
s303, setting Vi of new individual as length and XiThe equivalent array, Vi, copies Xi, i.e. is Vi＝Xi；
S304, crossing: firstly, taking any floating point number in 0-1 as the probability p generated randomly at present1(ii) a Then, the probability p of random generation at present is judged1If the cross factor is smaller than the cross factor CR, if so, performing cross, otherwise, not performing cross, wherein the cross step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1Genes up to Mth gene;
s305, mutation: firstly, taking any floating point number in 0-1 as the probability p generated randomly at present2(ii) a Then, the probability p of random generation at present is judged2If the number of the new individuals is less than the variation factor F, adopting a DE/rand strategy to carry out comparison on the new individuals V if the number of the new individuals is less than the variation factor FiPerforming mutation, if not, wherein a DE/rand strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s3051, randomly selecting an integer Z from an integer range (0, M)2(ii) a S3052, randomly selecting M-Z2+1 gene length data set C, the data string corresponding to each gene in the data set C conforming to the regulation set for the gene data string; s3053, New individual ViInner Z2The character strings of the genes from the Mth gene to the Mth gene are sequentially changed into data in a data group C;
s306, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s307, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
S308judging whether the error rate Fi _ new of the new individual Vi is smaller than the error rate G1[ i ] of the parent Ui]If yes, command G1[ i]Equal to Fi _ new and A1iReplication Vi(i.e., so that the genes in the ith individual Ai are exactly the same as Vi); if not, G1[ i ] is not changed]；
S309, judging whether i is equal to N, if so, performing the step S310, and if not, performing the step S301;
s310, comparing the error rates G1[1 ] corresponding to all individuals in the first new population A1]、G1[2]、……G1[i]……、G1[M-1]、G1[M]Size, selecting the minimum error rate min G1[ Z ]3]And find the minimum error rate min G1[ Z ]3]Corresponding individual
Figure BDA0001591975640000031
Thus obtaining a first population A1Of the optimal individual
Figure BDA0001591975640000032
The present invention also provides a method for finding a network computation structure with minimal error for dynamic ECG data, comprising the steps of:
step one, initialization:
s11, initializing each individual Xi, wherein Xi is an array comprising N genes, M Xi individuals are arranged, i is more than or equal to 1 and is less than or equal to M, three data strings are arranged in each gene, one data string represents a calculation formula, the other two data strings represent nodes connected forwards, and one digit in each data string of an expression node is 0 or 1 for representing the expression or non-expression of the data string;
s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1;
s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1;
s14, initializing algebraic gen, wherein nominal gen is 0;
step two, calculating the error Fitness [ i ] of the first generation individual Xi:
s21, where i is 1;
s22, adding the new individual ViDecoding the gene data into a convolution spiritVia network fabric CNN 1;
s23, carrying out iterative convergence on the weight of the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data to obtain an optimal weight matrix W1 of the convolutional neural network structure CNN 1;
s24, inputting the current electrocardiogram data into the convolutional neural network structure CNN1 with the optimal weight matrix W1, and calculating the error rate Fi_new1；
S25, MinFitness [ i ]]Is equal to Fi_new1；
S26、i＝i+1；
S27, judging whether i is larger than M, if not, performing S22, and if so, performing a third step;
step three, setting a second species group A2Each individual is A2i，A2iIs length and XiEqual array, A2iReplication Xi, i.e. A2i＝Xi(1≤i≤M)，G2[i]＝Fitness[i]Obtaining a second species A2Comprises the following steps:
s411, calculating gen as gen + 1;
s412, i ═ gen;
s413, setting Vi of new individual as length and XiThe equivalent array, Vi, copies Xi, i.e. is Vi＝Xi；
S414, crossing: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present3(ii) a Then, the probability P of random generation at present is judged3If the cross factor is smaller than the cross factor CR, if so, performing cross, otherwise, not performing cross, wherein the cross step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1Genes up to Mth gene;
s415, mutation: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present4(ii) a Then, the probability P of random generation at present is judged4Whether or not toIf the variation factor is smaller than the variation factor F, adopting a DE/best strategy to carry out on the new individual ViPerforming mutation, if not, wherein a DE/best strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s4151 randomly selecting three integers R from the range of integers (0, M)1、R2And R3(ii) a S4152 New Individual ViData of each gene represented therein and
Figure BDA0001591975640000041
and
Figure BDA0001591975640000042
the data of the internal expression gene are sequentially operated as the formula (1-1),
Figure BDA0001591975640000043
wherein, the middle bracket 2]The representation is rounded by rounding
Figure BDA0001591975640000044
Calculated result of (A), XimnRepresenting a new individual ViThe nth data in the mth gene,
Figure BDA0001591975640000045
representing a new individual
Figure BDA0001591975640000046
The nth data in the mth gene,
Figure BDA0001591975640000047
representing a new individual
Figure BDA0001591975640000048
The nth data in the mth gene,
Figure BDA0001591975640000049
representing a new individual
Figure BDA00015919756400000410
The nth data in the mth gene is set asThe data volume of each gene is Y, so that M is more than or equal to 1 and less than or equal to M, and n is more than or equal to 1 and less than or equal to Y;
s416, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s417, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s418, judging whether the error rate Fi _ new of the new individual Vi is smaller than the error rate G2[ i ] of the parent Ui]If yes, command G2[ i]Equal to Fi _ new and hit A2iReplication Vi(i.e. so that the ith individual A1iThe internal gene is completely linked with ViThe same); if not, G2[ i ] is not changed]；
S419, judging whether i is equal to N, if so, performing step S420, and if not, performing step S411;
s420, comparing the first new population A2Error rates G2[1 ] corresponding to all individuals]、G2[2]、……G2[i]……、G2[n-1]、G2[N]Size, selecting the minimum error rate min G2[ Z ]4]And find the minimum error rate min G2[ Z ]4]Corresponding individual
Figure BDA00015919756400000411
Obtaining a second species A2Global optimal individual of (2)
Figure BDA00015919756400000412
Compared with the prior art, the invention has the following beneficial effects:
the method realizes the calculation steps of changing each calculation formula in the convolutional neural network structure at any time, judges the various convolutional neural network structures by using the currently acquired electrocardio data and electrocardio data results, calculates the error between the judgment result of each convolutional neural network structure and the real electrocardio data result, obtains the most suitable convolutional neural network structure through the error comparison between each convolutional neural network structure, enables the current individual electrocardio data to be more adaptively judged, and avoids the occurrence of misjudgment results.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a schematic diagram of a convolutional neural network structure;
FIG. 2 shows an individual Xi formed by the convolutional neural network structure of FIG. 1.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the functions of the invention clearer and easier to understand, the invention is further described by combining the following specific embodiments:
as shown in fig. 1 and fig. 2, the present invention provides a method for finding an error minimization network calculation structure for ECG data, comprising the following steps:
step one, initialization:
s11, initializing each individual Xi (each individual Xi corresponds to a convolutional neural network structure), wherein each Xi is an array comprising N genes, M Xi individuals are arranged, i is more than or equal to 1 and less than or equal to M, three data strings are arranged in each gene, one data string represents a calculation formula, the other two data strings represent nodes connected forwards, and one digit in each data string of an expression node is 0 or 1 for representing the expression or non-expression of the data string; for example:
the main sub-elements stored (calculation 18, then 0 to 17 can be used to represent each calculation) include ConvBlock, ResBlock, Max firing, Average firing, Summation, collocation, input, output. Where ConvBlock and ResBlock both include three outputs 32, 64, 128 and there are two convolution kernels, 3 x 3 and 5 x 5 respectively. In the figure, since input is definitely the first gene, input in the genes is represented by 0, 7 genes are shown in fig. 2, each of three boxes represents one gene, and the first box is a code number (also referred to as "code number") for representing the present calculation formulaWhich is a specific calculation formula in the database), the first digit in the second small box represents whether the data is expressed or not, the second digit is the serial number of the calculation formula connected to the previous calculation formula (i.e. the gene in Xi), the first digit in the third small box represents whether the data is expressed or not, and the second digit represents the serial number of the calculation formula connected to the previous calculation formula; the code numbers of the respective calculation formulas in the database are as follows: input has a calculation formula number of 0, conv (64, 3) (ConvBlock with convolution kernel of 3X 3 output of 64) has a calculation formula number of 1, pool (avg) has a calculation formula number of 2, conv (32, 5) (ConvBlock with convolution kernel of 5X 5 output of 32) has a calculation formula number of 3, conv (32, 3) (ConvBlock with convolution kernel of 3X 3 output of 32) has a calculation formula number of 6, sum has a calculation formula number of 4, output has a calculation formula number of 5, and one of the individual X isiData set with 7 genes, Xi＝(011001 021001 031102 060112 041314 030113051506)，XiA data set of 7 genes, each gene having five data, the second number in each gene being 0 indicating that the fourth number is not expressed, 1 indicating that the fourth number is expressed, the fourth number in each gene being 0 indicating that the fourth number is not expressed, 1 indicating that the fourth number is expressed; then each individual Xi has 6 x 7 digits;
s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1;
s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1;
s14, initializing algebraic gen, wherein nominal gen is 0;
step two, calculating the error Fitness [ i ] of the individual Xi:
s21, where i is 1;
s22, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN 1;
s23, carrying out iterative convergence on the weight of the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data (including a detection result), and obtaining an optimal weight matrix W1 of the convolutional neural network structure CNN 1;
s24, comparing the current electrocardioInputting data into a convolutional neural network structure CNN1 with an optimal weight matrix W1, and calculating an error rate Fi_new1；
S25, MinFitness [ i ]]Is equal to Fi_new1；
S26、i＝i+1；
S27, judging whether i is larger than M, if not, performing S22, and if so, performing a third step;
step three, setting a first population A1Each individual is A1i，A1iIs length and XiEqual array, A1iReplication Xi, i.e. A1i＝Xi(1. ltoreq. i.ltoreq.M, i is 1 to M and is copied), G1[ i]＝Fitness[i]Obtaining a first population A1The step of optimizing the individual of (a) comprises:
s301, calculating gen + 1;
s302, where, i ═ gen;
s303, setting Vi of new individual as length and XiThe equivalent array, Vi, copies Xi, i.e. is Vi＝Xi(i takes 1 to M and copies);
s304, crossing: firstly, taking any floating point number in 0-1 as the probability p generated randomly at present1(improving randomness); then, the probability p of random generation at present is judged1Whether the cross factor is smaller than a cross factor CR (providing a basis for some subsequent non-crossing), if so, crossing is carried out, otherwise, the crossing is not carried out, wherein the crossing step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1From gene to gene M (e.g., V in FIG. 2)iAn X with equal length design and each bit number according to a modular setting ruleRSo that data ViIs an array XR) (7 genes, as shown in FIG. 2 and FIG. 1, any integer from 1 to 7, such as Z, is selected1Equal to 5, then V will beiFrom the 5 th gene to the 7 th gene of (A) by XRFrom gene 5 to gene 7Thus);
s305, mutation: firstly, taking any floating point number in 0-1 as the probability p generated randomly at present2(ii) a Then, the probability p of random generation at present is judged2If the number of the new individuals is less than the variation factor F, adopting a DE/rand strategy to carry out comparison on the new individuals V if the number of the new individuals is less than the variation factor FiPerforming mutation, if not, wherein a DE/rand strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s3051, randomly selecting an integer Z from an integer range (0, M)2(ii) a S3052, randomly selecting M-Z2+1 gene length data set C, the data string corresponding to each gene in the data set C conforming to the regulation set for the gene data string; s3053, New individual ViInner Z2The character strings of the genes from the Mth gene to the Mth gene are sequentially changed into data in a data group C; (in FIG. 2 and FIG. 1, if there are 7 genes, any integer from 1 to 7 is selected, for example, Z2Equal to 6, then a 10-bit array is randomly generated (4150450216) and V is assignediThe data of the 6 th gene and the 7 th gene in (4) are replaced with a 10-bit array (4150450216)
S306, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s307, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s308, judging whether the error rate Fi _ new of the new individual Vi is smaller than the error rate G1[ i ] of the parent Ui]If yes, command G1[ i]Equal to Fi _ new and A1iReplication Vi(i.e., so that the genes in the ith individual Ai are exactly the same as Vi); if not, G1[ i ] is not changed]；
S309, judging whether i is equal to N, if so, performing the step S310, and if not, performing the step S301;
s310, comparing the error rates G1[1 ] corresponding to all individuals in the first new population A1]、G1[2]、……G1[i]……、G1[M-1]、G1[M]Size, select the smallestError rate min G1[ Z ]3]And find the minimum error rate min G1[ Z ]3]Corresponding individual
Figure BDA0001591975640000071
Thus obtaining a first population A1Of the optimal individual
Figure BDA0001591975640000072
In order to calculate a wider variety of convolutional neural network structures, so that the convolutional neural network structures that cannot be found by the first group are found out in each convolutional neural network structure according to different rules, and the accuracy of the found convolutional neural network structures is further improved, the method further comprises the following steps: obtaining a first population A after the third step1And a second species A2A step of co-optimizing the individuals;
obtaining a first population A1And a second species A2The step of co-optimizing the individuals comprises:
s41, setting a second species A2Each individual is A2i，A2iIs length and XiEqual array, A2iReplication Xi, i.e. A2i＝Xi(1≤i≤M)，G2[i]＝Fitness[i]Obtaining a second species A2Comprises the following steps:
s411, calculating gen as gen + 1;
s412, i ═ gen;
s413, setting Vi of new individual as length and XiThe equivalent array, Vi, copies Xi, i.e. is Vi＝Xi；
S414, crossing: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present3(ii) a Then, the probability P of random generation at present is judged3If the cross factor is smaller than the cross factor CR, if so, performing cross, otherwise, not performing cross, wherein the cross step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1Gene to MthAll genes are replicated to ViZ of (a)1Genes up to Mth gene;
s415, mutation: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present4(ii) a Then, the probability P of random generation at present is judged4If the number of the new individuals V is less than the variation factor F, adopting a DE/best strategy to carry out comparison on the new individuals V if the number of the new individuals V is less than the variation factor FiPerforming mutation, if not, wherein a DE/best strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s4151 randomly selecting three integers R from the range of integers (0, M)1、R2And R3(ii) a S4152 New Individual ViData of each gene represented therein and
Figure BDA0001591975640000081
and
Figure BDA0001591975640000082
the data of the internal expression gene are sequentially operated as the formula (1-1),
Figure BDA0001591975640000083
wherein, the middle bracket 2]The representation is rounded by rounding
Figure BDA0001591975640000084
Calculated result of (A), XimnRepresenting a new individual ViThe nth data in the mth gene,
Figure BDA0001591975640000085
representing a new individual
Figure BDA0001591975640000086
The nth data in the mth gene,
Figure BDA0001591975640000087
representing a new individual
Figure BDA0001591975640000088
The nth data in the mth gene,
Figure BDA0001591975640000089
representing a new individual
Figure BDA00015919756400000810
Setting the data quantity of each gene as Y for the nth data in the mth gene, wherein M is more than or equal to 1 and less than or equal to M, and n is more than or equal to 1 and less than or equal to Y; (in FIG. 2 and FIG. 1, if there are 7 genes, then each ViAre arrays of 6 x 7 digits,
Figure BDA00015919756400000811
Figure BDA00015919756400000812
Figure BDA00015919756400000813
f is 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, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s417, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s418, judging whether the error rate Fi _ new of the new individual Vi is smaller than the error rate G2[ i ] of the parent Ui]If yes, command G2[ i]Equal to Fi _ new and hit A2iReplication Vi(i.e. so that the ith individual A1iThe internal gene is completely linked with ViThe same); if not, G2[ i ] is not changed]；
S419, judging whether i is equal to N, if so, performing step S420, and if not, performing step S411;
s420, comparing the first new population A2Error rates G2[1 ] corresponding to all individuals]、G2[2]、……G2[i]……、G2[n-1]、G2[N]Size, selecting the minimum error rate min G2[ Z ]4]And find the minimum error rate min G2[ Z ]4]Corresponding individual AZ4Obtaining a second species A2Global optimal individual A ofZ4；
S42, population exchange: first population A1And a second species A2Carrying out communication: comparing the first population A1Min G1[ Z ] minimum error rate3]And a second species A2Min G2[ Z ] minimum error rate4]Size of (d), small min G1[ Z ]3]Or minG2[ Z ]4]The corresponding individual is a first kind of population A1And a second species A2To the co-optimal individual.
The present embodiment also provides a method for finding a network computation structure with minimal error for ECG data, comprising the steps of:
step one, initialization:
s11, initializing each individual Xi, wherein Xi is an array comprising N genes, M Xi individuals are arranged, i is more than or equal to 1 and is less than or equal to M, three data strings are arranged in each gene, one data string represents a calculation formula, the other two data strings represent nodes connected forwards, and one digit in each data string of an expression node is 0 or 1 for representing the expression or non-expression of the data string;
s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1;
s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1;
s14, initializing algebraic gen, wherein nominal gen is 0;
step two, calculating the error Fitness [ i ] of the first generation individual Xi:
s21, where i is 1;
s22, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN 1;
s23, carrying out iterative convergence on the weight of the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data to obtain an optimal weight matrix W1 of the convolutional neural network structure CNN 1;
s24, inputting the current electrocardiogram data into the convolutional neural network structure CNN1 with the optimal weight matrix W1, and calculating the error rate Fi_new1；
S25, MinFitness [ i ]]Is equal to Fi_new1；
S26、i＝i+1；
S27, judging whether i is larger than M, if not, performing S22, and if so, performing a third step;
step three, setting a second species group A2Each individual is A2i，A2iIs length and XiEqual array, A2iReplication Xi, i.e. A2i＝Xi(1≤i≤M)，G2[i]＝Fitness[i]Obtaining a second species A2Comprises the following steps:
s411, calculating gen as gen + 1;
s412, i ═ gen;
s413, setting Vi of new individual as length and XiThe equivalent array, Vi, copies Xi, i.e. is Vi＝Xi；
S414, crossing: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present3(ii) a Then, the probability P of random generation at present is judged3If the cross factor is smaller than the cross factor CR, if so, performing cross, otherwise, not performing cross, wherein the cross step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1Genes up to Mth gene;
s415, mutation: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present4(ii) a Then, the probability P of random generation at present is judged4If the number of the new individuals V is less than the variation factor F, adopting a DE/best strategy to carry out comparison on the new individuals V if the number of the new individuals V is less than the variation factor FiPerforming mutation, if not, wherein a DE/best strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s4151 randomly selecting three integers R from the range of integers (0, M)1、R2And R3(ii) a S4152 New Individual ViData of each gene represented therein and
Figure BDA0001591975640000101
and
Figure BDA0001591975640000102
the data of the internal expression gene are sequentially operated as the formula (1-1),
Figure BDA0001591975640000103
wherein, the middle bracket 2]The representation is rounded by rounding
Figure BDA0001591975640000104
Calculated result of (A), XimnRepresenting a new individual ViThe nth data in the mth gene,
Figure BDA0001591975640000105
representing a new individual
Figure BDA0001591975640000106
The nth data in the mth gene,
Figure BDA0001591975640000107
representing a new individual
Figure BDA0001591975640000108
The nth data in the mth gene,
Figure BDA0001591975640000109
representing a new individual
Figure BDA00015919756400001010
Setting the data quantity of each gene as Y for the nth data in the mth gene, wherein M is more than or equal to 1 and less than or equal to M, and n is more than or equal to 1 and less than or equal to Y;
s416, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then calculating the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram dataIteratively converging the weights to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s417, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s418, judging whether the error rate Fi _ new of the new individual Vi is smaller than the error rate G2[ i ] of the parent Ui]If yes, command G2[ i]Equal to Fi _ new and hit A2iReplication Vi(i.e. so that the ith individual A1iThe internal gene is completely linked with ViThe same); if not, G2[ i ] is not changed]；
S419, judging whether i is equal to N, if so, performing step S420, and if not, performing step S411;
s420, comparing the first new population A2Error rates G2[1 ] corresponding to all individuals]、G2[2]、……G2[i]……、G2[n-1]、G2[N]Size, selecting the minimum error rate min G2[ Z ]4]And find the minimum error rate min G2[ Z ]4]Corresponding individual AZ4Obtaining a second species A2Global optimal individual A ofZ4。
Finally, the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.

Claims (3)

1. A method for finding a network computation structure of minimum errors for ambulatory ECG data, comprising the steps of:
step one, initialization:
s11, initializing each individual Xi，XiFor an array containing N genes, M X genes are providediIndividuals, i is more than or equal to 1 and less than or equal to M, three data strings are arranged in each gene, one data string represents a calculation formula, and the other two data strings represent nodes connected forwards for expressionEach data string of the node has a digit which is 0 or 1 for representing the expression or non-expression of the data string;
s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1;
s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1;
s14, initializing algebra gen, wherein gen = 0;
step two, calculating an individual XiError of [1 ] Fitness [ i ]]：
S21, nominal = 1;
s22, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN 1;
s23, carrying out iterative convergence on the weight of the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data to obtain an optimal weight matrix W1 of the convolutional neural network structure CNN 1;
s24, inputting the current electrocardiogram data into the convolutional neural network structure CNN1 with the optimal weight matrix W1, and calculating the error rate Fi_new1；
S25, MinFitness [ i ]]Is equal to Fi_new1；
S26、i=i+1；
S27, judging whether i is larger than M, if not, performing S22, and if so, performing a third step;
step three, setting a first population A1Each individual is A1i，A1iIs length and XiEqual array, A1iReplication of XiIs namely A1i=XiAnd 1. ltoreq. i.ltoreq.M, G1[ i]= Fitness[i]Obtaining a first population A1The step of optimizing the individual of (a) comprises:
s301, calculating gen = gen + 1;
s302, life = gen;
s303, setting a new individual ViIs length and XiEqual array, ViReplication of XiI.e. is Vi=Xi；
S304, crossing: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present1(ii) a Then, it makes an judgmentBreaking the current randomly generated probability P1If the cross factor is smaller than the cross factor CR, if so, performing cross, otherwise, not performing cross, wherein the cross step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1Genes up to Mth gene;
s305, mutation: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present2(ii) a Then, the probability P of random generation at present is judged2If the number of the new individuals is less than the variation factor F, adopting a DE/rand strategy to carry out comparison on the new individuals V if the number of the new individuals is less than the variation factor FiPerforming mutation, if not, wherein a DE/rand strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s3051, randomly selecting an integer Z from an integer range (0, M)2(ii) a S3052, randomly selecting M-Z2+1 gene length data set C, the data string corresponding to each gene in the data set C conforming to the regulation set for the gene data string; s3053, New individual ViInner Z2The character strings of the genes from the Mth gene to the Mth gene are sequentially changed into data in a data group C;
s306, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s307, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s308, judging a new individual ViWhether the error rate Fi _ new of (a) is less than the error rate G1[ i ] of the parent Ui]If yes, command G1[ i]Equal to Fi _ new and A1iReplication ViWherein A is1iReplication ViI.e. such that the ith individual A1iThe internal gene is completely linked with ViThe same is true; if not, G1[ i ] is not changed]；
S309, judging whether i is equal to N, if so, performing the step S310, and if not, performing the step S301;
s310, comparing the error rates G1[1 ] corresponding to all individuals in the first new population A1]、G1[2]、……G1[i]……、G1[M-1]、G1[M]Size, selecting the minimum error rate min G1[ Z ]3]And find the minimum error rate min G1[ Z ]3]Corresponding individual
Figure 416517DEST_PATH_IMAGE001
Obtaining a first population A1Of the optimal individual
Figure 578508DEST_PATH_IMAGE001
。
2. The method of claim 1, further comprising: obtaining a first population A after the third step1And a second species A2A step of co-optimizing the individuals;
obtaining a first population A1And a second species A2The step of co-optimizing the individuals comprises:
s41, setting a second species A2Each individual is A2i，A2iIs length and XiEqual array, A2iReplication of XiIs namely A2i=XiAnd 1. ltoreq. i.ltoreq.M, G2[ i]= Fitness[i]Obtaining a second species A2Comprises the following steps:
s411, calculating gen = gen + 1;
s412, life = gen;
s413, setting a new individual ViIs length and XiEqual array, ViReplication of XiI.e. is Vi=Xi；
S414, crossing: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present3(ii) a Then, the probability P of random generation at present is judged3If the cross factor is smaller than the cross factor CR, if so, the crossing is carried out, otherwise, the crossing is not carried outAnd crossing, wherein the crossing step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1Genes up to Mth gene;
s415, mutation: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present4(ii) a Then, the probability P of random generation at present is judged4If the number of the new individuals V is less than the variation factor F, adopting a DE/best strategy to carry out comparison on the new individuals V if the number of the new individuals V is less than the variation factor FiPerforming mutation, if not, wherein a DE/best strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s4151 randomly selecting three integers R from the range of integers (0, M)1、R2And R3(ii) a S4152 New Individual ViData of each gene represented therein and
Figure 640005DEST_PATH_IMAGE002
、
Figure 924356DEST_PATH_IMAGE003
and
Figure 336883DEST_PATH_IMAGE004
the data of the internal expression gene are sequentially operated as the formula (1-1),
Figure 36986DEST_PATH_IMAGE005
(1-1), wherein the parenthesis [ alpha ], []The representation is rounded by rounding
Figure 218568DEST_PATH_IMAGE006
As a result of the calculation of (a),
Figure 408241DEST_PATH_IMAGE007
representing a new individual ViThe nth data in the mth gene,
Figure 308064DEST_PATH_IMAGE008
representing a new individual
Figure 77437DEST_PATH_IMAGE002
The nth data in the mth gene,
Figure 379105DEST_PATH_IMAGE009
representing a new individual
Figure 739679DEST_PATH_IMAGE003
The nth data in the mth gene,
Figure 126798DEST_PATH_IMAGE010
representing a new individual
Figure 434283DEST_PATH_IMAGE004
Setting the data quantity of each gene as Y for the nth data in the mth gene, wherein M is more than or equal to 1 and less than or equal to M, and n is more than or equal to 1 and less than or equal to Y;
s416, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s417, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s418, judging a new individual ViWhether the error rate Fi _ new of (a) is less than the error rate G2[ i ] of the parent Ui]If yes, command G2[ i]Equal to Fi _ new and hit A2iReplication ViWherein it is A2iReplication ViI.e. such that the ith individual A2iThe internal gene is completely linked with ViThe same is true; if not, G2[ i ] is not changed]；
S419, judging whether i is equal to N, if so, performing step S420, and if not, performing step S411;
s420, comparing the firstNew population A2Error rates G2[1 ] corresponding to all individuals]、G2[2]、……G2[i]……、G2[n-1]、G2[N]Size, selecting the minimum error rate min G2[ Z ]4]And find the minimum error rate min G2[ Z ]4]Corresponding individual
Figure 324879DEST_PATH_IMAGE011
Obtaining a second species A2Global optimal individual of (2)
Figure 121933DEST_PATH_IMAGE012
；
S42, population exchange: first population A1And a second species A2Carrying out communication: comparing the first population A1Min G1[ Z ] minimum error rate3]And a second species A2Min G2[ Z ] minimum error rate4]Size of (d), small min G1[ Z ]3]Or min G2[ Z4]The corresponding individual is a first kind of population A1And a second species A2To the co-optimal individual.
3. A method for finding a network computation structure of minimum errors for ambulatory ECG data, comprising the steps of:
step one, initialization:
s11, initializing each individual Xi，XiFor an array containing N genes, M X genes are providediAn individual, i is more than or equal to 1 and less than or equal to M, three data strings are arranged in each gene, one data string represents a calculation formula, the other two data strings represent nodes connected forwards, and one digit in each data string of an expression node is 0 or 1 for representing the expression or non-expression of the data string;
s12, initializing the crossover factor CR: CR takes any floating point numerical value in 0-1;
s13, initializing a variation factor F: f, taking any floating point numerical value in 0-1;
s14, initializing algebra gen, wherein gen = 0;
step two, calculating the first generation individuals XiError of [1 ] Fitness [ i ]]：
S21, nominal = 1;
s22, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN 1;
s23, carrying out iterative convergence on the weight of the calculation formula in the convolutional neural network structure CNN1 by using a gradient descent mode and the current electrocardiogram data to obtain an optimal weight matrix W1 of the convolutional neural network structure CNN 1;
s24, inputting the current electrocardiogram data into the convolutional neural network structure CNN1 with the optimal weight matrix W1, and calculating the error rate Fi_new1；
S25, MinFitness [ i ]]Is equal to Fi_new1；
S26、i=i+1；
S27, judging whether i is larger than M, if not, performing S22, and if so, performing a third step;
step three, setting a second species group A2Each individual is A2i，A2iIs length and XiEqual array, A2iReplication of XiIs namely A2i=XiAnd 1. ltoreq. i.ltoreq.M, G2[ i]= Fitness[i]Obtaining a second species A2Comprises the following steps:
s411, calculating gen = gen + 1;
s412, life = gen;
s413, setting a new individual ViIs length and XiEqual array, ViReplication of XiI.e. is Vi=Xi；
S414, crossing: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present3(ii) a Then, the probability P of random generation at present is judged3If the cross factor is smaller than the cross factor CR, if so, performing cross, otherwise, not performing cross, wherein the cross step is as follows: s3041, randomly selecting an integer R from an integer range (1, M), wherein R is not equal to i; s3042 randomly selecting an integer Z from the range of integers (0, M)1(ii) a S3043, mixing XRMiddle Z1All genes from gene M to gene M are duplicated to ViZ of (a)1Genes up to Mth gene;
S415. mutation: firstly, taking any floating point number in 0-1 as the probability P generated randomly at present4(ii) a Then, the probability P of random generation at present is judged4If the number of the new individuals V is less than the variation factor F, adopting a DE/best strategy to carry out comparison on the new individuals V if the number of the new individuals V is less than the variation factor FiPerforming mutation, if not, wherein a DE/best strategy is adopted to perform mutation on the new individual ViAnd (3) carrying out mutation: s4151 randomly selecting three integers R from the range of integers (0, M)1、R2And R3(ii) a S4152 New Individual ViData of each gene represented therein and
Figure 671382DEST_PATH_IMAGE002
、
Figure 844874DEST_PATH_IMAGE003
and
Figure 855556DEST_PATH_IMAGE004
the data of the internal expression gene are sequentially operated as the formula (1-1),
Figure 823512DEST_PATH_IMAGE005
(1-1), wherein the parenthesis [ alpha ], []The representation is rounded by rounding
Figure 857327DEST_PATH_IMAGE006
As a result of the calculation of (a),
Figure 568931DEST_PATH_IMAGE007
representing a new individual ViThe nth data in the mth gene,
Figure 434119DEST_PATH_IMAGE008
representing a new individual
Figure 572976DEST_PATH_IMAGE002
The nth data in the mth gene,
Figure 94087DEST_PATH_IMAGE009
representing a new individual
Figure 343803DEST_PATH_IMAGE003
The nth data in the mth gene,
Figure 329076DEST_PATH_IMAGE010
representing a new individual
Figure 576518DEST_PATH_IMAGE004
Setting the data quantity of each gene as Y for the nth data in the mth gene, wherein M is more than or equal to 1 and less than or equal to M, and n is more than or equal to 1 and less than or equal to Y;
s416, adding the new individual ViDecoding the gene data into a convolutional neural network structure CNN2, and then performing iterative convergence on a calculation formula in the convolutional neural network structure CNN2 by using a gradient descent mode and electrocardiogram data to obtain an optimal weight matrix W2 of the convolutional neural network structure CNN 2;
s417, verification and evaluation: inputting the electrocardio data into a convolutional neural network structure CNN2 with an optimal weight matrix W2, and calculating an error rate Fi _ new;
s418, judging a new individual ViWhether the error rate Fi _ new of (a) is less than the error rate G2[ i ] of the parent Ui]If yes, command G2[ i]Equal to Fi _ new and hit A2iReplication ViWherein it is A2iReplication ViI.e. such that the ith individual A2iThe internal gene is completely linked with ViThe same is true; if not, G2[ i ] is not changed]；
S419, judging whether i is equal to N, if so, performing step S420, and if not, performing step S411;
s420, comparing the first new population A2Error rates G2[1 ] corresponding to all individuals]、G2[2]、……G2[i]……、G2[n-1]、G2[N]Size, selecting the minimum error rate min G2[ Z ]4]And find the minimum error rate min G2[ Z ]4]Corresponding individual
Figure 647242DEST_PATH_IMAGE011
Obtaining a second species A2Global optimal individual of (2)
Figure 700649DEST_PATH_IMAGE012
。
CN201810191992.9A 2018-03-08 2018-03-08 Method for searching error minimum network computing structure for dynamic ECG data Active CN108470158B (en)

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 CN108470158A (en) 2018-08-31
CN108470158B true 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)

* Cited by examiner, † Cited by third party
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

Patent Citations (10)

* Cited by examiner, † Cited by third party
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
CN108470158A (en) 2018-08-31

Similar Documents

Publication Publication Date Title
Meng et al. PaDE: An enhanced Differential Evolution algorithm with novel control parameter adaptation schemes for numerical optimization
Rost et al. Prediction of protein secondary structure at better than 70% accuracy
Ahmad et al. NETASA: neural network based prediction of solvent accessibility
Hunley et al. The apportionment of human diversity revisited
CN109411016B (en) Gene variation site detection method, device, equipment and storage medium
CN109448794A (en) A kind of epistasis site method for digging based on heredity taboo and Bayesian network
CN112001887B (en) Full convolution genetic neural network method for infant brain medical record image segmentation
CN111077768B (en) Optimization device and control method of optimization device
CN113936738A (en) RNA-protein binding site prediction method based on deep convolutional neural network
CN113627471A (en) Data classification method, system, equipment and information data processing terminal
CN112164474A (en) Drug sensitivity prediction method based on self-expression model
CN108470158B (en) Method for searching error minimum network computing structure for dynamic ECG data
CN111930007A (en) Optimization device and method for controlling the same
Andalon-Garcia et al. Performance comparison of three topologies of the island model of a parallel genetic algorithm implementation on a cluster platform
CN115908909A (en) Evolutionary neural architecture searching method and system based on Bayes convolutional neural network
Hadikhani et al. Improved data clustering using multi-trial vector-based differential evolution with Gaussian crossover
CN116153396A (en) Non-coding variation prediction method based on transfer learning
Chang et al. Genetic architecture search for binarized neural networks
KR102659917B1 (en) Method for developing meta-gene based on non-negative matrix factorization and applications thereof
KR20200121039A (en) Electronic device for generating a gene feature vector for gene distributed representation based on a correlation between genes according to cancer and operating method thereof
CN109360602B (en) DNA coding sequence design method and device based on fuzzy priority
KR101897513B1 (en) Method, apparatus and computer-readable recording medium for selecting feature using relational matrix
CN112908416A (en) Biomedical data feature selection method and device, computing equipment and storage medium
Schwender et al. Package ‘scrime’
Clay Inverse population genetic problems with noise: inferring extent and structure of haplotype blocks from point allele frequencies

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