CN109726850B - Method for establishing resultant yarn quality prediction model - Google Patents
Method for establishing resultant yarn quality prediction model Download PDFInfo
- Publication number
- CN109726850B CN109726850B CN201811433170.3A CN201811433170A CN109726850B CN 109726850 B CN109726850 B CN 109726850B CN 201811433170 A CN201811433170 A CN 201811433170A CN 109726850 B CN109726850 B CN 109726850B
- Authority
- CN
- China
- Prior art keywords
- data
- input
- output
- value
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000012360 testing method Methods 0.000 claims abstract description 99
- 238000012937 correction Methods 0.000 claims abstract description 15
- 239000011159 matrix material Substances 0.000 claims description 102
- 238000013507 mapping Methods 0.000 claims description 71
- 238000011156 evaluation Methods 0.000 claims description 44
- 229920000742 Cotton Polymers 0.000 claims description 19
- 230000015572 biosynthetic process Effects 0.000 claims description 17
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000003786 synthesis reaction Methods 0.000 claims description 10
- 238000003780 insertion Methods 0.000 claims description 9
- 230000037431 insertion Effects 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000002194 synthesizing effect Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000013210 evaluation model Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- OVBPIULPVIDEAO-LBPRGKRZSA-N folic acid Chemical compound C=1N=C2NC(N)=NC(=O)C2=NC=1CNC1=CC=C(C(=O)N[C@@H](CCC(O)=O)C(O)=O)C=C1 OVBPIULPVIDEAO-LBPRGKRZSA-N 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 238000007619 statistical method Methods 0.000 claims 1
- 238000004519 manufacturing process Methods 0.000 abstract description 11
- 239000004753 textile Substances 0.000 abstract description 9
- 230000008901 benefit Effects 0.000 abstract description 5
- 230000002860 competitive effect Effects 0.000 abstract description 2
- 230000003631 expected effect Effects 0.000 abstract description 2
- 206010020112 Hirsutism Diseases 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000009960 carding Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 210000003746 feather Anatomy 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000010813 municipal solid waste Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Treatment Of Fiber Materials (AREA)
Abstract
The invention has proposed the establishment method of a yarn quality prediction model, regard all to obtain the expected effect data as the output data, input every parameter and output every index as a whole respectively, utilize the multi-mode reasoning method to set up a relation model between input data and output data, thus has set up the yarn quality prediction model, and has given the assessment decision-making method of the model, have carried on the error correction to the yarn quality prediction model; and verifying the input-output relationship by generating different output conditions under different conditions through a series of different input test data. The invention improves the quality of textile products of textile enterprises, enlarges the variety range, improves the production efficiency, reduces the labor intensity of operators, reduces the production cost of the enterprises, improves the economic benefit, enhances the adaptability and the competitive power of the textile products of the enterprises in the international market, and can meet the increasing demands of people according to different conditions.
Description
Technical Field
The invention belongs to the technical field of textile quality prediction, particularly relates to data processing and knowledge mining of cotton quality, and particularly relates to a method for establishing a finished yarn quality prediction model.
Background
In the rapid development of computer network technology, the traditional cotton yarn production is increasingly unable to meet the requirements of people. The quality index of cotton, the parameters of a machine in the process of producing cotton yarns, the field environment of a production workshop and the like all have certain influence on the quality of the produced cotton yarns, some are main influence factors, and some are secondary influence factors. The quality of the produced cotton yarn is evaluated and has quality indexes, such as strength, neps, evenness, hairiness value, single strength and the like. In many input-output relationships, the input-output relationships do not exhibit a distinct linear relationship, and instead of changing an input parameter, the output parameter changes accordingly. If the change is the main factor, the output result is obviously changed, if the change is the secondary factor, the output result is not obviously changed, the input parameters are coupled and not mutually independent, and under a certain input, some input parameters play the main role, some input parameters play the secondary role and some input parameters play no role, so when the input parameters are changed, if the change is the main factor, the output index is obviously changed. Some inputs and outputs have coupling relations, and the relations are complex and difficult to establish.
Disclosure of Invention
Aiming at the technical problems that the yarn forming process is a complex mapping relation with multiple inputs and multiple outputs and the prediction of the yarn forming quality is difficult, the invention provides the method for establishing the yarn forming quality prediction model, which improves the prediction effect of the yarn forming quality, reduces the cost, improves the economic benefit and greatly improves the productivity of enterprises.
In order to achieve the purpose, the technical scheme of the invention is realized as follows: a method for establishing a yarn quality prediction model comprises the following steps:
the method comprises the following steps: collecting data of raw cotton indexes in an enterprise as an original input database and data of finished yarn quality indexes as an original output database, and filtering and normalizing the data in the original input database and the original output database;
step two: dividing the data in the normalized input database into input test data and input sample data, wherein the data in the original output database corresponding to the input test data is output test data, and the data in the original output database corresponding to the input sample data is output sample data;
step three: yarn formation prediction model: establishing function transformation from input sample data to output sample data to obtain a first mapping relation matrix of the input sample data and the output sample data;
step four: and (3) model evaluation: establishing a single-factor evaluation value of each output factor in the output sample data corresponding to each input sample data, and forming a single-factor evaluation matrix by using all the single-factor evaluation values; judging each input factor according to the weight of each input sample data and an expert system in the aspect to obtain an evaluation matrix of the output factors as a second mapping relation matrix, and storing the evaluation matrix; performing a synthesis operation on each input sample data and the second mapping relation matrix, comparing the obtained calculation value with the corresponding output sample data, and if the calculation value is smaller than a given error threshold value, replacing the first mapping relation matrix in the third step by the second mapping relation matrix and storing the second mapping relation matrix; if the error value is larger than the given error threshold value, storing the obtained first mapping relation matrix in the step three; the step four is circulated, the machine learns all input sample data, and the trained mapping relation matrix is obtained and stored in the database;
step five: and (3) model verification: synthesizing input test data by using a trained mapping relation matrix stored in a database to obtain a test result, comparing the test result with output test data to obtain an error, and entering a sixth step if the error is larger than a given error; otherwise, returning to the step five, when the errors corresponding to all the input test data reach the set threshold, storing the yarn prediction model, and ending the program;
step six: correction of model evaluation matrix: and adjusting the mapping relation matrix by correcting the weight of the input test data, and returning to the step five.
The data filtering method in the first step comprises the following steps: only one group of data which are the same or similar is reserved according to the requirements of main parameters, and other groups of data are removed; the data normalization processing method comprises the following steps: processing data to [0,1 ]]The data in between, namely:
wherein xi represents the normalized data of the original input database or the original output database; x represents data in the raw input database or the raw output database; x is the number ofminRepresenting a minimum value in the raw input database or the raw output database; x is the number ofmaxRepresenting the maximum value in the raw input database or the raw output database.
The mapping relation matrix is synthesized by input sample data and output sample data:
wherein R isTFor the mapping relation matrix, U ═ U1,u2,...,unV ═ V } is input sample data1,v2,...,vmFor the output of the sample data, the output data is,
represents a synthesis operator, uiA value indicating each input index, i being 1,2, …, n, n being the number of input indexes; v. ofjAnd j is 1,2, …, and m is the number of output assessment indexes.
The model evaluation method comprises the following steps: factor u for each input sampleiMaking an assessment f (u) alonei) Evaluation of f (u)i) Is the input sample factor U in the input sample data UiA model mapping f to output sample data V, from whichInducing a model relation R from all input sample factors of input sample data U to output sample data VfFrom the model relation RfCan induce a model linear transformation from input sample data U to output sample data V
Where A is the normalized weight set of input sample data U, B is the calculated weight set corresponding to the output sample data, B is called the evaluation model.
The evaluation
Different models are obtained by different definitions of the synthesis operator, and the weight A of input sample data is equal to (a)1,a2,...an)：
(a) Model M (. cndot. /) prominent type of principal factor, bj＝/{(ai·rij)，1≤i≤n},j＝1,2,...,m；
(b) Model M (&C (+) prominent type of major factor, bj＝∑(ai&rij),1≤i≤n,j＝1,2,...,m；
(c) Model M (, +) weighted average model, bj＝∑(ai·rij),1≤i≤n,j＝1,2,...,m。
The weight of the input sample data is the influence coefficient of the input sample data on the output sample data, the weight is the influence value of the input data given by a plurality of experts, and the weight of the input sample data is determined by using a frequency statistics method:
(1) factor u for each input sample dataiAt a weight a given by k expertsijFind out the maximum value MiAnd minimum value miRespectively as follows:
Mi＝max{aijj is more than or equal to 1 and less than or equal to k; 1,2, n, and
mi＝min{aij|1≤j≤k；i＝1,2,...,n}；
(2) selecting proper positive integer p, and weighting aijDividing the obtained product into p groups from small to large, wherein the group distance S is as follows:
(3) calculating the frequency number and the frequency falling in each group of weights;
(4) taking the group median or neighborhood of the group in which the maximum frequency lies as the factor uiThe weight of (c).
The correction method of the trained mapping relation matrix comprises the following steps: establishing a mapping relation between input test data and output test data by using the trained mapping relation matrix:
wherein,
in order to input a matrix of test data,
in order to output a matrix of test data,
in order to be a matrix of the mapping relationship,
a weight matrix for input test data;
wherein, ω isp1+ωp2+...+ωpn＝1；
Correcting mapping relation matrix R by correcting weight of input test dataTThe weight of the input test data is omegal＝ωl-1+μ(ωl-ωl-1) (ii) a Wherein, ω islIs the current weight; omegal-1For the weight obtained from the previous calculation, the initial weight ω0Training according to a sample to obtain an empirical value, wherein mu is a learning rate of the weight;
gradually adjusting the mapping relation matrix R by verifying different input test data and output test dataT。
The correction method of the matrix Y for outputting the test data comprises the following steps: y isq＝y1+η(yq-y1) Wherein, ynIs the current output value, y1Is the sample output value, eta is the learning rate of the output sample; and (4) iterating different output test data according to a certain step length to obtain a relatively ideal output result.
Simultaneously correcting input test data and output test data by adding data fitting, wherein the data fitting method comprises the following steps:
L(z)＝L0(z)y0+L1(z)y0+L2(z)y0
wherein,
l (z) output insertion data indicating that the sample or test data corresponds to the input data z, L0(z) corresponds to input data z0Output insertion data of L1(z) corresponds to input data z1Output insertion data of L2(z) corresponds to input data z2The output of (2) inserts data, z1、z2、z0Respectively representing different input data, y0The representation corresponds to input data z0The output data of (1);
the formula is satisfied at the fitting node:
then:
when new test data is input for yarn-forming prediction modeling, a predicted value is output through a mapping relation matrix, and then the predicted output value is compared with an actually required output value to correct errors, wherein the method comprises the following steps:
1. several parameters (lambda) influencing the output of the user input1,λ2,...,λm) Performing linear fitting;
2. several parameters (lambda)1,λ2,...,λm) Corresponding output test value
Absolute values of differences from corresponding values in the corresponding samples;
(1) for each input parameter (lambda)1,λ2,...,λm) Making difference value, selecting several groups N of input values whose number is most and difference value is 01；
(2) On the basis of step (1), the groups N1Selecting a plurality of groups N of the corresponding input values with the minimum number of the median differences2≤N1；
(3) Selecting a plurality of groups N2The input value of (2) is corresponding to a sub mapping matrix R in the mapping relation matrixsInput test value of user
And a sub-mapping matrix RsFor synthesizing operations, i.e.
Is a different output test result, s 1,22；
(4) If the user gives an output result
Then the minimum closeness method is used to obtain the output test result
Is selected as the minimum value of
The output result with the minimum expected value
Or selecting the output test result
In
The group with the smallest number is the most input parameters; then, based on a certain main factor, the group of output calculation with the smallest difference value with the main factor is selected.
The invention has the beneficial effects that: all input parameters and all output indexes are respectively taken as a whole to establish a relation model between the input parameters and the output indexes, and input data are normalized by carrying out data acquisition, analysis and characteristic extraction on actually produced yarns; secondly, all the obtained expected effect data are used as output data, input parameters and output indexes are respectively used as a whole, a relation model between the input data and the output data is established by utilizing a multi-mode reasoning method, so that a yarn quality prediction model is established, and an evaluation decision method of the model is provided; the evaluation decision method carries out error correction on the resultant yarn quality prediction model, provides three error correction methods, and effectively avoids unnecessary loss caused by overlarge error; the input-output relationship can be verified by a series of correct different input data to generate different output conditions under different conditions. According to the invention, a more accurate finished yarn prediction model system is established and developed, the finished yarn quality prediction model is tested in the value range of each input parameter, the errors of each output index are controlled to be below 10%, the errors of the output strength, evenness and hairiness are controlled to be about 3%, the errors of 90% nep are controlled to be below 8%, and the errors of more than 80% of single strength are controlled to be below 1 unit, so that the basis is provided for the production of a subsequent cotton yarn factory. The experimental verification proves that: the invention improves the quality of textile products of textile enterprises, enlarges the variety range, improves the production efficiency, reduces the labor intensity of operators, reduces the production cost of the enterprises, improves the economic benefit, enhances the adaptability and the competitive power of the textile products of the enterprises in the international market, and can meet the increasing demands of people according to different conditions. According to different requirements of different customers, the cotton yarn can meet the requirements of the customers, different requirements of different groups are met, and the cotton yarn is different in the fields of fire fighting, military, labor protection, art, aerospace and the like and can be supported in the aspect of cotton yarn cloth.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a histogram of errors for an embodiment of the present invention.
FIG. 3 is a graph of the original versus predicted values for the brute force of an embodiment of the present invention.
FIG. 4 is a graph comparing the original value to the predicted value of neps according to embodiments of the invention.
FIG. 5 is a graph of raw versus predicted values for evenness according to an embodiment of the present invention.
FIG. 6 is a graph comparing the original value and the predicted value of the hair feather value according to the embodiment of the present invention.
FIG. 7 is a graph of the original versus predicted values for single intensity for an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive effort based on the embodiments of the present invention, are within the scope of the present invention.
As shown in fig. 1, a method for establishing a yarn formation quality prediction model, wherein a yarn formation process is a complex mapping relation with multiple inputs and multiple outputs, the inputs and the outputs are respectively regarded as a whole to establish a relation model, the yarn formation prediction model is evaluated through machine iterative learning, and error correction is performed according to comparison between a predicted output value and an actual value, and the method comprises the following steps:
the method comprises the following steps: the method comprises the steps of collecting data of raw cotton indexes in enterprises as an original input database and data of finished yarn quality indexes as an original output database, and filtering and normalizing the data in the original input database and the original output database.
And (3) combing the relation between multiple inputs and multiple outputs, preprocessing the test data and the sample data, removing the data which do not meet the actual production requirement, and then establishing a model and evaluating the model through learning the sample.
The data filtering method in the first step comprises the following steps: a large amount of data is obtained by calling all input data of an original input database, only one group of the same or similar data is reserved according to target requirements such as requirements of main parameters, and other groups of data are removed.
For the filtered input data, units of various types of data are not uniform in order to facilitate data processing, and therefore normalization processing is performed. The data normalization processing method comprises the following steps: processing data to [0,1 ]]The data in between, namely:
wherein x isiRepresenting the normalized data of the original input database or the original output database; x represents data in the raw input database or the raw output database; x is the number ofminRepresenting a minimum value in the raw input database or the raw output database; x is the number ofmaxRepresenting the maximum value in the raw input database or the raw output database.
Step two: and dividing the data in the normalized input database into input test data and input sample data, wherein the data in the original output database corresponding to the input test data is output test data, and the data in the original output database corresponding to the input sample data is output test data.
And inputting sample data and outputting the sample data to establish a finished yarn prediction model and model evaluation, and inputting test data and outputting the test data to verify the model. In a specific application, the ratio of sample data to test data is half of each.
Step three: establishing a finished yarn prediction model: and establishing functional transformation from input sample data to output sample data to obtain a mapping relation matrix of the input sample data and the output sample data.
Mapping relation matrix RTIs performed by a functional transformation T of input sample data U to output sample data V. The mapping relation matrix is synthesized by input sample data and output sample data:
wherein R isTFor the mapping relation matrix, U ═ U1,u2,...,unV ═ V } is input sample data1,v2,...,vmFor the output of the sample data, the output data is,
represents a synthesis operator, uiA value indicating each input index, i being 1,2, …, n, n being the number of input indexes; v. ofjAnd j is 1,2, …, and m is the number of output assessment indexes. By means of the synthesis operator, given input sample data U to output sample data V, a mapping relation matrix R can be determinedT。
Step four: and (3) model evaluation: establishing a single-factor evaluation value of each output factor in the output sample data corresponding to each input sample data, and forming a single-factor evaluation matrix by using all the single-factor evaluation values; and judging each input factor according to the weight of each input sample data and the expert system in the aspect to obtain an evaluation matrix of the output factors, wherein the evaluation matrix is used as a second mapping relation matrix and stored. Then, each piece of input sample data is used for carrying out synthesis operation with the input sample data, the obtained calculated value is compared with the corresponding output data, and if the calculated value is smaller than a given error threshold value, the relationship matrix storage in the third step is replaced; and if the error is larger than the given error threshold value, storing the mapping relation matrix in the third step. According to the same method, the machine learns all input sample data, and the trained mapping relation matrix is stored in the database.
In order to verify whether the yarn formation prediction model obtained in the step three is directly feasible and optimal for input and output, the established yarn formation prediction model needs to be evaluated.
The model evaluation method comprises the following steps: factor u for each input sampleiMaking an assessment f (u) alonei) Evaluation of f (u)i) Is the input sample factor U in the input sample data UiA model mapping f to the output sample data V, which can induce a model relation R from all input sample factors of the input sample data U to the output sample data VfFrom the model relation RfCan induce a model linear transformation from input sample data U to output sample data V
Where A is the normalized set of weights for the input samples U and B is the set of calculated weights corresponding to the output samples V, called the evaluation model.
Taking input sample data as a factor set U ═ U1,u2,...,unOutputting sample data as decision set V ═ V1,v2,...,vm}. For each input factor uiFirst, a single factor evaluation { r is establishedi1,ri2,...,rinI.e. rij(0≤rij1) represents the output factor vjFor input factor uiThe evaluation is performed such that a one-factor evaluation matrix R is obtainedf＝(rij)m×n. One-factor evaluation matrix RfAnd mappingRelation matrix RTHave the same meaning except that RfIs a finite matrix of elements, RTIs a matrix expandable into several elements, i.e. RfIs a mapping relation matrix RTBecause of rijAlso shown are the weights of each element, and the resulting weight matrix is denoted Rf。
The evaluation
Different definitions of the synthesis operator, different models are obtained, wherein the weight A of the input sample data is equal to (a)1,a2,...an)：
(a) Model M (. cndot. /) prominent type of principal factor, bj＝/{(ai·rij) I is more than or equal to 1 and less than or equal to n, and j is 1, 2. In model M (·, /), pair rijMultiplying by a weight a less than 1iIndicates aiIs when considering multiple factors rijThe correction values of (2) are related to the primary factors, thus ignoring the secondary factors.
(b) Model M (&C (+) prominent type of major factor, bj＝∑(ai&rij) I is not less than 1 and not more than n, j is 1,2,., m; model M (&And (+) also highlights the main factor, and in practical applications, it is recommended to adopt either model (a) or model (b) if the main factor plays a dominant role in the evaluation.
(c) Model M (, +) weighted average model, bj＝∑(ai·rij) I is not less than 1 and not more than n, j is 1, 2. The model M (·, +) gives consideration to all factors according to the weight balance, and is suitable for the condition of considering each factor to act.
By
It can be obtained that each measured value in the output sample data V corresponding to each input sample data U is cijI.e. cij(i 1, 2.. n.; j1, 2.. m) represents the value of the j-th output index of the i-th input factor, and then by normalizing:
obtaining a model evaluation matrix RT＝(rij)m×nThe corresponding is the element of the one-way evaluation matrix. The invention provides a specific evaluation matrix given by an expert system and a specific calculation method for how to obtain the elements of the single-factor evaluation matrix.
The weight of the input sample data is the influence coefficient of the input sample data on the output sample data, the weight is the influence value of the input data given by a plurality of experts, and the weight of the input sample data is determined by using a frequency statistics method:
(1) factor u for each input sample dataiAt a weight a given by k expertsijFind out the maximum value MiAnd minimum value miRespectively as follows:
Mi＝max{aijj is more than or equal to 1 and less than or equal to k; 1,2, n, and
mi＝min{aij|1≤j≤k；i＝1,2,...,n}；
(2) selecting proper positive integer p, and weighting aijDividing the obtained product into p groups from small to large, wherein the group distance S is as follows:
(3) calculating the frequency number and the frequency falling in each group of weights;
(4) taking the group median or neighborhood of the group in which the maximum frequency lies as the factor uiThe weight of (c).
Step five: and (3) model verification: synthesizing input test data by using a trained mapping relation matrix stored in a database to obtain a test result, comparing the test result with output test data to obtain an error, and entering a sixth step if the error is larger than a given error; otherwise, returning to the step five, when the errors corresponding to all the input test data reach the set threshold, storing the model, and ending the program.
Step six: correction of model evaluation matrix: and adjusting the model evaluation matrix by correcting the weight of the input test data, and returning to the step five.
And inducing a mapping relation matrix from input data to output data through function transformation T. The correction method for the model evaluation matrix which is also a mapping relation matrix comprises the following steps: establishing a mapping relation between input test data and output test data by using the model evaluation matrix:
wherein,
in order to input a matrix of test data,
in order to output a matrix of test data,
in order to be a matrix of the mapping relationship,
is a weight matrix of the input test data.
Wherein, ω isp1+ωp2+...+ωpn＝1；
According to the mapping relation, the correction of the mapping relation matrix R is realized by correcting the weight of the input test dataTThe weight of the input test data is omegal＝ωl-1+μ(ωl-ωl-1) (ii) a Wherein, ω islIs the current weight; omegal-1For the weight obtained from the previous calculation, the initial weight ω0And (4) obtaining an empirical value according to sample training, wherein mu is the learning rate of the weight. Gradually adjusting the mapping relation matrix R by verifying different samples and testing input data and output dataT。
The correction method of the matrix Y for outputting the test data comprises the following steps: y isq＝y1+η(yq-y1) Wherein, ynIs the current output value, y1Is the sample output value, eta is the learning rate of the output sample; different output test data are iterated according to a certain step lengthThe step length is selected to ensure that iteration is carried out for a limited number of times, and a relatively ideal output result is obtained.
Taking the correction of the matrix Y of the output test data as an example, when the learning rate eta is 0.75 according to data analysis, machine learning and sample training, the prediction result of the data is better. When a new batch of test data is put into training, the learning is further carried out, and meanwhile, the learning rate is controlled within the range of 0.8-0.95 in order to control errors. When a new batch of new data is added, the learning rate is adjusted to 0.85 through training, and not only is the mapping relation matrix R corrected for the second timeTAnd meanwhile, fitting the data by machine learning, and adjusting the learning rate and the step length.
Because the data volume is very little and some data have null values or large jump, in order to enable the machine to learn the change rule of the data, a data fitting process is added, and input and output are corrected. Simultaneously correcting input test data and output test data by adding data fitting, wherein the data fitting method comprises the following steps:
L(z)＝L0(z)y0+L1(z)y0+L2(z)y0
wherein,
l (z) output insertion data indicating that the sample or test data corresponds to z, L0(z) corresponds to z0Output insertion data of L1(z) corresponds to z1Output insertion data of L2(z) corresponds to z2The output of (2) inserts data, z1、z2、z0Respectively representing different input data, y0The representation corresponds to z0The output data of (1).
The formula is satisfied at the fitting node:
then:
when new test data is input for yarn-forming prediction modeling, a predicted value is output through a mapping relation matrix, and then the predicted output value is compared with an actually required output value to correct errors, wherein the method comprises the following steps:
1. several parameters (lambda) influencing the output of the user input1,λ2,...,λm) Performing linear fitting;
2. several parameters (lambda)1,λ2,...,λm) Corresponding output test value
Absolute values of differences from corresponding values in the corresponding samples;
(1) for each input parameter (lambda)1,λ2,...,λm) Making difference value, selecting several groups N of input values whose number is most and difference value is 01；
(2) On the basis of step (1), the groups N1Selecting a plurality of groups N of the corresponding input values with the minimum number of the median differences2≤N1；
(3) Selecting a plurality of groups N2The input value of (2) is corresponding to a sub mapping matrix R in the mapping relation matrixsInput test value of user
And a sub-mapping matrix RsFor synthesizing operations, i.e.
Is a different output test result, s 1,22；
(4) If the user gives an output result
Then the minimum closeness method is used to obtain the output test result
Is selected as the minimum value of
The output result with the minimum expected value
Or selecting the output test result
In
The group with the smallest number is the most input parameters; then, based on a certain main factor, the group of output calculation with the smallest difference value with the main factor is selected. The meaning of the test output is the same as before, except that here the calculated output value corresponding to the new input of the user is derived.
A yarn quality prediction system is developed on the basis of the method or algorithm of each step provided by the invention, and the prediction system is applied to the yarn forming process of a certain textile group production test. The micronaire value of the raw cotton index is 4.59, the strength of the raw cotton is 29.9, the average length of the right half part of the raw cotton is 29.02, the length uniformity is 83, the trash content is 1.05, the maturity is 0.87, the short staple rate of the cotton carding AFIS is 8.6, the neps of the cotton carding AFIS are 35, and the production area is Xinjiang; the actual results of the resultant yarn quality obtained from the raw cotton were 249.7 in tenacity, 8 in neps, 11.5 in yarn levelness CV%, 3.19 in hairiness H value, and 6 in single tenacity CV%. The output results of the test by the method of the invention are 229.987 strong, 15.367 neps, 8.427 yarn levelness CV%, 9.5596 hairiness H value and 15.42 single strong CV%. The test result shows that: the cost of an enterprise is reduced by about 25%, the economic benefit is improved by about 20%, and the productivity of the enterprise is greatly improved.
And (4) performing five times of error correction, namely correcting output, correcting a function matrix, correcting a learning rate, correcting an iteration step length and correcting data fitting. As shown in Table 1, the errors of the predicted five parameter indexes can be controlled to be less than 10% and are slightly larger than the respective errors, wherein the errors of the output strength, the evenness and the hairiness are controlled to be about 3%; the error of neps is mostly controlled below 10%; the error of more than 80% of single intensity is controlled below 1 unit, and the individual error is slightly larger. Fig. 2 shows the size of the error by using a bar chart, and fig. 3-7 show that the size of the error basically fluctuates within a certain range, according to the production test of a certain textile group, the cost is reduced by 25%, the benefit is improved by 20%, the productivity of an enterprise is greatly improved, and various requirements of people can be continuously met.
TABLE 1 error data
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (6)
1. A method for establishing a yarn quality prediction model is characterized by comprising the following steps:
the method comprises the following steps: collecting data of raw cotton indexes in an enterprise as an original input database and data of finished yarn quality indexes as an original output database, and filtering and normalizing the data in the original input database and the original output database;
step two: dividing the data in the normalized input database into input test data and input sample data, wherein the data in the original output database corresponding to the input test data is output test data, and the data in the original output database corresponding to the input sample data is output sample data;
step three: yarn formation prediction model: establishing function transformation from input sample data to output sample data to obtain a first mapping relation matrix of the input sample data and the output sample data;
step four: and (3) model evaluation: establishing a single-factor evaluation value of each output factor in the output sample data corresponding to each input sample data, and forming a single-factor evaluation matrix by using all the single-factor evaluation values; judging each input factor according to the weight of each input sample data and an expert system in the aspect to obtain an evaluation matrix of the output factors as a second mapping relation matrix, and storing the evaluation matrix; performing a synthesis operation on each input sample data and the second mapping relation matrix, comparing the obtained calculation value with the corresponding output sample data, and if the calculation value is smaller than a given error threshold value, replacing the first mapping relation matrix in the third step by the second mapping relation matrix and storing the second mapping relation matrix; if the error value is larger than the given error threshold value, storing the obtained first mapping relation matrix in the step three; the step four is circulated, the machine learns all input sample data, and the trained mapping relation matrix is obtained and stored in the database;
step five: and (3) model verification: synthesizing input test data by using a trained mapping relation matrix stored in a database to obtain a test result, comparing the test result with output test data to obtain an error, and entering a sixth step if the error is larger than a given error; otherwise, returning to the step five, when the errors corresponding to all the input test data reach the set threshold, storing the yarn prediction model, and ending the program;
step six: correction of model evaluation matrix: adjusting the mapping relation matrix by correcting the weight of the input test data, and returning to the step five;
the mapping relation matrix is synthesized by input sample data and output sample data:
wherein R isTFor the mapping relation matrix, U ═ U1,u2,...,unV ═ V } is input sample data1,v2,...,vmFor the output of the sample data, the output data is,
represents a synthesis operator, uiA value indicating each input index, i being 1,2, …, n, n being the number of input indexes; v. ofjThe output meets the assessment indexes, j is 1,2, …, m is the number of the output assessment indexes;
the model evaluation method comprises the following steps: factor u for each input sampleiMaking an assessment f (u) alonei) Evaluation of f (u)i) Is the input sample factor U in the input sample data UiA model mapping f to the output sample data V, which can induce a model relation R from all input sample factors of the input sample data U to the output sample data VfFrom the model relation RfCan induce a model linear transformation from input sample data U to output sample data V
Where A is the normalized weight set of input sample data U, B is the calculated weight set corresponding to the output sample data, B is called the evaluation model;
the evaluation model
Different models are obtained by different definitions of the synthesis operator, and the weight A of input sample data is equal to (a)1,a2,...an)：
(a) Model M (. cndot. /) prominent type of principal factor, bj＝/{(ai·rij)，1≤i≤n},j＝1,2,...,m；
(b) Model M (&C (+) prominent type of major factor, bj＝∑(ai&rij),1≤i≤n,j＝1,2,...,m；
(c) Model M (, +) weighted average model, bj＝∑(ai·rij),1≤i≤n,j＝1,2,...,m。
2. The method for building the yarn quality prediction model according to claim 1, wherein the data filtering method in the first step is as follows: only one group of data which are the same or similar is reserved according to the requirements of main parameters, and other groups of data are removed; the data normalization processing method comprises the following steps: processing data to [0,1 ]]The data in between, namely:
wherein x isiRepresenting the normalized data of the original input database or the original output database; x represents data in the raw input database or the raw output database; x is the number ofminRepresenting a minimum value in the raw input database or the raw output database; x is the number ofmaxRepresenting the maximum value in the raw input database or the raw output database.
3. The method for building the yarn quality prediction model according to claim 1, wherein the weight of the input sample data is an influence coefficient of the input sample data on the output sample data, the weight is an influence value of the input data given by a plurality of experts, and the weight of the input sample data is determined by a frequency statistical method:
(1) factor u for each input sample dataiAt a weight a given by k expertsij1Find out the maximum value MiAnd minimum value miRespectively as follows:
Mi＝max{aij1j1 is more than or equal to |1 and less than or equal to k; 1,2, n, and
mi＝min{aij1|1≤j1≤k；i＝1,2,...,n}；
(2) selecting proper positive integer p, and weighting aij1Dividing the obtained product into p groups from small to large, wherein the group distance S is as follows:
(3) calculating the frequency number and the frequency falling in each group of weights;
(4) taking the group median or neighborhood of the group in which the maximum frequency lies as the factor uiThe weight of (c).
4. The method for establishing the yarn quality prediction model according to claim 1, wherein the method for correcting the trained mapping relation matrix comprises the following steps: establishing a mapping relation between input test data and output test data by using the trained mapping relation matrix:
wherein,
in order to input a matrix of test data,
in order to output a matrix of test data,
in order to be a matrix of the mapping relationship,
a weight matrix for input test data;
wherein, ω isp1+ωp2+...+ωpn＝1；
Correcting mapping relation matrix R by correcting weight of input test dataTThe weight of the input test data is omegal＝ωl-1+μ(ωl-ωl-1) (ii) a Wherein, ω islIs the current weight; omegal-1For the weight obtained from the previous calculation, the initial weight ω0Training according to a sample to obtain an empirical value, wherein mu is a learning rate of the weight;
gradually adjusting the mapping relation matrix R by verifying different input test data and output test dataT。
5. The method for building a yarn quality prediction model according to claim 4, wherein the matrix Y of the output test data is corrected by: y isq＝y1+η(yq-y1) Wherein, yqIs the current output value, y1Is the sample output value, eta is the learning rate of the output sample; different output test data are iterated according to a certain step length to obtain a relatively ideal output result;
simultaneously correcting input test data and output test data by adding data fitting, wherein the data fitting method comprises the following steps:
L(z)＝L0(z)y0+L1(z)y1+L2(z)y2；
wherein,
l (z) output insertion data indicating that the sample or test data corresponds to the input data z, L0(z) corresponds to input data z0Output insertion data of L1(z) corresponds to input data z1Output insertion data of L2(z) corresponds to input data z2The output of (2) inserts data, z1、z2、z0Respectively representing different input data, y0The representation corresponds to input data z0The output data of (1);
the formula is satisfied at the fitting node:
then:
6. the method for building a yarn quality prediction model according to claim 5, wherein when new test data is input for yarn predictive modeling, a predicted value is output through the mapping relation matrix, and then an error is corrected by comparing the predicted output value with an actually required output value, and the method comprises the steps of:
1. several parameters (lambda) influencing the output of the user input1,λ2,...,λm1) Performing linear fitting;
2. several parameters (lambda)1,λ2,...,λm1) Corresponding output test value
Absolute values of differences from corresponding values in the corresponding samples;
(1) for each input parameter (lambda)1,λ2,...,λm1) Making difference value, selecting several groups N of input values whose number is most and difference value is 01；
(2) On the basis of step (1), the groups N1Selecting a plurality of groups N of the corresponding input values with the minimum number of the median differences2≤N1；
(3) Selecting a plurality of groups N2The input value of (2) is corresponding to a sub mapping matrix R in the mapping relation matrixsInput test value of user
And a sub-mapping matrix RsFor synthesizing operations, i.e.
Is a different output test result, s 1,22；
(4) If the user gives an output result
Then the minimum closeness method is used to obtain the output test result
Is selected as the minimum value of
The output result with the minimum expected value
Or selecting the output test result
In
The group with the smallest number is the most input parameters; then, based on a certain main factor, the group of output calculation with the smallest difference value with the main factor is selected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811433170.3A CN109726850B (en) | 2018-11-28 | 2018-11-28 | Method for establishing resultant yarn quality prediction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811433170.3A CN109726850B (en) | 2018-11-28 | 2018-11-28 | Method for establishing resultant yarn quality prediction model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109726850A CN109726850A (en) | 2019-05-07 |
| CN109726850B true CN109726850B (en) | 2020-10-30 |
Family
ID=66295807
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811433170.3A Active CN109726850B (en) | 2018-11-28 | 2018-11-28 | Method for establishing resultant yarn quality prediction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109726850B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1997031262A1 (en) * | 1996-02-21 | 1997-08-28 | Lawson-Hemphill, Incorporated | System and method for electronically evaluating predicted fabric qualities |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108469180B (en) * | 2018-04-09 | 2020-07-21 | 华北理工大学 | Method for establishing sintering end point prediction system based on big data and machine learning |
-
2018
- 2018-11-28 CN CN201811433170.3A patent/CN109726850B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1997031262A1 (en) * | 1996-02-21 | 1997-08-28 | Lawson-Hemphill, Incorporated | System and method for electronically evaluating predicted fabric qualities |
Non-Patent Citations (1)
| Title |
|---|
| 《Predicting the Hairiness of Cotton Rotor Spinning Yarns by Artificial Intelligence》;Morteza Vadood 等;《JOURNAL OF TEXTILES AND POLYMERS》;20180131;第6卷(第1期);正文第15-21页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109726850A (en) | 2019-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416695B (en) | Power load probability density prediction method, system and medium based on deep learning | |
| CN109543737B (en) | Information system health degree evaluation method based on FAHP-FCA combined empowerment | |
| CN106055918B (en) | Method for identifying and correcting load data of power system | |
| CN108898479B (en) | Credit evaluation model construction method and device | |
| CN103853786B (en) | The optimization method and system of database parameter | |
| CN109117380B (en) | Software quality evaluation method, device, equipment and readable storage medium | |
| CN108805743A (en) | A kind of power grid enterprises' sale of electricity company operation Benefit Evaluation Method | |
| CN109144986A (en) | A kind of importance appraisal procedure of industrial equipment data | |
| CN108898259A (en) | Adaptive Evolutionary planning Methods of electric load forecasting and system based on multi-factor comprehensive | |
| CN112907026A (en) | Comprehensive evaluation method based on editable mesh index system | |
| CN112819322A (en) | Power transmission line path scheme evaluation method based on improved fuzzy analytic hierarchy process | |
| CN111126499A (en) | Secondary clustering-based power consumption behavior pattern classification method | |
| CN111008790A (en) | Hydropower station group power generation electric scheduling rule extraction method | |
| CN116127695A (en) | Production line construction method and system based on comprehensive performance evaluation | |
| CN114202252A (en) | Multi-target optimization method for spun yarn process parameters | |
| CN113327047B (en) | Power marketing service channel decision method and system based on fuzzy comprehensive model | |
| CN111177495A (en) | Method for intelligently identifying data content and generating corresponding industry report | |
| Pourchot et al. | To share or not to share: A comprehensive appraisal of weight-sharing | |
| CN115861671A (en) | Double-layer self-adaptive clustering method considering load characteristics and adjustable potential | |
| CN112184412A (en) | Modeling method, device, medium and electronic equipment of credit rating card model | |
| CN114118592B (en) | Smart power grids power consumption end short-term energy consumption prediction system | |
| CN103345552A (en) | Method and device for assessing reliability of power ICT communication network | |
| CN109726850B (en) | Method for establishing resultant yarn quality prediction model | |
| CN117592656A (en) | Carbon footprint monitoring method and system based on carbon data accounting | |
| CN117151488A (en) | Method, system, storage medium and equipment for expanding cold tide and strong wind weather sample |
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 |