CN111027706B - Method and device for measuring surface roughness of household paper - Google Patents
Method and device for measuring surface roughness of household paper Download PDFInfo
- Publication number
- CN111027706B CN111027706B CN201911135254.3A CN201911135254A CN111027706B CN 111027706 B CN111027706 B CN 111027706B CN 201911135254 A CN201911135254 A CN 201911135254A CN 111027706 B CN111027706 B CN 111027706B
- Authority
- CN
- China
- Prior art keywords
- surface roughness
- paper
- model
- household paper
- household
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B21/00—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant
- G01B21/30—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant for measuring roughness or irregularity of surfaces
Abstract
The invention discloses a method for measuring the surface roughness of household paper, which comprises the following steps: acquiring original data of the household paper to be detected, and preprocessing the original data by adopting an abnormal value detection method to obtain preprocessed data; screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors; establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model; adjusting parameters of the living paper surface roughness measurement model according to preset parameters to obtain an optimal living surface roughness measurement model; and inputting the maximum influence factors and the preprocessing data into the optimal living surface roughness measurement model to obtain the roughness of the surface of the living paper to be measured. The method for measuring the surface roughness of the household paper can accurately measure the surface roughness of the household paper, and is beneficial to improving the service efficiency of the household paper.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a method and a device for measuring surface roughness of household paper.
Background
The production of tissue is a complex process. The production of the household paper is mainly divided into four sub-processes: forming, dewatering, wrinkling and coiling. The pulp after beating is spouted from the head box, gets into the woollen blanket, after the preliminary dehydration of shaping portion, gets into the drying part, further dewaters in the drying part. After entering the drying part, the wet paper web is subjected to preliminary dewatering by pressing and then is dried in contact with a yankee dryer. The surface of the yankee dryer is coated with a layer of special chemical reagent, a cylinder sticking agent and a stripping agent in advance, and the coating nozzle sprays the chemical reagent on the dry part of the yankee dryer to form a chemical protective layer. The internal steam heats the surface of the yankee dryer, the heat brings away the moisture on the paper surface through the yankee dryer, and the moisture on the paper is left about 5 percent. When the paper web reaches the other end of the drying cylinder, the wrinkling scraper is in contact with the paper on the surface of the drying cylinder, so that the paper web is separated from the surface of the drying cylinder, the paper web and the scraper generate violent interaction at the same time, the paper is wrinkled, and in the wrinkling process, a fiber network is damaged, so that the paper web shrinks, the tensile property of the paper is improved, and the bulk, the softness and the water absorption of the paper are increased.
At present, the existing method for measuring the surface roughness of the household paper cannot accurately measure the surface roughness of the household paper, and cannot measure the household paper meeting the requirement to be further processed into various products, so that the use efficiency of the household paper cannot be improved.
Disclosure of Invention
The method and the device for measuring the surface roughness of the household paper provided by the embodiment of the invention can accurately measure the surface roughness of the household paper, thereby being beneficial to measuring the household paper meeting the requirement to further process the household paper into various products and being beneficial to improving the service efficiency of the household paper.
In one aspect, an embodiment of the present invention provides a method for measuring surface roughness of household paper, including:
acquiring original data of the household paper to be detected, and preprocessing the original data by adopting an abnormal value detection method to obtain preprocessed data;
screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors;
establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
adjusting parameters of the living paper surface roughness measurement model according to preset parameters to obtain an optimal living surface roughness measurement model;
and inputting the maximum influence factors and the preprocessed data into the optimal living surface roughness measurement model to obtain the roughness of the surface of the living paper to be measured.
Further, the acquiring of the original data of the household paper to be tested specifically includes:
measuring the to-be-measured household paper by using a household paper wrinkle quality analyzer to obtain the original data of the to-be-measured household paper, wherein the original data comprises the surface roughness, the wrinkle depth and the wrinkle frequency of the household paper.
Further, the preprocessing is performed on the original data by using an abnormal value detection method to obtain preprocessed data, specifically:
and detecting an abnormal value of the original data by adopting a 3 sigma detection method, and deleting the abnormal value to obtain preprocessed data.
Further, the influencing factors of the surface roughness of the household paper comprise the selection, proportioning and beating of fiber raw materials, the surface state of a drying cylinder, chemicals, parameters and states of a scraper and the running state of a paper machine;
screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors, wherein the method specifically comprises the following steps:
keeping other influence factors unchanged, and researching the influence of each factor on the roughness of the household paper one by one according to a control variable method to obtain a preset number of influence factors which have the greatest influence on the roughness of the household paper and record the influence factors as first influence factors;
and judging the influence of the first influence factor on the roughness of the household paper according to a correlation analysis method to obtain the maximum influence factor of a second preset number.
Further, the establishing of the life paper surface roughness measurement model based on the gradient enhanced regression tree model specifically includes:
selecting preprocessing data with a preset proportional quantity as a data set;
initializing a gradient enhancement regression tree model according to the loss function of the data set to obtain an initialized model;
calculating a negative gradient value of the loss function according to the initialization model, and taking the negative gradient value as an estimated value of a residual error;
generating a corresponding regression tree according to the estimated value of the residual error, and calculating the gradient descending step length of the Mth tree;
and updating the initialization model according to the gradient descent step length and the learning efficiency, and establishing a domestic paper surface roughness measurement model based on a gradient enhancement regression tree model.
Further, the preset parameters comprise a learning rate, a tree depth and iteration times; the method comprises the following steps of adjusting parameters of the household paper surface roughness measurement model according to preset parameters to obtain an optimal household surface roughness measurement model, and specifically comprises the following steps:
dividing the data set into a preset number of subsets;
testing each subset, and taking the rest subsets of the selected subsets as a training set to finish the testing and training of a preset number;
determining the value ranges of the iteration times and the learning rate according to the sample size of the data set, training and testing by adopting the depths of different numerical trees, and acquiring the depth of the tree with the minimum error of a preset number of test results as an alternative depth;
and adjusting the learning rate and the iteration number according to the alternative depths until the precision of the surface roughness measurement model of the household paper cannot be improved any more, so as to obtain the optimal surface roughness measurement model of the household paper.
On the other hand, another embodiment of the invention provides a device for measuring the surface roughness of household paper, which comprises a preprocessing module, a screening module, a model establishing module, a parameter adjusting module and a measuring module;
the preprocessing module is used for acquiring original data of the household paper to be detected and preprocessing the original data by adopting an abnormal value detection method to obtain preprocessed data;
the screening module is used for screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors;
the model establishing module is used for establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
the parameter adjusting module is used for adjusting parameters of the living paper surface roughness measuring model according to preset parameters to obtain an optimal living surface roughness measuring model;
and the measuring module is used for inputting the maximum influence factors and the preprocessing data into the optimal living surface roughness measuring model to obtain the roughness of the surface of the living paper to be measured.
Further, the screening module is specifically configured to keep other influence factors unchanged, and explore the influence of each factor on the roughness of the household paper one by one according to a controlled variable method to obtain a preset number of influence factors having the greatest influence on the roughness of the household paper, and record the influence factors as first influence factors;
and judging the influence of the first influence factor on the roughness of the household paper according to a correlation analysis method to obtain the maximum influence factor of a second preset number.
Further, the model building module comprises a selecting unit, an initializing unit, a first calculating unit, a second calculating unit and a model updating unit;
the selecting unit is used for selecting the preprocessed data with the quantity in a preset proportion as a data set;
the initialization unit is used for initializing the gradient enhanced regression tree model according to the loss function of the data set to obtain an initialization model;
the first calculation unit calculates a negative gradient value of the loss function according to the initialization model, and uses the negative gradient value as an estimated value of a residual error;
the second calculating unit is configured to generate a corresponding regression tree according to the estimated value of the residual error, and calculate a gradient descent step length of the mth tree;
and the model updating unit is used for updating the initialization model according to the gradient descent step length and the learning efficiency and establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model.
Further, the preset parameters comprise a learning rate, a tree depth and an iteration number; the parameter adjusting module comprises a dividing unit, a testing and training unit, an obtaining unit and an adjusting unit;
the dividing unit is used for dividing the data set into a preset number of subsets;
the test and training unit is used for testing each subset, and using the rest subsets of the selected subsets as training sets to finish the test and training of preset quantity;
the acquisition unit is used for determining the value ranges of the iteration times and the learning rate according to the sample size of the data set, training and testing by adopting the depths of different numerical trees, and acquiring the depth of the tree with the minimum error of the test result as the alternative depth;
and the adjusting unit is used for adjusting the learning rate and the iteration number according to the alternative depths until the precision of the surface roughness measurement model of the household paper cannot be improved any more, so as to obtain the optimal surface roughness measurement model of the household paper.
The method for measuring the surface roughness of the household paper provided by the embodiment of the invention can accurately measure the surface roughness of the household paper, thereby being beneficial to measuring the household paper meeting the requirements to further process the household paper into various products and improving the service efficiency of the household paper.
Drawings
FIG. 1 is a schematic flow chart of a method for measuring the surface roughness of household paper provided by the invention;
FIG. 2 is a schematic flowchart of step S3 in an embodiment of a method for measuring surface roughness of household paper according to the present invention;
FIG. 3 is a schematic flowchart of step S4 in an embodiment of a method for measuring surface roughness of household paper according to the present invention;
FIG. 4 is a schematic diagram of correlation analysis of a method for measuring surface roughness of household paper provided by the invention;
FIG. 5 is a schematic diagram of a decision tree flow of a method for measuring surface roughness of household paper according to the present invention;
FIG. 6 shows the learning rate and R of the method for measuring the surface roughness of household paper provided by the present invention 2 A relationship diagram;
FIG. 7 shows the tree depth and R of the method for measuring the surface roughness of the household paper provided by the invention 2 A relationship diagram;
FIG. 8 shows data and R of a classifier for measuring surface roughness of household paper according to the method for measuring surface roughness of household paper of the present invention 2 A relationship diagram;
FIG. 9 is a schematic structural diagram of a device for measuring surface roughness of household paper according to 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 derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Please refer to fig. 1-8:
the present invention provides a first embodiment.
The embodiment of the invention provides a method for measuring the surface roughness of household paper, which comprises the following steps:
s1, acquiring original data of to-be-detected household paper, and preprocessing the original data by adopting an abnormal value detection method to obtain preprocessed data;
s2, screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors;
s3, establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
s4, adjusting parameters of the surface roughness measurement model of the household paper according to preset parameters to obtain an optimal household surface roughness measurement model;
and S5, inputting the maximum influence factors and the preprocessing data into the optimal living surface roughness measurement model to obtain the roughness of the surface of the living paper to be measured.
According to the embodiment of the invention, the quality detector of the household paper is used for detecting the household paper to be detected to obtain the original data, and the method for detecting the abnormal value is used for preprocessing the original data, so that the abnormal value in the original data can be deleted, more accurate original data can be obtained, and the result of measuring the surface roughness of the household paper can be effectively improved; the influence factors influencing the surface roughness in life are screened by a control variable method and a correlation analysis method respectively, a preset number of influence factors can be screened, the screened influence factors are factors influencing the surface roughness of the household paper most, and the accuracy of a measurement result can be further improved by screening the influence factors; by establishing the surface roughness measurement model of the household paper based on the gradient enhanced regression tree model, the calculated residual error is reduced towards the gradient direction, so that the influence of the residual error on the calculation result is reduced, the accuracy of the calculation result is improved, and the accuracy and the reliability of the measurement result are improved; and adjusting and optimizing the established household paper surface roughness measurement model according to preset parameters to obtain an optimal household surface roughness measurement model, inputting the screened maximum influence factors and the preprocessed data into the optimal household paper surface roughness measurement model, and effectively improving the household paper surface roughness measurement result output by the optimal household surface roughness measurement model, so that the household paper meeting the requirements can be measured to be further processed into various products, and the service efficiency of the household paper is further improved.
As a specific implementation manner of the embodiment of the present invention, the obtaining of the original data of the to-be-measured household paper specifically includes:
and measuring the to-be-measured household paper by using a household paper wrinkle quality analyzer to obtain the original data of the to-be-measured household paper.
In the embodiment of the present invention, the raw data includes the surface roughness, wrinkle depth and wrinkle frequency of the paper for daily use.
As a specific implementation manner of the embodiment of the present invention, the method for detecting an abnormal value is adopted to preprocess the original data to obtain preprocessed data, and specifically, the method includes:
and detecting abnormal values of the original data by adopting a 3 sigma detection method, and deleting the abnormal values to obtain preprocessed data.
In the embodiment of the invention, after the original data is acquired, in order to ensure the quality of the data, the data needs to be cleaned, abnormal values in the original data are processed, and abnormal data with isolation points, outliers and the like which are obviously different from other data in the number of the original data are detected and deleted. The method for detecting abnormal data comprises the following steps: 3 sigma test method, T test method and cluster analysis method. The embodiment of the invention adopts a 3 sigma test method, the 3 sigma test method is based on probability statistics theory, and for a certain variable, the probability that the deviation value of the mean value exceeds 3 sigma does not exceed 11.11 percent, so if the absolute value of the difference between the variable value and the mean value exceeds 3 sigma, the variable value is considered to be an abnormal value.
The data standardization processing is a form of data transformation, and can eliminate the problems caused by different variables in dimension. Normalization of data, for a data matrix X n×m Mean value of mean 1×m Standard deviation of σ 1×m Normalized by zero-mean:
according to the embodiment of the invention, the abnormal value in the original data can be accurately detected by the 3 sigma detection method, and the detected abnormal value is deleted, so that the reliability of the original data is favorably improved, and the measurement result of the surface roughness of the household paper is favorably improved.
As a specific implementation manner of the embodiment of the present invention, the influence factors of the surface roughness of the household paper are screened according to a controlled variable method and a correlation analysis method, so as to obtain a preset number of maximum influence factors, which specifically are:
keeping other influence factors unchanged, and researching the influence of each factor on the roughness of the household paper one by one according to a control variable method to obtain the maximum influence factor of a preset number on the roughness of the household paper, and recording the maximum influence factor as a first influence factor;
and judging the influence of the first influence factor on the roughness of the household paper according to a correlation analysis method to obtain the maximum influence factor of a second preset number.
In the embodiment of the invention, the influencing factors of the surface roughness of the household paper comprise the selection, proportioning and beating of fiber raw materials, the surface state of a drying cylinder, chemicals, the parameters and the state of a scraper, the running state of a paper machine and the like; the method comprises the steps of keeping other influence factors unchanged, researching the influence of each factor on the roughness of the household paper one by one according to a controlled variable method to obtain the influence degree of each influence factor on the roughness of the household paper, sequencing according to the influence degree, selecting the influence factors with the preset number arranged in the front as first influence factors according to the sequencing sequence, and primarily screening the influence factors with the largest influence on the roughness of the household paper. According to the embodiment of the invention, the influence of the first influence factor on the roughness of the household paper is judged by the correlation analysis method according to the correlation analysis method, and the maximum influence factor of the second preset number is obtained. The correlation analysis method comprises a mutual information estimation method, a Spearman rank correlation coefficient method, a distance correlation method and the like, the Pearson correlation coefficient is used for analysis in the embodiment of the invention, and the expression of the correlation coefficient r is as follows:
the correlation coefficient r can reflect the correlation between the two variables x and y, typically between-1 and-1. The larger the absolute value of r, the more closely x correlates with y. The closer r is to 1,x, the stronger the positive correlation with y is; the closer r is to-1,x, the stronger the negative correlation with y; r equals 0,x is uncorrelated with y.
The embodiment of the invention respectively uses a single-factor experimental method and a characteristic selection method to research factors influencing the surface roughness of the household paper, and can determine that the factors such as the use time of the wrinkling scraper, the wrinkling rate and the like have larger influence on the surface roughness of the household paper. The determined factors having a large influence on the surface roughness of the household paper are used as the input amount of the surface roughness measurement model of the household paper of the present invention.
As a specific implementation manner of the embodiment of the present invention, a life paper surface roughness measurement model based on a gradient enhanced regression tree model is established, specifically:
s31, selecting preprocessing data with a preset proportional quantity as a data set;
s32, initializing the gradient enhancement regression tree model according to the loss function of the data set to obtain an initialized model;
s33, calculating to obtain a negative gradient value of the loss function according to the initialization model, and taking the negative gradient value as an estimated value of the residual error;
s34, generating a corresponding regression tree according to the estimated value of the residual error, and calculating the gradient descending step length of the Mth tree;
and S35, updating the initialization model according to the gradient descent step length and the learning efficiency, and establishing a domestic paper surface roughness measurement model based on the gradient enhancement regression tree model.
In the embodiment of the present invention, the gradient enhanced regression tree model (GBRT model) is a kind of Boosting algorithm, and is also an improvement of the Boosting algorithm. The original Boosting algorithm is that each sample is assigned with an equal weight value at the beginning of the algorithm, and all the basic learners have the same importance degree at first. The model obtained in each training can lead the estimation of the data points to have difference, and after each step is finished, the weight value is processed, and the weight of the misclassification point is increased, and the weight of the correct classification point is reduced, so that some points can be seriously concerned if being misclassified for a long time, and a very high weight is given. After N iterations are carried out, N simple basic learners are obtained, the N simple basic learners are combined, and the weighted basic learners are weighted (the weighted value of the basic classifier with larger error rate is smaller, and the weighted value of the basic classifier with smaller error rate is larger) or voted to obtain a final model.
Please refer to fig. 5, which illustrates a decision tree process generated by leaf nodes according to an embodiment of the present invention. The specific generation process of the GBRT model is as follows: suppose the dataset is D = { (x) 1 ，y 1 )，(x 2 ，y 2 )，...，(x n ，y n ) H, the loss function is L (y, f (x)), the leaf node number of each regression tree is J, and the input space is divided into J disjoint areas R 1m ，R 2m ，...，R jm And estimating a constant value b for each region jm Regression tree g m (x) Is formulated as:
initializing the gradient enhancement regression tree model according to the loss function of the data set to obtain an initialized model:
calculating according to the initialization model to obtain a negative gradient value of the loss function, and taking the negative gradient value as an estimated value of residual error r im The expression of (a) is as follows:
generating a corresponding regression tree according to the estimated value of the residual error, and calculating the gradient descending step length of the Mth tree;
generating a regression tree g according to the estimated value of the residual error m (x) The input space of the m-th tree is partitioned into J disjoint regions R 1m ，R 2m ，...，R jm And, calculating the step size of gradient descent:
updating the initialization model according to the gradient descending step length and the learning efficiency, and establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
f m (x)＝f m-1 (x)+I r ×ρg m (x i ) (6)
wherein, I r Indicating the learning efficiency.
In the embodiment of the invention, the prediction precision of the GBRT model is mainly influenced by the quantity M of the regression trees and the learning efficiency I r The influence of each base model on the final result can be effectively reduced by setting the learning rate to prevent the model from being over-fitted.
In the embodiment of the invention, the GBRT model is greatly different from the conventional Boosting, and the core of the GBRT model is that each calculation is to reduce the residual error of the last time, and a new model can be established in the Gradient (Gradient) direction in which the residual error is reduced. The new model established according to the GBRT model can reduce the residual error of the previous model to the gradient direction, and can effectively improve the calculation precision, thereby effectively improving the measurement accuracy of the surface roughness of the household paper.
As a specific implementation manner of the embodiment of the present invention, the preset parameters include a learning rate, a tree depth, and an iteration number; adjusting parameters of the living paper surface roughness measurement model according to preset parameters to obtain an optimal living surface roughness measurement model, which specifically comprises the following steps:
s41, dividing the data set into subsets with preset quantity;
s42, testing each subset, and taking the rest subsets of the selected subsets as training sets to finish the testing and training of preset quantity;
s43, determining the value ranges of the iteration times and the learning rate according to the sample size of the data set, training and testing by adopting the depths of different numerical trees, and obtaining the depth of the tree with the minimum error of the test result of the preset number as an alternative depth;
and S44, adjusting the learning rate and the iteration number according to the alternative depths until the precision of the surface roughness measurement model of the household paper cannot be improved any more, and obtaining the optimal surface roughness measurement model of the household paper.
In the embodiment of the invention, a cross-validation method is adopted, a data set of original data is disordered and divided into C subsets, one subset is selected as a test set each time, samples of the rest C-1 subsets are used as training sets, and training and testing are completed; selecting another subset as a test set for the next test to finish training and testing; repeating the training and the testing for C times, and bending the testing of each subset and the corresponding training; the soft measurement model for the surface roughness of the household paper comprises 3 parameters to be determined, namely learning rate eta, tree depth d and iteration number N d 。
Presetting adjustment ranges of 3 parameters to be determined, wherein the preset learning rate is more than 0.01 and less than 1; the depth of the tree is 2 < d < 10; number of iterations N d Is more than 1; fixing the iteration times and the learning rate according to the size of the sample size, and selecting a proper tree depth, which specifically comprises the following steps: r when observing depths d of different trees 2 Determining the minimum error which can be achieved after multiple iterations when the depth is different; the minimum error is smaller and is determined as the alternative depth; adjusting learning rate eta and iteration number N under proper tree depth d And when the learning rate is reduced every time, the iteration times are expanded in the same proportion until the model precision cannot be improved.
The embodiment of the invention has the following beneficial effects:
according to the embodiment of the invention, the quality detector of the household paper is used for detecting the household paper to be detected to obtain the original data, and the method for detecting the abnormal value is used for preprocessing the original data, so that the abnormal value in the original data can be deleted, more accurate original data can be obtained, and the result of measuring the surface roughness of the household paper can be effectively improved; the influence factors influencing the surface roughness in life are screened by a control variable method and a correlation analysis method respectively, a preset number of influence factors can be screened, the screened influence factors are the factors influencing the surface roughness of the household paper most, and the accuracy of the measurement result can be further improved by screening the influence factors; by establishing the surface roughness measurement model of the household paper based on the gradient enhanced regression tree model, the calculated residual error is reduced towards the gradient direction, so that the influence of the residual error on the calculation result is reduced, the accuracy of the calculation result is improved, and the accuracy and the reliability of the measurement result are improved; and adjusting and optimizing the established household paper surface roughness measurement model according to preset parameters to obtain an optimal household surface roughness measurement model, inputting the screened maximum influence factors and the preprocessed data into the optimal household paper surface roughness measurement model, and effectively improving the household paper surface roughness measurement result output by the optimal household surface roughness measurement model, so that the household paper meeting the requirements can be measured to be further processed into various products, and the service efficiency of the household paper is further improved.
Please refer to fig. 4-9:
the present invention provides a second embodiment.
Referring to fig. 9, an embodiment of the present invention provides a device for measuring surface roughness of household paper, including a preprocessing module 101, a screening module 201, a model building module 301, a parameter adjusting module 401, and a measuring module 501;
the preprocessing module 101 is configured to obtain original data of the paper for daily use to be detected, and preprocess the original data by using an abnormal value detection method to obtain preprocessed data;
the screening module 201 is used for screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors;
the model establishing module 301 is used for establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
the parameter adjusting module 401 is configured to adjust parameters of the living paper surface roughness measurement model according to preset parameters to obtain an optimal living surface roughness measurement model;
and the measuring module 501 is used for inputting the maximum influence factor and the preprocessing data into the optimal living surface roughness measuring model to obtain the roughness of the surface of the living paper to be measured.
According to the embodiment of the invention, the preprocessing module 101 is used for detecting the to-be-detected household paper by using the household paper quality detector to obtain the original data, and preprocessing the original data by using the abnormal value detection method, so that abnormal values in the original data can be deleted, more accurate original data can be obtained, and the result of measuring the surface roughness of the household paper can be effectively improved; the screening module 201 screens the influence factors influencing the surface roughness in life by respectively adopting a control variable method and a correlation analysis method, so that a preset number of influence factors can be screened out, the screened influence factors are the factors influencing the surface roughness of the household paper to the maximum extent, and the accuracy of the measurement result can be further improved by screening the influence factors; the model establishing module 301 is used for establishing a life paper surface roughness measurement model based on a gradient enhanced regression tree model, so that the calculated residual error is reduced towards the gradient direction, the influence of the residual error on the calculation result is reduced, the accuracy of the calculation result is improved, and the accuracy and the reliability of the measurement result are improved; the parameter adjusting module 401 is used for adjusting and optimizing the established household paper surface roughness measurement model according to preset parameters, an optimal household surface roughness measurement model can be obtained, the measurement module 501 is used for inputting the screened maximum influence factors and the preprocessed data into the optimal household paper surface roughness measurement model, the household paper surface roughness measurement result output by the optimal household surface roughness measurement model can be effectively improved, the household paper meeting the requirements can be measured to be further processed into various products, and the service efficiency of the household paper is further improved.
As a specific implementation manner of the embodiment of the present invention, the screening module 201 is specifically configured to keep other influencing factors unchanged, and explore the influence of each factor on the roughness of the household paper one by one according to a control variable method, to obtain a preset number of influencing factors that have the greatest influence on the roughness of the household paper, and to record the influencing factors as first influencing factors;
and judging the influence of the first influence factor on the roughness of the household paper according to a correlation analysis method to obtain the maximum influence factor of a second preset number.
In the embodiment of the invention, the influencing factors of the surface roughness of the household paper comprise the selection, proportioning and beating of fiber raw materials, the surface state of a drying cylinder, chemicals, the parameters and the state of a scraper, the running state of a paper machine and the like; the influence of each factor on the roughness of the household paper is explored one by one according to a control variable method by keeping other influencing factors unchanged, the influence degree of each influencing factor on the roughness of the household paper is obtained, sorting is carried out according to the influence degree, the influencing factors with the preset number arranged in the front are selected as first influencing factors in the sorting sequence, and the influencing factors with the largest influence on the roughness of the household paper are screened primarily. According to the embodiment of the invention, the influence of the first influence factor on the roughness of the household paper is judged by the correlation analysis method according to the correlation analysis method, and the maximum influence factor of the second preset number is obtained. The correlation analysis method comprises a mutual information estimation method, a Spearman rank correlation coefficient method, a distance correlation method and the like, the Pearson correlation coefficient is used for analysis in the embodiment of the invention, and the expression of the correlation coefficient r is as follows:
the correlation coefficient r can reflect the correlation between the two variables x and y, typically between-1 and-1. The larger the absolute value of r, the more closely x correlates with y. The closer r is to 1,x, the stronger the positive correlation with y is; the closer r is to-1,x, the stronger the negative correlation with y; r equals 0,x is uncorrelated with y.
The embodiment of the invention respectively uses a single-factor experimental method and a characteristic selection method to research factors influencing the surface roughness of the household paper, and can determine that the factors such as the use time of the wrinkling scraper, the wrinkling rate and the like have larger influence on the surface roughness of the household paper. The determined factors having a large influence on the surface roughness of the household paper are used as the input amount of the surface roughness measurement model of the household paper of the present invention.
As a specific implementation manner of the embodiment of the present invention, the model establishing module 301 includes a selecting unit, an initializing unit, a first calculating unit, a second calculating unit, and a model updating unit;
the selecting unit is used for selecting the preprocessing data with the quantity of a preset proportion as a data set;
the initialization unit is used for initializing the gradient enhancement regression tree model according to the loss function of the data set to obtain an initialization model;
the first calculation unit calculates a negative gradient value of the loss function according to the initialization model, and the negative gradient value is used as an estimated value of the residual error;
the second calculation unit is used for generating a corresponding regression tree according to the estimated value of the residual error and calculating the gradient descending step length of the Mth tree;
and the model updating unit is used for updating the initialization model according to the gradient descent step length and the learning efficiency and establishing a living paper surface roughness measurement model based on the gradient enhanced regression tree model.
In the embodiment of the present invention, the gradient enhanced regression tree model (GBRT model) is a kind of Boosting algorithm, and is also an improvement of the Boosting algorithm. The original Boosting algorithm is that each sample is assigned with an equal weight value at the beginning of the algorithm, and all the basic learners have the same importance degree at first. The model obtained in each training can lead the estimation of the data points to have difference, and after each step is finished, the weight value is processed, and the weight of the misclassification point is increased, and the weight of the correct classification point is reduced, so that some points can be seriously concerned if being misclassified for a long time, and a very high weight is given. After N iterations are performed, N simple basic learners are obtained, the N simple basic learners are combined, and the weighted learners are weighted (the weighted value of the basic classifier with larger error rate is smaller, and the weighted value of the basic classifier with smaller error rate is larger) or voted to obtain a final model.
Please refer to FIG. 5, which illustrates the leaf node generation according to the embodiment of the present inventionThe decision tree process of (1). The specific generation process of the GBRT model is as follows: suppose the dataset is D = { (x) 1 ，y 1 )，(x 2 ，y 2 )，...，(x n ，y n ) H, the loss function is L (y, f (x)), the leaf node number of each regression tree is J, and the input space is divided into J disjoint areas R 1m ，R 2m ，...，R jm And estimating a constant value b for each region jm Regression tree g m (x) Is formulated as:
initializing the gradient enhancement regression tree model according to the loss function of the data set to obtain an initialized model;
calculating according to the initialization model to obtain a negative gradient value of the loss function, and taking the negative gradient value as an estimated value of residual error r im The expression of (a) is as follows:
generating a corresponding regression tree according to the estimated value of the residual error, and calculating the gradient descending step length of the Mth tree;
generating a regression tree g according to the estimated value of the residual error m (x) The input space of the m-th tree is partitioned into J disjoint regions R 1m ，R 2m ，...，R jm And, calculating the step size of gradient descent:
updating the initialization model according to the gradient descending step length and the learning efficiency, and establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
f m (x)＝f m-1 (x)+I r ×ρg m (x i ) (6)
wherein, I r Indicating the learning efficiency.
In the embodiment of the invention, the prediction precision of the GBRT model is mainly influenced by the quantity M of the regression trees and the learning efficiency I r The influence of each base model on the final result can be effectively reduced by setting the learning rate to prevent the model from being overfitted.
In the embodiment of the invention, the GBRT model is greatly different from the conventional Boosting, and the core of the GBRT model is that each calculation is to reduce the residual error of the last time, and a new model can be established in the Gradient (Gradient) direction in which the residual error is reduced. The new model established according to the GBRT model can reduce the residual error of the previous model to the gradient direction, and can effectively improve the calculation precision, thereby effectively improving the measurement accuracy of the surface roughness of the household paper.
As a specific implementation manner of the embodiment of the present invention, the preset parameters include a learning rate, a tree depth, and an iteration number; the parameter adjusting module 401 includes a dividing unit, a testing and training unit, an obtaining unit and an adjusting unit:
a dividing unit for dividing the data set into a preset number of subsets;
the test and training unit is used for testing each subset, and using the rest subsets of the selected subsets as training sets to finish the test and training of preset quantity;
the acquisition unit is used for determining the value ranges of the iteration times and the learning rate according to the sample size of the data set, training and testing by adopting the depths of different numerical trees, and acquiring the depth of the tree with the minimum error of the test result as an alternative depth;
and the adjusting unit is used for adjusting the learning rate and the iteration number according to the alternative depths until the precision of the surface roughness measurement model of the household paper cannot be improved any more, so that the optimal surface roughness measurement model of the household paper is obtained.
In the embodiment of the invention, a cross-validation method is adopted, a data set of original data is disturbed and divided into C subsets, one subset is selected as a test set each time, samples of the other C-1 subsets are used as a training set, and training and testing are completed; selecting another subset as a test set for the next test to finish training and testing; repeating the training and testing for C times, and the testing and corresponding training of each subset of the deflection; the soft measurement model for the surface roughness of the household paper comprises 3 parameters to be determined, namely learning rate eta, tree depth d and iteration number N d 。
Presetting adjustment ranges of 3 parameters to be determined, wherein the preset learning rate is more than 0.01 and less than 1; the depth of the tree is 2 < d < 10; number of iterations N d Is more than 1; fixing the iteration times and the learning rate according to the size of the sample size, and selecting a proper tree depth, which specifically comprises the following steps: r when observing depths d of different trees 2 Determining the minimum error which can be achieved after multiple iterations when the depth is different; the minimum error is smaller and is determined as the alternative depth; adjusting learning rate eta and iteration number N under proper tree depth d And when the learning rate is reduced every time, the iteration times are expanded in the same proportion until the model precision cannot be improved.
The embodiment of the invention has the following beneficial effects:
according to the embodiment of the invention, the preprocessing module 101 is used for detecting the to-be-detected household paper by using the household paper quality detector to obtain the original data, and preprocessing the original data by using the abnormal value detection method, so that abnormal values in the original data can be deleted, more accurate original data can be obtained, and the result of measuring the surface roughness of the household paper can be effectively improved; the screening module 201 is used for screening the influence factors influencing the surface roughness in life by respectively adopting a control variable method and a correlation analysis method, so that a preset number of influence factors can be screened out, the screened influence factors are the factors influencing the surface roughness of the household paper to the maximum extent, and the accuracy of the measurement result can be further improved by screening the influence factors; the model building module 301 builds a gradient-enhanced regression tree model-based living paper surface roughness measurement model, so that the calculated residual error is reduced towards the gradient direction, thereby reducing the influence of the residual error on the calculation result, being beneficial to improving the accuracy of the calculation result and further being beneficial to improving the accuracy and the reliability of the measurement result; parameter adjustment and optimization are carried out on the established household paper surface roughness measurement model through the parameter adjustment module 401 according to preset parameters, an optimal household surface roughness measurement model can be obtained, the measurement module 501 inputs the screened maximum influence factors and the preprocessed data into the optimal household paper surface roughness measurement model, the household paper surface roughness measurement result output by the optimal household surface roughness measurement model can be effectively improved, the household paper meeting the requirements can be measured to be further processed into various products, and the service efficiency of the household paper is further improved
The foregoing is a preferred embodiment of the present invention, and it should be noted that it would be apparent to those skilled in the art that various modifications and enhancements can be made without departing from the principles of the invention, and such modifications and enhancements are also considered to be within the scope of the invention.
Claims (10)
1. A method for measuring the surface roughness of household paper is characterized by comprising the following steps:
acquiring original data of the household paper to be detected, and preprocessing the original data by adopting an abnormal value detection method to obtain preprocessed data;
screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors;
establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
adjusting parameters of the living paper surface roughness measurement model according to preset parameters to obtain an optimal living surface roughness measurement model;
and inputting the maximum influence factors and the preprocessing data into the optimal living surface roughness measurement model to obtain the roughness of the surface of the living paper to be measured.
2. The method for measuring the surface roughness of the household paper as claimed in claim 1, wherein the obtaining of the raw data of the household paper to be measured specifically comprises:
measuring the living paper to be detected by adopting a living paper wrinkle quality analyzer to obtain the original data of the living paper to be detected, wherein the original data comprises the surface roughness, wrinkle depth and wrinkle frequency of the living paper.
3. The method for measuring the surface roughness of the household paper as claimed in claim 1, wherein the method for detecting the abnormal value is used for preprocessing the original data to obtain preprocessed data, and specifically comprises:
and detecting an abnormal value of the original data by adopting a 3 sigma detection method, and deleting the abnormal value to obtain preprocessed data.
4. The method for measuring the surface roughness of the household paper as claimed in claim 1, wherein the influence factors of the surface roughness of the household paper are screened according to a control variable method and a correlation analysis method to obtain the maximum influence factors, and specifically the method comprises the following steps:
keeping other influence factors unchanged, and exploring the influence of each factor on the roughness of the household paper one by one according to a controlled variable method to obtain a preset number of influence factors which have the largest influence on the roughness of the household paper and record the influence factors as first influence factors;
and judging the influence of the first influence factor on the roughness of the household paper according to a correlation analysis method to obtain the maximum influence factor of a second preset number.
5. The method for measuring the surface roughness of the household paper as claimed in claim 1, wherein the establishing of the gradient enhanced regression tree model-based household paper surface roughness measurement model specifically comprises:
selecting preprocessing data with a preset proportional quantity as a data set;
initializing a gradient enhancement regression tree model according to the loss function of the data set to obtain an initialized model;
calculating a negative gradient value of the loss function according to the initialization model, and taking the negative gradient value as an estimated value of a residual error;
generating a corresponding regression tree according to the estimated value of the residual error, and calculating the gradient descending step length of the Mth tree;
and updating the initialization model according to the gradient descent step length and the learning efficiency, and establishing a domestic paper surface roughness measurement model based on a gradient enhancement regression tree model.
6. The method for measuring the surface roughness of household paper according to claim 5, wherein the preset parameters include a learning rate, a tree depth and an iteration number; the method comprises the following steps of adjusting parameters of the household paper surface roughness measurement model according to preset parameters to obtain an optimal household surface roughness measurement model, and specifically comprises the following steps:
dividing the data set into a preset number of subsets;
testing each subset, and taking the rest subsets of the selected subsets as a training set to finish the testing and training of a preset number;
determining the value ranges of the iteration times and the learning rate according to the sample size of the data set, training and testing by adopting the depths of different numerical trees, and acquiring the depth of the tree with the minimum error of a preset number of test results as an alternative depth;
and adjusting the learning rate and the iteration number according to the alternative depth until the precision of the domestic paper surface roughness measurement model can not be improved any more, so as to obtain the optimal domestic paper surface roughness measurement model.
7. The device for measuring the surface roughness of the household paper is characterized by comprising a preprocessing module, a screening module, a model establishing module, a parameter adjusting module and a measuring module;
the preprocessing module is used for acquiring original data of the household paper to be detected and preprocessing the original data by adopting an abnormal value detection method to obtain preprocessed data;
the screening module is used for screening the influence factors of the surface roughness of the household paper according to a control variable method and a correlation analysis method to obtain the maximum influence factors;
the model establishing module is used for establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model;
the parameter adjusting module is used for adjusting parameters of the living paper surface roughness measuring model according to preset parameters to obtain an optimal living surface roughness measuring model;
and the measuring module is used for inputting the maximum influence factors and the preprocessing data into the optimal living surface roughness measuring model to obtain the roughness of the surface of the living paper to be measured.
8. The device for measuring the surface roughness of the household paper as claimed in claim 7, wherein the screening module is specifically configured to keep other influencing factors unchanged, and explore the influence of each factor on the roughness of the household paper one by one according to a controlled variable method to obtain a preset number of influencing factors which have the greatest influence on the roughness of the household paper and record the influencing factors as first influencing factors;
and judging the influence of the first influence factor on the roughness of the household paper according to a correlation analysis method to obtain the maximum influence factor of a second preset number.
9. The apparatus for measuring surface roughness of household paper according to claim 7, wherein the model establishing module comprises a selecting unit, an initializing unit, a first calculating unit, a second calculating unit and a model updating unit;
the selecting unit is used for selecting the preprocessing data with the preset proportional quantity as a data set;
the initialization unit is used for initializing the gradient enhanced regression tree model according to the loss function of the data set to obtain an initialization model;
the first calculation unit calculates a negative gradient value of the loss function according to the initialization model, and uses the negative gradient value as an estimated value of a residual error;
the second calculating unit is configured to generate a corresponding regression tree according to the estimated value of the residual error, and calculate a gradient descent step length of the mth tree;
and the model updating unit is used for updating the initialization model according to the gradient descent step length and the learning efficiency and establishing a domestic paper surface roughness measurement model based on a gradient enhanced regression tree model.
10. The apparatus for measuring surface roughness of household paper as claimed in claim 9, wherein the preset parameters include learning rate, tree depth and iteration number; the parameter adjusting module comprises a dividing unit, a testing and training unit, an obtaining unit and an adjusting unit;
the dividing unit is used for dividing the data set into a preset number of subsets;
the test and training unit is used for testing each subset, and using the rest subsets of the selected subsets as training sets to finish the test and training of preset quantity;
the acquisition unit is used for determining the value ranges of the iteration times and the learning rate according to the sample size of the data set, training and testing by adopting the depths of different numerical trees, and acquiring the depth of the tree with the minimum error of the test result as the alternative depth;
and the adjusting unit is used for adjusting the learning rate and the iteration number according to the alternative depths until the precision of the surface roughness measurement model of the household paper cannot be improved any more, so that the optimal surface roughness measurement model of the household paper is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911135254.3A CN111027706B (en) | 2019-11-19 | 2019-11-19 | Method and device for measuring surface roughness of household paper |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911135254.3A CN111027706B (en) | 2019-11-19 | 2019-11-19 | Method and device for measuring surface roughness of household paper |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111027706A CN111027706A (en) | 2020-04-17 |
| CN111027706B true CN111027706B (en) | 2023-03-10 |
Family
ID=70205828
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911135254.3A Active CN111027706B (en) | 2019-11-19 | 2019-11-19 | Method and device for measuring surface roughness of household paper |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111027706B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292051A (en) * | 2017-07-07 | 2017-10-24 | 湘潭大学 | A kind of carbide chip chemically-mechanicapolish polishes the Forecasting Methodology of surface roughness |
| CN109828532A (en) * | 2019-01-29 | 2019-05-31 | 贵州大学 | A kind of Prediction of Surface Roughness method and process parameter optimizing method based on GA-GBRT |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8527246B1 (en) * | 2010-08-27 | 2013-09-03 | Agilent Technologies, Inc. | Method for modeling conductor surface roughness |
-
2019
- 2019-11-19 CN CN201911135254.3A patent/CN111027706B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292051A (en) * | 2017-07-07 | 2017-10-24 | 湘潭大学 | A kind of carbide chip chemically-mechanicapolish polishes the Forecasting Methodology of surface roughness |
| CN109828532A (en) * | 2019-01-29 | 2019-05-31 | 贵州大学 | A kind of Prediction of Surface Roughness method and process parameter optimizing method based on GA-GBRT |
Non-Patent Citations (1)
| Title |
|---|
| 孟子薇 ; 洪蒙纳 ; 李继庚 ; 满奕 ; .基于梯度增强回归树算法的磨浆过程打浆度软测量模型.2019,(01),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111027706A (en) | 2020-04-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416695B (en) | Power load probability density prediction method, system and medium based on deep learning | |
| CN108280552B (en) | Power load prediction method and system based on deep learning and storage medium | |
| CN111222290B (en) | Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment | |
| CN109407649B (en) | Fault type matching method based on fault characteristic variable selection | |
| CN108073158A (en) | Based on PCA and KNN density algorithm Wind turbines Method for Bearing Fault Diagnosis | |
| CN109992921B (en) | On-line soft measurement method and system for thermal efficiency of boiler of coal-fired power plant | |
| CN109409425B (en) | Fault type identification method based on neighbor component analysis | |
| CN108399434B (en) | Analysis and prediction method of high-dimensional time series data based on feature extraction | |
| US5942689A (en) | System and method for predicting a web break in a paper machine | |
| CN116167640B (en) | LCP film production quality detection data analysis method and system | |
| CN112116198A (en) | Data-driven process industrial state perception network key node screening method | |
| CN111027706B (en) | Method and device for measuring surface roughness of household paper | |
| CN112001115B (en) | Soft measurement modeling method of semi-supervised dynamic soft measurement network | |
| CN109389313B (en) | Fault classification diagnosis method based on weighted neighbor decision | |
| CN110910528A (en) | Method and device for predicting tensile strength of paper sheet | |
| CN112069621B (en) | Method for predicting residual service life of rolling bearing based on linear reliability index | |
| CN106055883B (en) | Transient stability evaluation input feature validity analysis method based on improved Sammon mapping | |
| CN110967184B (en) | Gearbox fault detection method and system based on vibration signal distribution characteristic recognition | |
| Mamatov et al. | Mathematical model for calculating the temperature of cotton in a direct-flow drying drum | |
| CN109583323B (en) | Subway vibration signal identification method based on door control circulation unit | |
| CN115308102A (en) | Paper air permeability measuring method and paper air permeability measuring system | |
| CN108038256B (en) | Real-time temperature estimation method in wet cloth heat setting process | |
| CN111024557B (en) | Soft measurement method for water absorption of paper sheet | |
| CN115081937B (en) | Method for solving operational parameters of sizing machine based on quality analysis | |
| CN117612644B (en) | Air safety evaluation method and system |
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 |