CN107633274A - A kind of clustering method of rolling mill vibration operating mode division - Google Patents
A kind of clustering method of rolling mill vibration operating mode division Download PDFInfo
- Publication number
- CN107633274A CN107633274A CN201710957276.2A CN201710957276A CN107633274A CN 107633274 A CN107633274 A CN 107633274A CN 201710957276 A CN201710957276 A CN 201710957276A CN 107633274 A CN107633274 A CN 107633274A
- Authority
- CN
- China
- Prior art keywords
- sample
- rolling mill
- clustering
- local
- follows
- 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.)
- Pending
Links
Abstract
The invention discloses a kind of clustering method of rolling mill vibration operating mode division, comprise the following steps：1）Build the cluster sample of historical state data；2）Cluster sample is filtered, obtains characteristic signal；3) nondimensionalization processing is carried out to the characteristic signal for clustering sample；4) determine based on how popular the local geometry measure information and multiple manifold measuring similarity being locally linear embedding into；5) dimension-reduction treatment is carried out to sample space；6) determine to optimize cluster centre；7) data set after the dimension-reduction treatment is classified using nearest neighbor classifier, the vibration operating mode to rolling machine system divides.The present invention is directed to rolling machine system dynamics, cluster analysis is carried out to rolling mill vibration operating mode using milling train historical state data, establish energy complete characterization rolling machine system vibration performance model, the vibration operating mode complicated and changeable to milling train divides, and playing good technical support to accurate, comprehensive assurance milling train dynamic behavior acts on.
Description
Technical Field
The invention relates to a clustering method for partitioning vibration working conditions of a rolling mill.
Background
China is a large country for producing and consuming steel, and the quality, the yield and the production technical level of steel products play a crucial role in the development of national economy, national defense and other key fields. Since the 21 st century, the steel industry in China has rapidly developed, and by 2016, the crude steel yield in China is continuously 9 years first in the world, and the yield has broken through 8 hundred million tons. However, in the last 10 years, the domestic metallurgical industry is low, the capacity is seriously excessive, and the market competition has entered the albefaction, so that the product structure is greatly adjusted, and the band steel products are more and more abundant. Meanwhile, with the continuous improvement of the level of mechanical equipment, the rolling speed of the rolling mill is faster and faster, the wide application of a hydraulic servo technology, the rapid development of automatic thickness control, the improvement of the quality requirement of a thin strip product and the increase of the processing difficulty of the thin strip product, and the phenomenon of strong vibration of the rolling mill frequently occurs in the rolling process. Vibration of a rolling mill causes vibration marks on the surface of a thin strip, so that the thickness tolerance of the thin strip exceeds the allowable range, vibration is aggravated after vibration marks are generated on the surface of a roller, subsequent rolling is influenced, accidents such as steel piling and strip breakage can be caused in serious cases, the performance and continuous production of equipment efficiency are greatly influenced, and research and development and production of products with higher added values are seriously frustrated.
The rolling mill is a key device for producing steel products, the rolling mill system has multiple operating conditions, the large-scale equipment has the characteristics of variable states, huge load and the like, and the dynamic characteristic difference of the system is obvious in the process of rolling high-strength alloy thin strips with different specifications and different types, so that the clustering of the vibration conditions of the rolling mill is performed, the dynamic behavior of the rolling mill system in the process of rolling high-strength alloy thin strips with different specifications and different types is facilitated to be analyzed, the clustering device is a basis and key for establishing rolling mill system models under different operating conditions, and has important theoretical research significance and engineering application value.
Disclosure of Invention
In order to solve the technical problems, the invention provides a clustering method for partitioning the vibration working conditions of the rolling mill, which can extract the complicated and changeable vibration working condition types of the rolling mill by analyzing the historical state data of the rolling mill and play a good role in theoretically supporting the dynamics behavior of the rolling mill accurately and comprehensively.
The technical scheme adopted by the invention is as follows: a clustering method for partitioning vibration working conditions of a rolling mill comprises the following steps:
(1) Obtaining rolling mill vibration signal, plate and strip specification, technological parameter, control signal and force parameter signal to form cluster sample, the cluster sample is represented by X,
X＝[X 1 ，X 2 ，…X i …，X k ]
where k is the number of signals, X i For the data sequence of the i-th group of signals, i =1,2 … … k, and X i ＝[x 1 ，x 2 ，…，x n ]N is the number of points collected in one pass rolling time of the rolling mill;
(2) Preprocessing the clustering samples by using a multi-scale morphological filtering method to obtain characteristic signals;
(3) Carrying out dimensionless processing on the filtered characteristic signals of the clustering samples;
(4) Determining a multi-manifold local geometry information metric and a multi-manifold similarity metric based on local linear embedding;
(5) Carrying out dimension reduction processing on the sample space;
(6) Determining an optimized clustering center;
(7) And classifying the data set subjected to the dimension reduction treatment by adopting a nearest neighbor classifier, and dividing the vibration working condition of the rolling mill system.
In the clustering method for dividing the vibration working conditions of the rolling mill, the specific operation steps of the step (2) are as follows:
selecting a structural element B close to the characteristics of the signal to be analyzed, and determining the length scale lambda l The specific method comprises the following steps:
firstly, sample signal X i ＝[x 1 ，x 2 ，…，x n ]Removing DC component of signal and searching for signal X i A positive peak sequence is represented as follows:
P＝{p in |in＝1,2,…,n p in the formula, n P Is the total number of peaks;
the time interval T of two adjacent positive peaks is calculated:
T＝{t in |t in ＝p in+1 -p in ,in＝1,2,…,n p -1}，
let the length scale lambda of the multi-scale morphological structure element l Respectively is lambda lmin And λ lmax ，
In the formula (I), the compound is shown in the specification,in order to round down the operator,the rounding-up operator.
The structural element length scale λ is thus l Comprises the following steps:
λ l ＝{λ lmin ,λ lmin +1,…,λ lmax -1,λ lmax }；
determining a height dimension λ of a structuring element h The specific method comprises the following steps:
p n respectively is p nmin And p nmax Define λ h ＝{β·[p nmin +j·(p nmax -p nmin )/(λ lmax -λ lmin )]}
Wherein j =0,1,2, …, λ lmax -λ lmin Beta is the coefficient of proportionality of height, 0<β<1；
Determining the structural element length scale lambda l And height dimension λ h Then, the length scale and the height scale are used for jointly determining the multi-scale morphological analysis structural element lambda B, and the calculation method comprises the following steps:
then using the structural elements and the sample signal X i Averaging after performing the open-close operation and the close-open operation, the formula is as follows:
in the formula, the closed operation is expressed,indicating an on operation.
In the clustering method for dividing the vibration working conditions of the rolling mill, the step (4) is specifically operated as follows:
(4-1) measuring the local geometric structure information of multiple streams based on local linear embedding, which specifically comprises the following operations:
(4-1-1) calculating the minimum linear reconstruction sparse matrix M of all sample points i And further obtaining a minimum linear reconstruction weight matrix M, wherein an objective function of the minimum linear reconstruction weight matrix M is written into the following form:
G J ＝minY(I-M)(I-M) T Y T ，
wherein, I = YY T Y is the projected low dimensional sample space;
(4-1-2) the multi-stream local geometry information metric based on local linear embedding is:
J L ＝Y(I-M)(I-M) T Y T ；
(4-2) establishing heterogeneous nearest multi-manifold similarity measurement, which specifically operates as follows:
(4-2-1) selecting heterogeneous neighbor points for any sample point based on the sample category information;
(4-2-2) calculating the mean value of heterogeneous neighborsWherein: y is ij Represents an arbitrary sample Y i T heterogeneous neighbors of (a);
(4-2-3) establishing a variance matrix of the sample point and the mean value of the heterogeneous neighbor points of each sample point to obtain heterogeneous nearest neighbor-based multi-manifold similarity measurement, which is expressed as follows:
in the clustering method for partitioning rolling mill vibration working conditions, in the step (4-1-1), the minimum linear reconstruction sparse matrix M of the sample points i The calculation method is as follows:
(4-1-1-1) selecting a plurality of adjacent points of the sample points, and establishing a local adjacent map;
(4-1-1-2) performing minimum linear reconstruction according to the local neighborhood graph, and calculating a minimum linear reconstruction sparse matrix M of the sample points i An objective function ofWherein: t is the number of neighbor points of the sample taken when the local neighbor map is built, Y i Is the ith set of samples in the low-dimensional sample space.
In the clustering method for partitioning the vibration working conditions of the rolling mill, the specific operation of the step (5) is as follows:
(5-1) introducing a linear transformation relationship Y = A T Z, establishing a mapping relation between high-dimensional sample data and low-dimensional sample data thereof, wherein A is a linear transformation matrix;
(5-2) transforming the multi-threaded local geometry information metric of step (4-1-4) into:
J L ＝A T Z(I-M)(I-M) T Z T A，
wherein:the mean value of t heterogeneous neighbor points of the ith group of data samples in the dimensionless sample is obtained;
(5-3) transforming the multi-manifold similarity measure of step (4-2-3) into:
(5-4) realizing heterogeneous nearest neighbor-based multi-manifold metric maximization, and establishing an objective function of the NDML method based on multi-manifold local geometric structure information metric and multi-manifold similarity metric, wherein the method specifically comprises the following steps:
(5-4) applying Lagrangian number multiplication to transform the objective function in step (5-4) into a generalized eigenvalue decomposition problem:
(5-5) arranging the eigenvalues from large to small according to the order of magnitude of the eigenvalues, and taking eigenvectors corresponding to the first k eigenvalues to form a linear projection matrix A = [ A ] 1 ，A 2 ，…，A k ]I.e. by the linear projection relation Y = a T And Z, obtaining linear embedding of the high-dimensional data in a low-dimensional space.
In the clustering method for classifying the vibration conditions of the rolling mill, the step (6) specifically operates as follows:
(6-1) initializing a population, in the dimensionality reduction dataset, randomly determining l elements as an initial cluster center M = [ M ] ("M") 1 ，m 2 ，…，m f ，…，m l ]Setting initial speed, iteration times and population scale of particles;
(6-2) calculating Euclidean distance from each group of data to the cluster center:
in the formula, d (Y) i ，M f ) For the element of the ith group in the data set after dimension reduction processing to the initial clustering center M f In Euclidean distance of, wherein M f ∈M；
(6-3) establishing an objective function, namely an effective value of Euclidean distance from the cluster center to each group of data:
(6-4) determining the fitness and the position of the clustering center according to the step (6-3), comparing the fitness and the position with the historical optimal position of the clustering center, and replacing the historical optimal position of the clustering center if the fitness and the position are better;
and (6-5) judging whether the constraint condition is met, if not, returning to the step (6-4), otherwise, obtaining the clustering center as the optimal clustering center.
Compared with the prior art, the invention has the beneficial effects that: according to the invention, aiming at the dynamics of the rolling mill system, the rolling mill vibration working condition is subjected to cluster analysis by using the historical state data of the rolling mill, a vibration characteristic model capable of completely representing the rolling mill system is established, the complex and variable vibration working conditions of the rolling mill are accurately divided, and a good technical support effect is achieved for accurately and comprehensively mastering the dynamics behavior of the rolling mill.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a rolling mill vibration signal according to an embodiment of the present invention; FIG. 2a is a time domain signal before filtering, and FIG. 2b is a frequency domain signal before filtering; fig. 2c is a filtered time domain signal and fig. 2d is a filtered frequency domain signal.
Fig. 3 is a fitness curve for determining an optimized cluster center by applying a particle swarm optimization according to an embodiment of the present invention.
FIG. 4 is a diagram illustrating data distribution of 110-dimensional data samples embedded in a 2-dimensional space according to an embodiment of the present invention.
Fig. 5 is a schematic diagram of a clustering result according to an embodiment of the present invention.
Detailed Description
An embodiment of the method of the present invention is further described in detail with reference to fig. 1.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer and more fully described, it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments of the present invention, belong to the protection scope of the present invention.
The invention provides a clustering method for partitioning vibration working conditions of a rolling mill, which comprises the following specific implementation steps of:
1. and constructing a clustering sample of the historical state data. Acquiring signals such as rolling mill vibration signals, plate and strip specifications, process parameters, control signals, force and energy parameters, and expressing clustering samples by X as follows:
X＝[X 1 ，X 2 ，…X i …，X k ]
where k is the number of signals, X i (i =1,2 … … k) is the data sequence of the i-th group signal, and X i ＝[x 1 ，x 2 ，…，x n ]And n is the number of points collected in one pass rolling time, can also be data of one shift, can be data of one day, and can even be data of one year.
2. And (3) carrying out self-adaptive preprocessing on the historical state data of the rolling mill by using a self-adaptive multi-scale morphological filtering method to obtain a denoised signal. In the implementation process, a section of 2s vibration acceleration signal is selected as a sample, and a time domain signal and a frequency domain signal are subjected to adaptive multi-scale morphological filtering, as shown in fig. 2. The specific operation is as follows:
(2-1) operation mode of selecting structural element B and morphological operator T
The structuring element B should be as close as possible to the characteristics of the sample signal to be analyzed. The structural element B is an isosceles triangle structural element, but the selected dimension lambda is self-adaptive, the original point is positioned at the peak position, and the dimension comprises the length dimension lambda l And height dimension λ h . For the length scale λ l In other words, λ l λ when =1 l B＝{0,1,0}；λ l When =2, λ l B＝{0,1,2,1,0}；λ l λ when =3 l B = {0,1,2,3,2,1,0}; and so on. Height dimension lambda h As well as so.
The basic morphological operator T mainly comprises expansion, corrosion, opening and closing, and the like, and specifically adopts a morphological operation average mode, namely, the basic morphological operators are connected in series, and structural elements and a sample signal X are utilized i Averaging after performing the open-close operation and the close-open operation, the formula is as follows:
wherein, represents a closed operation,indicating an on operation.
In the implementation process, firstly, the opening operation is carried out on the signal to be analyzed, then, the closing operation is carried out, further, the signals of the opening-closing operation and the closing-opening operation are obtained, and the detail signal (the filtered characteristic signal) is obtained after the average operation is carried out.
(2-2) determination of the structural element Length Scale λ l
Firstly, the signal X i ＝[x 1 ，x 2 ，…，x n ](n is the total number of data points) and searching for signal X i A positive peak sequence is represented as follows:
P＝{p in |in＝1,2,…,n p }，
in the formula, n P Is the total number of peaks.
The time interval T of two adjacent positive peaks is calculated:
T＝{t in |t in ＝p in+1 -p in ,in＝1,2,…,n p -1}，
let the length scale lambda of the multi-scale morphological structure element l Respectively is lambda lmin And λ lmax ，
In the formula (I), the compound is shown in the specification,in order to round down the operator,the rounding-up operator.
The structural element length scale λ is thus l Comprises the following steps:
λ l ＝{λ lmin ,λ lmin +1,…,λ lmax -1,λ lmax }
(2-3) determination of the height dimension λ of the structuring element h
Let p be n Respectively is p nmin And p nmax Defined as:
λ h ＝{β·[p nmin +j·(p nmax -p nmin )/(λ lmax -λ lmin )]}
wherein j =0,1,2, …, λ lmax -λ lmin Beta is the height proportionality coefficient (0)<β&And (1) taking 0.65.
(2-4) determination of Multi-Scale morphological analysis structural element λ B
Jointly determining a series of length and height progressively increasing structural elements, lambda, of a multi-scale morphological analysis using a length scale and a height scale l And λ h And the sizes are in one-to-one correspondence from small to large. The calculation method of the multi-scale morphological analysis structural element lambda B comprises the following steps:
(2-5) obtaining analysis results of multi-scale morphology
The tested signal contains components of a plurality of scale features, the scale of each structural element and the noise form scale of the tested signal are self-adaptive, so that the analysis result of each scale can inhibit the noise of the form scale, the analysis results of each scale are accumulated and averaged to obtain the final multi-scale form analysis result, most of the noise components are eliminated, and the feature information is highlighted. The effect of the test signal is shown in fig. 2.
3. On the basis of signals such as rolling mill vibration signals, plate and strip specifications, process parameters, control signals, force and energy parameters and the like, dimensionless standardization processing is carried out on the filtered characteristic signals, and the specific method comprises the following steps:
for signal sequence X i ＝[x 1 ，x 2 ，…，x n ]And (3) carrying out transformation:
wherein the content of the first and second substances,
at this time, the new sequence Z i ＝[z 1 ，z 2 ，…，z n ]Has a mean value of 0, a variance of 1, and is dimensionless, then the dimensionless sample space is Z = [ Z = 1 ，Z 2 ，…Z i …，Z k ]。
4. And determining the spatial multi-streamline local geometric structure information measurement and the multi-manifold similarity measurement of the dimensionless sample.
(4-1) establishing a multi-stream local geometric structure information metric based on Local Linear Embedding (LLE), specifically comprising:
(4-1-1) calculating the minimum linear reconstruction sparse matrix M of all sample points i And further obtaining a minimum linear reconstruction weight matrix M, wherein an objective function of the minimum linear reconstruction weight matrix M is written into the following form:
G J ＝minY(I-M)(I-M) T Y T ，
wherein, I = YY T And Y is the projected low-dimensional sample space.
Minimum linear reconstruction sparse matrix M of sample points i The calculation method is as follows:
(4-1-1-1) selecting a plurality of adjacent points of the sample points, and establishing a local adjacent map;
(4-1-1-2) performing minimum linear reconstruction according to the local neighborhood graph, and calculating a minimum linear reconstruction sparse matrix M of the sample points i An objective function ofWherein: t is the number of neighbor points of the sample taken when the local neighbor map is built, Y i Is the ith set of samples in the low-dimensional sample space.
(4-1-2) the local geometry information metric of multiple streamlines based on LLE is:
J L ＝Y(I-M)(I-M) T Y T 。
(4-2) establishing heterogeneous nearest multi-manifold similarity measurement, specifically operating as follows:
(4-2-1) selecting heterogeneous neighbor points for an arbitrary sample point based on the sample category information.
(4-2-2) meterCalculating mean values of heterogeneous neighborsWherein Y is ij Represents an arbitrary sample Y i T heterogeneous neighbors.
(4-2-3) establishing a variance matrix of the sample point and the mean value of the heterogeneous neighbor points of each sample point to obtain heterogeneous nearest neighbor-based multi-manifold similarity measurement, which is expressed as follows:
5. reducing the dimension of a sample space based on a non-parametric discrimination multi-popular learning method (NDML), and introducing a linear transformation relation Y = A T And Z, establishing a mapping relation between the high-dimensional sample data and the low-dimensional sample data thereof, wherein A is a linear transformation matrix. The specific operation is as follows:
(5-1) transforming the multi-threaded local geometry information metric of step (4-1-4) into:
J L ＝A T Z(I-M)(I-M) T Z T A，
wherein:the mean value of t heterogeneous neighbor points of the ith group of data samples in the dimensionless sample is obtained;
(5-2) transforming the multi-manifold similarity measure of step (4-2-3) into:
(5-3) realizing heterogeneous nearest neighbor-based multi-manifold metric maximization, and establishing an objective function of the NDML method based on multi-manifold local geometric structure information metric and multi-manifold similarity metric, wherein the method specifically comprises the following steps:
(5-4) applying Lagrange number multiplication to convert the target function obtained in the step (5-3) into a generalized eigenvalue decomposition problem, wherein the specific expression is as follows:
(5-5) arranging the eigenvalues from large to small according to the order of the magnitude of the eigenvalues (lambda) 1 ≥λ 2 ≥…≥λ k ) Taking eigenvectors corresponding to the first k eigenvalues to form a linear projection matrix A = [ A ] = 1 ，A 2 ，…，A k ]I.e. by the linear projection relation Y = a T And Z, obtaining linear embedding of the high-dimensional data in a low-dimensional space.
6. Determining an optimal clustering center by adopting a particle swarm algorithm, wherein a particle fitness curve in an implementation process is shown in figure 3, and the implementation steps are as follows:
and (6-1) initializing a clustering center and determining constraint conditions. Initializing a population, and randomly determining l elements as an initial clustering center M = [ M ] in the dimensionality reduction dataset 1 ，m 2 ，…，m f ，…，m l ]And setting the initial speed, the iteration times and the population size of the particles.
(6-2) calculating Euclidean distance of each set of data to the initial particle (cluster center):
in the formula, d (Y) i ，M f ) For the ith group of data in the data set after the dimension reduction processing to the initial clustering center M f In Euclidean distance of, wherein M f ∈M。
(6-3) establishing an objective function, namely effective values of Euclidean distances from the initial particles (cluster centers) to each group of data:
and (6-4) modifying the cluster center. And (4) according to the fitness and the position of the particle (cluster center) determined in the step (6-3), comparing the fitness and the position with the historical optimal position, and preferably replacing the historical optimal position of the particle (cluster center).
(6-5) judging whether the constraint condition is reached, and if not, returning to the step (6-4); otherwise, the obtained clustering center is the optimal clustering center.
7. Determining Euclidean distance d (Y) of each group of sample data i ，M f ) The data set Y after the dimensionality reduction is classified by a nearest neighbor classifier and visualized, as shown in fig. 4. If d (Y) i ，M f ) Is a minimum value, illustrate Y i To the clustering center M f Has the smallest Euclidean distance, then Y i And belongs to class f. The 110-dimensional data is reduced to a plane space, and the clustering result is shown in FIG. 5. The accuracy of the method is found to be about 96% through analysis of a given clustering sample, and the method has a good clustering effect in the implementation process.
It should be noted that: the above embodiments are merely illustrative of the technical solutions of the present invention and not restrictive thereof, and those skilled in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.
Claims (7)
1. A clustering method for partitioning vibration working conditions of a rolling mill comprises the following steps:
(1) Obtaining rolling mill vibration signal, plate and strip specification, technological parameter, control signal and force parameter signal to form cluster sample, the cluster sample is represented by X,
X＝[X 1 ，X 2 ，…X i …，X k ]，
where k is the number of signals, X i For the data sequence of the i-th group of signals, i =1,2 … … k, and X i ＝[x 1 ，x 2 ，…，x n ]N is the number of points collected in one pass of rolling time of the rolling mill;
(2) Preprocessing the clustering samples by using a multi-scale morphological filtering method to obtain characteristic signals;
(3) Carrying out dimensionless processing on the filtered characteristic signals of the clustering samples;
(4) Determining a multi-manifold local geometric structure information metric and a multi-manifold similarity metric based on local linear embedding;
(5) Carrying out dimension reduction processing on the sample space;
(6) Determining an optimized clustering center;
(7) And classifying the data set subjected to the dimensionality reduction by adopting a nearest neighbor classifier, and dividing the vibration working condition of the rolling mill system.
2. The clustering method for classifying the vibration working conditions of the rolling mill according to claim 1, wherein the specific operation steps of the step (2) are as follows:
selecting a structural element B which is close to the characteristics of a signal to be analyzed, wherein the structural element B is an upper triangular structural element, a semicircular structural element or a sinusoidal structural element; determining a length scale λ l The specific method comprises the following steps:
firstly, sample signal X i ＝[x 1 ，x 2 ，…，x n ]Removing DC component of signal and searching for signal X i A positive peak sequence is represented as follows:
P＝{p in |in＝1,2,…,n p }
in the formula, n P Is the total number of peaks;
the time interval T of two adjacent positive peaks is calculated:
T＝{t in |t in ＝p in+1 -p in ,in＝1,2,…,n p -1}
let the length scale lambda of the multi-scale morphological structure element l Respectively is lambda lmin And λ lmax
In the formula (I), the compound is shown in the specification,in order to round down the operator,the rounding-up operator.
The structural element length scale λ is thus l Comprises the following steps:
λ l ＝{λ lmin ,λ lmin +1,…,λ lmax -1,λ lmax }；
determining a structuring element height dimension λ h The specific method comprises the following steps:
p n respectively is p nmin And p nmax Define λ h ＝{β·[p nmin +j·(p nmax -p nmin )/(λ lmax -λ lmin )]}，
Wherein j =0,1,2, …, λ lmax -λ lmin Beta is the coefficient of proportionality of height, 0<β<1；
Determining the structural element length scale lambda l And height dimension λ h Then, the length scale and the height scale are used for jointly determining the multi-scale morphological analysis structural element lambda B, and the calculation method comprises the following steps:
then using the structural elements and the sample signal X i Averaging after performing the open-close operation and the close-open operation, the formula is as follows:
in the formula, the closed operation is expressed,indicating an open operation.
3. The clustering method for classifying the vibration working conditions of the rolling mill according to claim 1 or 2, wherein the specific operation steps of the step (3) are as follows:
for signal sequence X i ＝[x 1 ，x 2 ，…，x n ]And (3) carrying out transformation:
wherein:
obtaining dimensionless signal sequence Z i ＝[z 1 ，z 2 ，…，z n ]Obtaining a dimensionless sample as follows:
Z＝[Z 1 ，Z 2 ，…Z i …，Z k ]。
4. the clustering method for classifying the vibration working conditions of the rolling mill according to claim 3, wherein the step (4) is specifically operated as follows:
(4-1) measuring the local geometric structure information of multiple flow lines based on local linear embedding, which specifically comprises the following steps:
(4-1-1) calculating the minimum linear reconstruction sparse matrix M of all sample points i And further obtaining a minimum linear reconstruction weight matrix M, wherein an objective function of the minimum linear reconstruction weight matrix M is written into the following form:
G J ＝minY(I-M)(I-M) T Y T ，
wherein, I = YY T Y is the projected low-dimensional sample space;
(4-1-2) the multi-stream local geometry information metric based on local linear embedding is:
J L ＝Y(I-M)(I-M) T Y T ；
(4-2) establishing heterogeneous nearest multi-manifold similarity measurement, which specifically comprises the following operations:
(4-2-1) selecting heterogeneous neighbor points for an arbitrary sample point based on the sample category information;
(4-2-2) calculating the mean value of heterogeneous neighborsWherein Y is ij Represents an arbitrary sample Y i T heterogeneous neighbors of (a);
(4-2-3) establishing a variance matrix of the mean value of the sample point and the heterogeneous neighbor points of the sample point for each sample point to obtain heterogeneous nearest neighbor-based multi-manifold similarity measurement, which is expressed as follows:
5. the clustering method for vibration condition division of rolling mill according to claim 4, wherein in the step (4-1-1), the minimum linear reconstruction sparse matrix M of the sample points i The calculation method is as follows:
(4-1-1-1) selecting a plurality of adjacent points of the sample point, and establishing a local adjacent graph;
(4-1-1-2) performing minimum linear reconstruction according to the local neighborhood graph, and calculating a minimum linear reconstruction sparse matrix M of the sample points i An objective function ofWherein: t is the number of neighbor points of the sample taken when the local neighbor map is built, Y i Is the ith set of samples in the low-dimensional sample space.
6. The clustering method for classifying the vibration working conditions of the rolling mill according to claim 4 or 5, wherein the step (5) specifically operates as follows:
(5-1) introducing a linear transformation relationship Y = A T Z, establishing a mapping relation between high-dimensional sample data and low-dimensional sample data thereof, wherein A is a linear transformation matrix;
(5-2) transforming the multi-threaded local geometry information metric of step (4-1-4) into:
J L ＝A T Z(I-M)(I-M) T Z T A，
wherein:the mean value of t heterogeneous neighbor points of the ith group of data samples in the dimensionless sample is obtained;
(5-3) transforming the multi-manifold similarity measure of step (4-2-3) into:
(5-4) realizing the multi-manifold metric maximization based on heterogeneous nearest neighbor, and establishing an objective function of the NDML method based on the multi-manifold local geometric structure information metric and the multi-manifold similarity metric, wherein the method specifically comprises the following steps:
(5-4) applying Lagrange number multiplication, and (5-4) converting the target function into a generalized eigenvalue decomposition problem:
(5-5) arranging the eigenvalues from large to small according to the order of magnitude of the eigenvalues, and taking eigenvectors corresponding to the first k eigenvalues to form a linear projection matrix A = [ A ] 1 ，A 2 ，…，A k ]I.e. by linear projection relationsY＝A T And Z, obtaining linear embedding of the high-dimensional data in a low-dimensional space.
7. The rolling mill vibration condition clustering method according to claim 6, wherein the step (6) specifically operates as follows:
(6-1) initializing a population, in the dimensionality reduction dataset, randomly determining l elements as an initial cluster center M = [ M ] ("M") 1 ，m 2 ，…，m f ，…，m l ]Setting initial speed, iteration times and population scale of particles;
(6-2) calculating Euclidean distance of each group of data to the cluster center:
in the formula, d (Y) i ，M f ) For the element of the ith group in the data set after dimension reduction processing to the initial clustering center M f In Euclidean distance of where M f ∈M；
(6-3) establishing an objective function, namely an effective value of Euclidean distance from the cluster center to each group of data:
(6-4) determining the fitness and the position of the clustering center according to the step (6-3), comparing the fitness and the position with the historical optimal position, and replacing the clustering center if the fitness and the position are better;
and (6-5) judging whether the constraint condition is met, if not, returning to the step (6-4), otherwise, obtaining the clustering center as the optimal clustering center.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710957276.2A CN107633274A (en) | 2017-10-16 | 2017-10-16 | A kind of clustering method of rolling mill vibration operating mode division |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710957276.2A CN107633274A (en) | 2017-10-16 | 2017-10-16 | A kind of clustering method of rolling mill vibration operating mode division |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107633274A true CN107633274A (en) | 2018-01-26 |
Family
ID=61104241
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710957276.2A Pending CN107633274A (en) | 2017-10-16 | 2017-10-16 | A kind of clustering method of rolling mill vibration operating mode division |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107633274A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108573285A (en) * | 2018-04-25 | 2018-09-25 | 河海大学常州校区 | A kind of experimental parameter clustering method towards dredging energy consumption |
| CN112541538A (en) * | 2020-12-11 | 2021-03-23 | 中国电建集团华东勘测设计研究院有限公司 | Dam abnormity monitoring data automatic identification method under poor information condition |
| CN113657484A (en) * | 2021-08-13 | 2021-11-16 | 济南大学 | Method for dividing and identifying typical working conditions of cement grate cooler |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101334786A (en) * | 2008-08-01 | 2008-12-31 | 浙江大学 | Formulae neighborhood based data dimensionality reduction method |
-
2017
- 2017-10-16 CN CN201710957276.2A patent/CN107633274A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101334786A (en) * | 2008-08-01 | 2008-12-31 | 浙江大学 | Formulae neighborhood based data dimensionality reduction method |
Non-Patent Citations (4)
| Title |
|---|
| 李波等: "《高维数据的流形学习分析方法》", 31 May 2016, 武汉大学出版社 * |
| 温浩: "张量子空间人脸识别算法研究", 《中国博士学位论文全文数据库信息科技辑》 * |
| 王勇: "基于流形学习的分类与聚类方法及其应用研究", 《中国博士学位论文全文数据库信息科技辑》 * |
| 章立军等: "自适应多尺度形态学分析及其在轴承故障诊断中的应用", 《北京科技大学学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108573285A (en) * | 2018-04-25 | 2018-09-25 | 河海大学常州校区 | A kind of experimental parameter clustering method towards dredging energy consumption |
| CN112541538A (en) * | 2020-12-11 | 2021-03-23 | 中国电建集团华东勘测设计研究院有限公司 | Dam abnormity monitoring data automatic identification method under poor information condition |
| CN112541538B (en) * | 2020-12-11 | 2023-01-31 | 中国电建集团华东勘测设计研究院有限公司 | Dam abnormity monitoring data automatic identification method under poor information condition |
| CN113657484A (en) * | 2021-08-13 | 2021-11-16 | 济南大学 | Method for dividing and identifying typical working conditions of cement grate cooler |
| CN113657484B (en) * | 2021-08-13 | 2024-02-09 | 济南大学 | Method for dividing and identifying typical working conditions of cement grate cooler |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hao et al. | A steel surface defect inspection approach towards smart industrial monitoring | |
| Chen et al. | The fire recognition algorithm using dynamic feature fusion and IV-SVM classifier | |
| CN107633274A (en) | A kind of clustering method of rolling mill vibration operating mode division | |
| CN109255094B (en) | Commercial truck quality estimation method based on SVR-Adaboost improved algorithm | |
| Zhao et al. | Adaptive logit adjustment loss for long-tailed visual recognition | |
| CN102693452A (en) | Multiple-model soft-measuring method based on semi-supervised regression learning | |
| CN108416775B (en) | Ore granularity detection method based on deep learning | |
| CN111723876B (en) | Load curve integrated spectrum clustering method considering double-scale similarity | |
| CN111250548B (en) | Board convexity prediction method based on kernel partial least square combined support vector machine | |
| CN105740378B (en) | Digital pathology full-section image retrieval method | |
| CN109858386A (en) | A kind of microalgae cell recognition methods based on fluorescence microscope images | |
| CN110991547A (en) | Image significance detection method based on multi-feature optimal fusion | |
| Fang et al. | Identification of apple leaf diseases based on convolutional neural network | |
| Lin et al. | Steel surface defects detection based on deep learning | |
| CN111627505B (en) | Cluster structure type identification method | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN104200226B (en) | Particle filter method for tracking target based on machine learning | |
| CN109784142B (en) | Hyperspectral target detection method based on conditional random projection | |
| Liu et al. | Unbalanced classification method using least squares support vector machine with sparse strategy for steel surface defects with label noise | |
| CN111127407B (en) | Fourier transform-based style migration forged image detection device and method | |
| CN110701487B (en) | KPCA and Cas-SVDD-based multi-working-condition pipeline leakage detection method | |
| CN111914471A (en) | Rectification tower fault detection method based on rapid nuclear independent component analysis | |
| WO2023173770A1 (en) | Laplacian eigenmaps-based dimension reduction and visualization method for high-throughput calcium signal | |
| Zhul et al. | A fast quantum clustering approach for cancer gene clustering | |
| CN113780343A (en) | Bilateral slope DTW distance load spectrum clustering method based on LTTB dimension reduction |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180126 |