CN106839769A - Based on the semi-supervised local global electric melting magnesium furnace fault monitoring method of multiple manifold - Google Patents
Based on the semi-supervised local global electric melting magnesium furnace fault monitoring method of multiple manifold Download PDFInfo
- Publication number
- CN106839769A CN106839769A CN201710181754.5A CN201710181754A CN106839769A CN 106839769 A CN106839769 A CN 106839769A CN 201710181754 A CN201710181754 A CN 201710181754A CN 106839769 A CN106839769 A CN 106839769A
- Authority
- CN
- China
- Prior art keywords
- data
- semi
- supervised
- manifold
- dimensional space
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F27—FURNACES; KILNS; OVENS; RETORTS
- F27B—FURNACES, KILNS, OVENS, OR RETORTS IN GENERAL; OPEN SINTERING OR LIKE APPARATUS
- F27B14/00—Crucible or pot furnaces
- F27B14/08—Details peculiar to crucible or pot furnaces
- F27B14/20—Arrangement of controlling, monitoring, alarm or like devices
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F27—FURNACES; KILNS; OVENS; RETORTS
- F27M—INDEXING SCHEME RELATING TO ASPECTS OF THE CHARGES OR FURNACES, KILNS, OVENS OR RETORTS
- F27M2001/00—Composition, conformation or state of the charge
- F27M2001/01—Charges containing mainly non-ferrous metals
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F27—FURNACES; KILNS; OVENS; RETORTS
- F27M—INDEXING SCHEME RELATING TO ASPECTS OF THE CHARGES OR FURNACES, KILNS, OVENS OR RETORTS
- F27M2003/00—Type of treatment of the charge
- F27M2003/13—Smelting
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of based on the semi-supervised local global electric melting magnesium furnace fault monitoring method of multiple manifold, the method:The raw data set X of smelting process of electro-fused magnesia furnace is obtained, the raw data set according to smelting process of electro-fused magnesia furnace is set up based on the semi-supervised part of multiple manifold and the global process monitoring model for keeping, the test data x of Real-time Collection smelting process of electro-fused magnesia furnace0, using the process monitoring model of the semi-supervised local and overall situation holding based on multiple manifold to test data x0Carry out fault type diagnosis;The inventive method has taken into account the part and global structure of data simultaneously, learn to prevent local mistake simultaneously, consider the Partial diversity information between homogeneous data, it is ultimately constructed go out the semi-supervised fault diagnosis of multiple manifold optimal objective function, with good monitoring effect.
Description
Technical field
The invention belongs to fault detection and diagnosis technical field, and in particular to one kind based on the semi-supervised part of multiple manifold-it is complete
The electric melting magnesium furnace fault monitoring method of office.
Background technology
Electric melting magnesium furnace belongs to submerged slag furnace apparatus, as shown in figure 1, mainly including electric melting magnesium furnace body, main circuit equipment
With the part of control device three.Wherein, the body of electric melting magnesium furnace is mainly made up of body of heater, electrode jaw, electrode lifting mechanism etc..
Body of heater is made up of furnace shell and steel plate of furnace hearth, and furnace shell is generally circular in cross section, slightly taper, for ease of molten stone roller shelling, is welded on furnace shell wall
There are suspension ring.Electrode jaw can holding electrode, be easy to cable transmission electric current.In fusion process, with the fusing of furnace charge, stove
Pit level can rise steadily, and operator will at any time lift the purpose that electrode reaches adjustment arc length.Electrode lifting mechanism can
Electrode is vertically moved up and down along guide rail, reduce the situation that electrode is rocked, thermal power partition equilibrium in holding furnace, so as to reduce Lou
The generation of stove accident.Transformer and suspension belong to main circuit equipment, and are provided with control room, coordination electrode lifting on stove side.Stove
Moving cart is had, effect is will to melt the frit for completing to move on to fixed station, and cooling is come out of the stove.
The major product of electric melting magnesium furnace is fused magnesite, and the fusion process of fused magnesite is a sufficiently complex mistake
Journey, it is influenced by many factors.In the production process of fused magnesite, the main ranks such as melting, row's analysis, purification, crystallization can be experienced
Section, contains various physical and chemical changes.Can exist simultaneously due to the continuous fusing of furnace charge in smelting process, in molten bath solid
The variforms such as state, gaseous state, molten state, simultaneous oxidation magnesium powder can produce a large amount of gases in fusing, easily cause spray stove phenomenon,
The smelting process of fused magnesite is as shown in Figure 2.
Because current China majority electric melting magnesium furnace smelting process automaticity is also than relatively low, often in process of production can
Break down and the frequent situation about occurring of abnormal conditions.Wherein, because electrode actuator breaks down during traveling electrode
Or the reason such as double swerve shakiness causes the furnace wall of electrode distance electric melting magnesium furnace excessively near, the body of heater of electric melting magnesium furnace can be caused to melt
Change, that is, leak accident occur,
Further, since a large amount of gases that furnace charge is produced below in heating process can cause just in case cannot rapidly discharge
Furnace charge eruption, gently then has a strong impact on product yield and quality in body of heater, and economic loss, life that is heavy then threatening people are brought to enterprise
Life safety.This is accomplished by detecting the exception and failure occurred in electric melting magnesium furnace smelting process in time.
The content of the invention
In view of the shortcomings of the prior art, the present invention proposes a kind of electric melting magnesium furnace based on the semi-supervised part-overall situation of multiple manifold
Fault monitoring method.
The technical scheme is that:
A kind of electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold, comprises the following steps:
Step 1:Obtain the data and the unmarked sample class of u groups of the l group echo sample class of smelting process of electro-fused magnesia furnace
Data, composition raw data set X=(XL, XU)∈Rm×n, wherein, XL=[x1, x2..., xl], XU=[xl+1, xl+2...,
xl+u], l+u=n,It is flag data,It is Unlabeled data, m is data dimension;
Step 2:Raw data set according to smelting process of electro-fused magnesia furnace is set up semi-supervised local and global based on multiple manifold
The process monitoring model of holding;
Step 2.1:Raw data set X is mapped to by a core Hilbert sky for higher-dimension by nonlinear mapping function Φ
Between in H, obtain High dimensional space data collection Φ (X);
Step 2.2:Using each sample point in semi-supervised kernel locality preserving projections Algorithm for Solving High dimensional space data collection Φ (X)
Φ(xi) k neighbour domain point, obtain sample point Φ (xi) nearly Neighbourhood set Nk(Φ(xi)), i ∈ 1,2 ..., n;
Step 2.2.1:Using each sample in semi-supervised kernel locality preserving projections Algorithm for Solving High dimensional space data collection Φ (X)
Point Φ (xi) and neighborhood sample point Φ (xjThe distance between);
Step 2.2.2:Each sample point Φ (x in High dimensional space data collection Φ (X)i) and neighborhood sample point Φ (xj) between
Distance in the minimum k neighborhood point of selected distance, constitute sample point Φ (xi) nearly Neighbourhood set Nk(Φ(xi))。
Step 2.3:Using the weight matrix P of Partial Reconstruction method reconstruct higher dimensional space High dimensional space data collection Φ (X);
Step 2.3.1:The c class sample data streams for making flag data in raw data set X are
Wherein, c is the classification of exemplar, 1≤c≤M, ncIt is the number of c class sample data streams,
Step 2.3.2:According to sample point Φ (xi) and its nearly Neighbourhood set Nk(Φ(xi)) power of Partial Reconstruction flag data
The weight matrix of value matrix and Unlabeled data, obtains the weight matrix at data point and Neighborhood Number strong point in Different categories of samples data flow
Pij;
Step 2.3.3:The weights at data point and Neighborhood Number strong point in Different categories of samples data flow are solved using least square method
Matrix Pij, obtain the weight matrix U of flag datairWith the weight matrix of Unlabeled data
Step 2.4:The mapping square on Different categories of samples data manifold is projected to according to kernel method solution High dimensional space data collection
Battle array WΦ={ W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A;
Step 2.4.1:Set up the semi-supervised kernel localized target function on Different categories of samples data manifold in higher dimensional space;
Step 2.4.2:Set up the semi-supervised global objective function on Different categories of samples data manifold in higher dimensional space;
Step 2.4.3:With reference to semi-supervised kernel localized target function and semi-supervised global objective function, solved according to kernel method
Project to the mapping matrix W on Different categories of samples data manifoldΦ={ W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A, WcΦ
It is the mapping matrix in c class sample data manifolds.
Step 2.5:Mapping matrix W according to High dimensional space data collection on Different categories of samples data manifoldΦ={ W1Φ,
W2Φ..., WcΦ..., WMΦAnd coefficient matrices A, set up the low-dimensional embedded coordinate y={ y of High dimensional space data1, y2...,
yc..., yM, that is, obtain the process monitoring model of and overall situation holding semi-supervised local based on multiple manifold;
Step 3:The test data x of Real-time Collection smelting process of electro-fused magnesia furnace0, using based on multiple manifold it is semi-supervised local and
The process monitoring model that the overall situation keeps is to test data x0Carry out fault type diagnosis;
Step 3.1:The test data x of Real-time Collection smelting process of electro-fused magnesia furnace0;
Step 3.2:By nonlinear mapping function Φ by test data x0It is mapped to a core Hilbert space for higher-dimension
In H, test data High dimensional space data Φ (x are obtained0);
Step 3.3:Construction test data High dimensional space data Φ (x0) on Different categories of samples data manifold low-dimensional insertion
Coordinate yc 0;
Step 3.4:Solve test data High dimensional space data Φ (x0) in embedded space in each sample data manifold
On reconstructed error value errorc(Φ(x0));
Step 3.5:By test data High dimensional space data Φ (x0) in embedded space in each sample data manifold
Reconstructed error value errorc(Φ(x0)) sample type of data manifold c belonging to minimum value as the test data failure
Type.
Beneficial effects of the present invention:
The present invention proposes a kind of electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold, traditional
Data classification algorithm assumes that greatly data are located in single manifold, when data set includes multiclass and different category structure, single stream
Shape assumes it is difficult to ensure that the performance of classification, based on this, present invention assumes that being located in a single manifold respectively per class data, pin
To above mentioned problem, we have proposed a kind of electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold, half supervises
Superintend and direct locality preserving projections (MM-SSLPP) algorithm principle and be so that in similar sample that distance reduces, distance increase between foreign peoples's sample,
Classification learning process is instructed using unmarked sample simultaneously, the intrinsic manifold of low-dimensional of Various types of data is found;
Conventional method takes into consideration only the partial structurtes information of data, but have ignored the overall structure of data, may result in
Gained must project subspace and can not well portray the global association of sample, the global guarantor of the inventive method combination PCA methods
Thought is held, a kind of improved semi-supervised local and global method for diagnosing faults for keeping of multiple manifold is formd, the method is simultaneously simultaneous
The part and global structure of data are turned round and look at, while learning to prevent local mistake, it is contemplated that the partial diverse between homogeneous data
Property information, it is ultimately constructed go out the semi-supervised fault diagnosis of multiple manifold optimal objective function, and carried out by taking Tennessee Eastman Process as an example
Emulation experiment and analysis, verify that this chapter carries the good monitoring effect of algorithm.
Brief description of the drawings
Fig. 1 is electric melting magnesium furnace structural representation;
Wherein, 1 is transformer, and 2 is short net, and 3 is electrode jaw, and 4 is electrode, and 5 is furnace shell, and 6 is car body, and 7 is electric arc,
8 is furnace charge, and 9 is controller;
Fig. 2 is the smelting process flow chart of fused magnesite;
Fig. 3 is the electric melting magnesium furnace malfunction monitoring based on the semi-supervised part-overall situation of multiple manifold in the specific embodiment of the invention
The flow chart of method;
Fig. 4 is semi-supervised kernel pca method, semi-supervised kernel locality preserving projections side in the specific embodiment of the invention
Method and the inventive method data characteristics figure;
Wherein, (a) is the perspective view of the first and second principal component vectors of semi-supervised kernel pca method data;
B () is the perspective view of the first two feature of semi-supervised kernel locality preserving projections method data;
C () is the projecting direction figure of the inventive method first, second;
Fig. 5 is that the precision of different high-dimensional lower distinct methods and different marker samples numbers in the specific embodiment of the invention is shown
It is intended to;
Wherein, (a) for it is different it is high-dimensional under semi-supervised core pivot element analysis method, semi-supervised kernel locality preserving projections method and
The precision schematic diagram of the inventive method;
1- is the precision schematic diagram of different high-dimensional lower the inventive method;
2- is the precision schematic diagram of different high-dimensional lower semi-supervised kernel locality preserving projections methods;
3- be it is different it is high-dimensional under semi-supervised core pivot element analysis method precision schematic diagram;
B () is the precision schematic diagram of different high-dimensional lower the inventive method difference marker samples numbers;
4- is the precision schematic diagram of the exemplar of the inventive method 25%;
5- is the precision schematic diagram of the exemplar of the inventive method 20%;
6- is the precision schematic diagram of the exemplar of the inventive method 15%;
Fig. 6 is semi-supervised kernel pca method, semi-supervised kernel locality preserving projections side in the specific embodiment of the invention
Method and the inventive method monitoring effect figure;
Wherein, (a) is semi-supervised kernel pca method monitoring effect figure;
B () is semi-supervised kernel locality preserving projections method monitoring effect figure;
C () is the inventive method monitoring effect figure.
Specific embodiment
The specific embodiment of the invention is described in detail below in conjunction with the accompanying drawings.
A kind of electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold, as shown in figure 3, including with
Lower step:
Step 1:Obtain the data and the unmarked sample class of u groups of the l group echo sample class of smelting process of electro-fused magnesia furnace
Data, composition raw data set X=(XL, XU)∈Rm×n, wherein, XL=[x1, x2..., xl], XU=[xl+1, xl+2...,
xl+u], l+u=n,It is flag data,It is Unlabeled data, m is data dimension.
In present embodiment, initial data is input voltage value, three-phase electricity flow valuve, furnace temperature value, electrode relative position, gas
13 process variables such as flow, pressure, weight, in production process, when current setting value is constant, feed particles length change compares
When big, when electrode can be caused to move, the gap size produced between raw material is improper, and air pressure can be lost due to the discharge of gas in stove
Go balance.Cause electrode in stove, bath surface big ups and downs, so as to cause arc resistance acute variation, liquation occur with gas
The phenomenon outside stove is sprayed together.Above-mentioned abnormal gas exhaust operating mode is used as failure 1;
Bath surface rapid increase during the reduction of raw material fusing point, causes arc resistance to reduce, and current value is raised comparatively fast, if now
Current setting value is constant, then current following error is larger or very big, when bath surface is long lasting for rapid increase, can lead
Causing the impurity in molten bath cannot thoroughly separate out, and cause deterioration in quality, single ton of energy consumption to raise, and above-mentioned heating cycle of crossing is used as event
Barrier 2;
The situation of nominal situation is represented with failure 0, is due to obtaining the data with knowledge information in the industrial process
It is extremely difficult, still choose respectively per 50 samples of class and 3000 groups of data of unknown running status are used as training data
Carry out off-line modeling.
That is the c class sample data streams of flag data are in raw data set XWherein, c is mark
The classification of signed-off sample sheet, in the present embodiment, the sampled data to being obtained under normal, failure 1 and the running status of failure 2 is used to build
Mould is analyzed, i.e. 1≤c≤3, ncIt is the number of c class sample data streams,
Step 2:Raw data set according to smelting process of electro-fused magnesia furnace is set up semi-supervised local and global based on multiple manifold
The process monitoring model of holding.
Step 2.1:Raw data set X is mapped to by a core Hilbert sky for higher-dimension by nonlinear mapping function Φ
Between in H, obtain High dimensional space data collection Φ (X).
Step 2.2:Using each sample point in semi-supervised kernel locality preserving projections Algorithm for Solving High dimensional space data collection Φ (X)
Φ(xi) k neighbour domain point, obtain sample point Φ (xi) nearly Neighbourhood set Nk(Φ(xi)), i ∈ 1,2 ..., n.
Step 2.2.1:Using each sample in semi-supervised kernel locality preserving projections Algorithm for Solving High dimensional space data collection Φ (X)
Point Φ (xi) and neighborhood sample point Φ (xjThe distance between).
In present embodiment, each sample point Φ (xi) and neighborhood sample point Φ (xj) the distance between d (Φ (xi), Φ
(xj)) shown in formula such as formula (1):
d(Φ(xi), Φ (xj))=| | Φ (xi)-Φ(xj)|| (1)
Step 2.2.2:Each sample point Φ (x in High dimensional space data collection Φ (X)i) and neighborhood sample point Φ (xj) between
Distance in the minimum k neighborhood point of selected distance, constitute sample point Φ (xi) nearly Neighbourhood set Nk(Φ(xi))。
Step 2.3:Using the weight matrix P of Partial Reconstruction method reconstruct higher dimensional space High dimensional space data collection Φ (X).
Step 2.3.1:The c class sample data streams for making flag data in raw data set X are
Wherein, c is the classification of exemplar, 1≤c≤M, ncIt is the number of c class sample data streams,
Step 2.3.2:According to sample point Φ (xi) and its nearly Neighbourhood set Nk(Φ(xi)) power of Partial Reconstruction flag data
The weight matrix of value matrix and Unlabeled data, obtains the weight matrix at data point and Neighborhood Number strong point in Different categories of samples data flow
Pij。
In present embodiment, Different categories of samples data flow xiWith the weight matrix P between ambient data pointijAs shown in formula (2):
Wherein, UirIt is reconstruct Unlabeled data xiWeight matrix,σ is control
The pace of change of weight matrix, e is the truth of a matter of natural logrithm,It is Unlabeled data xiK1Individual unmarked neighbour domain,It is reconstruct flag dataWeight matrix, It is flag data
K2Individual unmarked neighbour domain, j=r, o, xrIt is and xiK1R-th point in neighbour domain, r=1,2 ... k1,Be withTogether
The k of class2O-th point in neighbour domain, o=1,2 ..., k2。
Step 2.3.3:The weights at data point and Neighborhood Number strong point in Different categories of samples data flow are solved using least square method
Matrix Pij, obtain the weight matrix U of flag datairWith the weight matrix of Unlabeled data
In present embodiment, the power at data point and Neighborhood Number strong point in Different categories of samples data flow is solved using least square method
Value matrix PijFormula such as formula (3) and formula (4) shown in:
Step 2.4:The mapping square on Different categories of samples data manifold is projected to according to kernel method solution High dimensional space data collection
Battle array WΦ={ W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A.
Step 2.4.1:Set up the semi-supervised kernel localized target function on Different categories of samples data manifold in higher dimensional space.
In present embodiment, the semi-supervised kernel localized target function J in higher dimensional space on Different categories of samples data manifoldL(WcΦ)
As shown in formula (5):
Wherein,It is the flag data semi-supervised kernel localized target function such as formula in c class sample data manifolds
(6) shown in,For Unlabeled data semi-supervised kernel localized target function such as formula (7) Suo Shi:
Wherein, matrix D is diagonal matrix, and its diagonal entry isWherein, PiiTake uirAnd Sio, diagonal matrixLrIt is figure Laplacian Matrix, Laplacian Matrix
Step 2.4.2:Set up the semi-supervised global objective function on Different categories of samples data manifold in higher dimensional space.
In present embodiment, the semi-supervised global objective function J in higher dimensional space on Different categories of samples data manifoldG(WcΦ) such as
Shown in formula (8):
Wherein,Shown in the semi-supervised global objective function such as formula (9) of c class sample data streams,
Shown in the semi-supervised global objective function such as formula (10) of Unlabeled data:
Step 2.4.3:With reference to semi-supervised kernel localized target function and semi-supervised global objective function, solved according to kernel method
Project to the mapping matrix W on Different categories of samples data manifoldΦ={ W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A, WcΦ
It is the mapping matrix in c class sample data manifolds.
In present embodiment, the mapping matrix W in c class sample data manifoldscΦAs shown in formula (11) and (12):
WcΦ=arg max (α JL(WcΦ)+βJG(WcΦ)) (11)
Wherein, α is Partial controll parameter, and β is global control parameter, alpha+beta=1.
When α=0, as multiple manifold semi-supervised kernel locality preserving projections algorithm (multiple manifold SSKLPP);
When β=0, as multiple manifold semi-supervised kernel pivot analysis algorithm (multiple manifold SSKPCA).
By method of Lagrange multipliers to the mapping matrix W on Different categories of samples data manifoldΦ={ W1Φ, W2Φ...,
WcΦ..., WMΦ}。
Ask shown in local derviation such as formula (13), coefficient matrices A=[a can be obtained1, a2..., ad]。
Wherein, d is the dimension after Data Dimensionality Reduction.
Step 2.5:Mapping matrix W according to High dimensional space data collection on Different categories of samples data manifoldΦ={ W1Φ,
W2Φ..., WcΦ..., WMΦAnd coefficient matrices A, set up the low-dimensional embedded coordinate y={ y of High dimensional space data1, y2...,
yc..., yM, that is, obtain the process monitoring model of and overall situation holding semi-supervised local based on multiple manifold.
In present embodiment, the mapping matrix W according to High dimensional space data collection on Different categories of samples data manifoldΦ=
{W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A, set up the low-dimensional embedded coordinate y={ y of High dimensional space data1,
y2..., yc..., yMFormula such as formula (14) shown in:
Step 3:The test data x of Real-time Collection smelting process of electro-fused magnesia furnace0, using based on multiple manifold it is semi-supervised local and
The process monitoring model that the overall situation keeps is to test data x0Carry out fault type diagnosis.
Step 3.1:The test data x of Real-time Collection smelting process of electro-fused magnesia furnace0。
Step 3.2:By nonlinear mapping function Φ by test data x0It is mapped to a core Hilbert space for higher-dimension
In H, test data High dimensional space data Φ (x are obtained0)。
Step 3.3:Construction test data High dimensional space data Φ (x0) on Different categories of samples data manifold low-dimensional insertion
Coordinate yc 0。
In present embodiment, test data High dimensional space data Φ (x0) on Different categories of samples data manifold low-dimensional insertion
Coordinate yc 0As shown in formula (15):
Step 3.4:Solve test data High dimensional space data Φ (x0) in embedded space in each sample data manifold
On reconstructed error value errorc(Φ(x0))。
In present embodiment, test data High dimensional space data Φ (x0) in embedded space in each sample data manifold
On reconstructed error value errorc(Φ(x0)) as shown in formula (16):
errorc(Φ(x0))=| | yc 0-yc||2 (16)
Step 3.5:By test data High dimensional space data Φ (x0) in embedded space in each sample data manifold
Reconstructed error value errorc(Φ(x0)) sample type of data manifold c belonging to minimum value as the test data failure
Type.
In present embodiment, determine shown in the formula such as formula (17) of test data fault type:
M(Φ(x0))=arg min errorc(Φ(x0)) (17)
In present embodiment, about 960 groups of test sample is taken per class, semi-supervised kernel pca method is respectively adopted
(SSKPCA) multiple manifold that, semi-supervised kernel locality preserving projections method (SSKLPP) and invention are proposed is semi-supervised local and global
The electric melting magnesium furnace fault monitoring method (SSK (LPP-PCA)) of holding extracts the feature of Various types of data, and is mapped in lower dimensional space,
The essential connection of data between analysis homogeneous data and class.
Wherein, semi-supervised kernel pca method (SSKPCA) considers from the whole geometry structure of data, for extracting number
According to the principal component vector that variance is maximum and time big, the of data then on data projection to the first two principal component vector direction, will be obtained
One and second principal component vector perspective view, shown in such as Fig. 4 (a).
Semi-supervised kernel locality preserving projections method (SSKLPP) are used for keeping the local geometric information between data, initial data
Information keeps constant by reconstructing weight matrix between neighborhood sample in space, is mapped to corresponding data point in low-dimensional embedded space
Between partial structurtes information it is still constant, extract sample the first and second features, obtain the projection of the first two feature of data
Figure, shown in such as Fig. 4 (b).
Multiple manifold is semi-supervised local and the global electric melting magnesium furnace fault monitoring method (SSK (LPP-PCA)) for keeping not only from
The global structure of data sets out, and allows also for the local characteristicses structure between data, extracts the first two feature of data, obtains
First, second projecting direction figure, shown in such as Fig. 4 (c).
In Fig. 4, lower triangle, circle, upper triangle represent normal data, the data of failure 1, the data of failure 2 respectively.
It is extremely due to obtaining the data with expertise knowledge information in actual industrial process in present embodiment
Difficult, in order to verify influence of the different knowledge information marker samples to process monitoring algorithm accuracy.Selection knowledge information
Marker samples are respectively 15%, 20%, the 25% of total number of samples, and the overall number of training sample set is identical and often class selection
Sample number is identical, and the influence to algorithm dimensionality reduction accuracy is analyzed by contrasting the marker samples number of different proportion.In Fig. 5 (a)
Three curves illustrate influence of the marker samples information of different proportion to process monitoring algorithm dimensionality reduction accuracy, by Fig. 5 (b)
It can be seen that when per class, mark sample number is minimum, accuracy is worst.Illustrate that marker samples quantity is more, the accuracy of algorithm is got over
Good, when number of labels reaches certain value, the accuracy of algorithm almost no longer changes.Simultaneously when marker samples number is fixed, three
The precision value of bar curve drops to the increase of lower dimensional space dimension with initial data feature, and selected characteristic number increases,
The data message of loss is fewer, and the accuracy of system is higher, and the performance of dimensionality reduction is better.From the foregoing, marker samples number is more
Monitoring effect is better, but 2.5% or 3.5% marker samples for the emulation experiment enough, almost can be intactly
Represent the whole information content of initial data.
In present embodiment, troubleshooting issue is actually many classification problems, and faults frequent occurs in industrial process,
Monitoring and diagnosis to failure are extremely necessary.We are modeled and real using the training data of above-mentioned Tennessee Eastman Process
When on-line monitoring sample residing for running status.Test data set then acquires two groups of each 450 sampled datas, the table of test set 1
Show that preceding 200 sample points, in the state of failure 1, about recover normal in the 201st point or so failure 1.Before test set 2 is represented
200 points are in the state of failure 2, and about recovering normal operating condition, i.e. failure 2 in the 201st point or so system eliminates.
In Fig. 6, wherein abscissa represents the test sample point at current time, and the test data is projected into each class respectively
Accordingly low-dimensional coordinate is obtained on other proper subspace, chooses nearest with it around the coordinate points respectively by Euclidean distance
K neighborhood point, the specific category belonging to the point is determined according to neighbour's reconstructed error value minimum, and ordinate represents reconstructed error value
Size, lowermost row represents the shape corresponding to current time test data to that should put distance and minimum with k Neighbor Points of surrounding
The current state of step response, i.e. batch processing operation.In figure 6, test set 1 is one group of 450 sampled data, including production process
In normal and data of failure 1, Fig. 6 (a) (b) (c) respectively illustrates SSKPCA algorithms, SSKLPP algorithms and inventive algorithm
Influence to malfunction monitoring.
Claims (4)
1. it is a kind of to be based on the semi-supervised part-overall situation of multiple manifold and electric melting magnesium furnace fault monitoring method, it is characterised in that including following
Step:
Step 1:Obtain the data and the number of the unmarked sample class of u groups of the l group echo sample class of smelting process of electro-fused magnesia furnace
According to composition raw data set X=(XL, XU)∈Rm×n, wherein, XL=[x1, x2..., xl], XU=[xl+1, xl+2..., xl+u],
L+u=n,It is flag data,It is Unlabeled data, m is data dimension;
Step 2:Raw data set according to smelting process of electro-fused magnesia furnace is set up based on the semi-supervised part of multiple manifold and global holding
Process monitoring model;
Step 2.1:Raw data set X is mapped to by a core Hilbert space H for higher-dimension by nonlinear mapping function Φ
In, obtain High dimensional space data collection Φ (X);
Step 2.2:Using each sample point Φ in semi-supervised kernel locality preserving projections Algorithm for Solving High dimensional space data collection Φ (X)
(xi) k neighbour domain point, obtain sample point Φ (xi) nearly Neighbourhood set Nk(Φ(xi)), i ∈ 1,2 ..., n;
Step 2.3:Using the weight matrix P of Partial Reconstruction method reconstruct higher dimensional space High dimensional space data collection Φ (X);
Step 2.4:The mapping matrix W on Different categories of samples data manifold is projected to according to kernel method solution High dimensional space data collectionΦ
={ W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A;
Step 2.5:Mapping matrix W according to High dimensional space data collection on Different categories of samples data manifoldΦ={ W1Φ, W2Φ...,
WcΦ..., WMΦAnd coefficient matrices A, set up the low-dimensional embedded coordinate y={ y of High dimensional space data1, y2..., yc..., yM,
Obtain the process monitoring model of and overall situation holding semi-supervised local based on multiple manifold;
Step 3:The test data x of Real-time Collection smelting process of electro-fused magnesia furnace0, using semi-supervised local and global based on multiple manifold
The process monitoring model of holding is to test data x0Carry out fault type diagnosis;
Step 3.1:The test data x of Real-time Collection smelting process of electro-fused magnesia furnace0;
Step 3.2:By nonlinear mapping function Φ by test data x0It is mapped in a core Hilbert space H for higher-dimension,
Obtain test data High dimensional space data Φ (x0);
Step 3.3:Construction test data High dimensional space data Φ (x0) low-dimensional embedded coordinate on Different categories of samples data manifold
yc 0;
Step 3.4:Solve test data High dimensional space data Φ (x0) weight in embedded space in each sample data manifold
Structure error amount errorc(Φ(x0));
Step 3.5:By test data High dimensional space data Φ (x0) reconstruct in embedded space in each sample data manifold
Error amount errorc(Φ(x0)) sample type of data manifold c belonging to minimum value as the test data fault type.
2. the electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold according to claim 1, it is special
Levy and be, the step 2.2 is comprised the following steps:
Step 2.2.1:Using each sample point Φ in semi-supervised kernel locality preserving projections Algorithm for Solving High dimensional space data collection Φ (X)
(xi) and neighborhood sample point Φ (xjThe distance between);
Step 2.2.2:Each sample point Φ (x in High dimensional space data collection Φ (X)i) and neighborhood sample point Φ (xj) between away from
From the k neighborhood point that middle selected distance is minimum, sample point Φ (x are constitutedi) nearly Neighbourhood set Nk(Φ(xi))。
3. the electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold according to claim 1, it is special
Levy and be, the step 2.3 is comprised the following steps:
Step 2.3.1:The c class sample data streams for making flag data in raw data set X areWherein,
C is the classification of exemplar, 1≤c≤M, ncIt is the number of c class sample data streams,
Step 2.3.2:According to sample point Φ (xi) and its nearly Neighbourhood set Nk(Φ(xi)) the weights square of Partial Reconstruction flag data
The weight matrix of battle array and Unlabeled data, obtains the weight matrix P at data point and Neighborhood Number strong point in Different categories of samples data flowij;
Step 2.3.3:The weight matrix at data point and Neighborhood Number strong point in Different categories of samples data flow is solved using least square method
Pij, obtain the weight matrix U of flag datairWith the weight matrix of Unlabeled data
4. the electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold according to claim 1, it is special
Levy and be, the step 2.4 is comprised the following steps:
Step 2.4.1:Set up the semi-supervised kernel localized target function on Different categories of samples data manifold in higher dimensional space;
Step 2.4.2:Set up the semi-supervised global objective function on Different categories of samples data manifold in higher dimensional space;
Step 2.4.3:With reference to semi-supervised kernel localized target function and semi-supervised global objective function, solved according to kernel method and projected
Mapping matrix W on to Different categories of samples data manifoldΦ={ W1Φ, W2Φ..., WcΦ..., WMΦAnd coefficient matrices A, WcΦIt is c
Mapping matrix in class sample data manifold.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710181754.5A CN106839769B (en) | 2017-03-24 | 2017-03-24 | Electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710181754.5A CN106839769B (en) | 2017-03-24 | 2017-03-24 | Electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106839769A true CN106839769A (en) | 2017-06-13 |
CN106839769B CN106839769B (en) | 2018-11-27 |
Family
ID=59130820
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710181754.5A Active CN106839769B (en) | 2017-03-24 | 2017-03-24 | Electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106839769B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107817745A (en) * | 2017-10-31 | 2018-03-20 | 东北大学 | Industrial process fault monitoring method based on clump stream forming core linear discriminant analysis |
CN107831743A (en) * | 2017-10-31 | 2018-03-23 | 东北大学 | Industrial process method for diagnosing faults based on credible soft label linear discriminant analysis |
CN107861481A (en) * | 2017-10-31 | 2018-03-30 | 东北大学 | The method for diagnosing faults excavated based on the semi-supervised new class of industrial big data figure |
CN108182445A (en) * | 2017-12-13 | 2018-06-19 | 东北大学 | Procedure fault recognition methods based on big data intelligence core independent component analysis |
CN109961027A (en) * | 2019-03-13 | 2019-07-02 | 东北大学 | Subspace multi-angle of view electric melting magnesium furnace fault monitoring method is shared based on similarity measurement |
CN111061257A (en) * | 2019-12-30 | 2020-04-24 | 杭州电子科技大学 | Industrial process monitoring method based on dynamic global LPP |
CN111913460A (en) * | 2019-05-20 | 2020-11-10 | 宁波大学 | Fault monitoring method based on sequence correlation local preserving projection algorithm |
WO2023097808A1 (en) * | 2021-12-03 | 2023-06-08 | 株洲瑞德尔冶金设备制造有限公司 | Fault monitoring method and apparatus for sintering device |
-
2017
- 2017-03-24 CN CN201710181754.5A patent/CN106839769B/en active Active
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107831743A (en) * | 2017-10-31 | 2018-03-23 | 东北大学 | Industrial process method for diagnosing faults based on credible soft label linear discriminant analysis |
CN107861481A (en) * | 2017-10-31 | 2018-03-30 | 东北大学 | The method for diagnosing faults excavated based on the semi-supervised new class of industrial big data figure |
CN107831743B (en) * | 2017-10-31 | 2019-12-10 | 东北大学 | Industrial process fault diagnosis method based on credible soft label linear discriminant analysis |
CN107817745A (en) * | 2017-10-31 | 2018-03-20 | 东北大学 | Industrial process fault monitoring method based on clump stream forming core linear discriminant analysis |
CN108182445A (en) * | 2017-12-13 | 2018-06-19 | 东北大学 | Procedure fault recognition methods based on big data intelligence core independent component analysis |
CN108182445B (en) * | 2017-12-13 | 2020-05-19 | 东北大学 | Process fault identification method based on big data intelligent core independent element analysis |
CN109961027A (en) * | 2019-03-13 | 2019-07-02 | 东北大学 | Subspace multi-angle of view electric melting magnesium furnace fault monitoring method is shared based on similarity measurement |
CN109961027B (en) * | 2019-03-13 | 2023-01-13 | 东北大学 | Similarity measurement based multi-view electric smelting magnesium furnace fault monitoring method for shared subspace |
CN111913460B (en) * | 2019-05-20 | 2022-03-18 | 宁波大学 | Fault monitoring method based on sequence correlation local preserving projection algorithm |
CN111913460A (en) * | 2019-05-20 | 2020-11-10 | 宁波大学 | Fault monitoring method based on sequence correlation local preserving projection algorithm |
CN111061257A (en) * | 2019-12-30 | 2020-04-24 | 杭州电子科技大学 | Industrial process monitoring method based on dynamic global LPP |
CN111061257B (en) * | 2019-12-30 | 2021-02-19 | 杭州电子科技大学 | Industrial process monitoring method based on dynamic global LPP |
WO2023097808A1 (en) * | 2021-12-03 | 2023-06-08 | 株洲瑞德尔冶金设备制造有限公司 | Fault monitoring method and apparatus for sintering device |
Also Published As
Publication number | Publication date |
---|---|
CN106839769B (en) | 2018-11-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106839769B (en) | Electric melting magnesium furnace fault monitoring method based on the semi-supervised part-overall situation of multiple manifold | |
CN107065828B (en) | The semi-supervised kernel discriminant analysis method for diagnosing faults of knowledge based data | |
CN107817745B (en) | Industrial process fault monitoring method based on clustering manifold kernel linear discriminant analysis | |
CN108181891A (en) | Industrial big data method for diagnosing faults based on intelligent core pivot element analysis | |
CN108536128A (en) | A kind of machine learning fault diagnosis system of parameter optimization | |
CN107748901B (en) | Industrial process fault diagnosis method based on similarity local spline regression | |
CN108182445A (en) | Procedure fault recognition methods based on big data intelligence core independent component analysis | |
CN107861481A (en) | The method for diagnosing faults excavated based on the semi-supervised new class of industrial big data figure | |
CN113234874B (en) | Blast furnace fault monitoring and self-healing control system based on dynamic Bayesian network | |
CN106599450B (en) | Priori knowledge-based fault monitoring method for flexible manifold embedded electric smelting magnesium furnace | |
CN104914854B (en) | Industrial process fault diagnosis method based on KPCA | |
CN105843212B (en) | A kind of blast furnace fault diagnosis system and method | |
CN107831743B (en) | Industrial process fault diagnosis method based on credible soft label linear discriminant analysis | |
CN109885032A (en) | The isomeric data collaborative modeling industry fault detection method kept based on neighborhood projection | |
CN106709149A (en) | Neural network-based method and system for predicting shapes of three-dimensional hearths of aluminum cells in real time | |
CN109978031B (en) | Multi-view manifold analysis fault diagnosis method based on image feature regression | |
CN109961027B (en) | Similarity measurement based multi-view electric smelting magnesium furnace fault monitoring method for shared subspace | |
CN110009620A (en) | The method and system of power grid stockbridge damper intelligent recognition are carried out based on transfer learning algorithm | |
Wei | Power grid facility thermal fault diagnosis via object detection with synthetic infrared imagery | |
CN115938493A (en) | Online optimization method of high-temperature solid oxide electrolytic hydrogen production system | |
CN109885028B (en) | Method for detecting faults in electric smelting magnesium oxide smelting process based on related entropy induction measurement | |
CN109885027A (en) | Industrial process method for diagnosing faults based on the sparse orthogonal discriminant analysis of bidirectional two-dimensional | |
Bao et al. | Automatic identification and defect diagnosis of transmission line insulators based on YOLOv3 network | |
CN110417350B (en) | Fault diagnosis method for photovoltaic array | |
CN109060400B (en) | Method for sampling and detecting mechanical property authenticity of torn steel plate of water turbine seat ring |
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 |