CN108334898A - A kind of multi-modal industrial process modal identification and Fault Classification - Google Patents
A kind of multi-modal industrial process modal identification and Fault Classification Download PDFInfo
- Publication number
- CN108334898A CN108334898A CN201810062893.0A CN201810062893A CN108334898A CN 108334898 A CN108334898 A CN 108334898A CN 201810062893 A CN201810062893 A CN 201810062893A CN 108334898 A CN108334898 A CN 108334898A
- Authority
- CN
- China
- Prior art keywords
- sample point
- point
- sample
- detected
- historical
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of multi-modal industrial process modal identification and Fault Classifications, including：Acquire industrial process history normal data and fault data under different modalities；Classified offline to different mode；Acquire industrial process data to be detected；Two characteristic quantities of sample to be detected are calculated according to standard Euclidean distance：The local density of point and the minimum range to high local density's point；According to the distribution situation of two characteristic quantities, the mode belonging to sample to be detected is judged online；The present invention is by acquiring the malfunction history data under different modalities；Classified offline to different failures；Acquire fault sample to be detected；The fault type of sample to be detected is judged online；It can pick out data mode and fault type under conditions of not needing priori, and cluster centre and clusters number need not be specified at cluster, greatly reduce calculation amount.
Description
Technical field
The invention belongs to multi-modal industrial process modal identifications and failure modes field, more particularly, to a kind of multimode
State industrial process modal identification and Fault Classification.
Background technology
For a large scale industry system, due to the change of production strategy and production environment, often show multi-modal, more
The characteristics of failure.Need to establish different models for different mode, thus before modeling to multi-modal industrial process into
Row modal identification and failure modes have great importance.
Current most popular modal identification and sorting technique are the methods based on data-driven.Its main method has two
Kind, one is the clustering algorithm based on pure mathematics, one is based on PCA (Principal Component Analysis) or
Similarity between PLS (Partial Least Square) model carries out modal identification.
Traditional clustering algorithm is mainly k-means clustering algorithms and the derivative algorithm of k-means.Based on k-means
Clustering method be a kind of clustering algorithm having supervision, need default clusters number and initialize cluster centre, and initialize
Cluster centre is the mean value computation according to data, so k-means clusters are more sensitive to abnormal data, cluster result is often
It is inaccurate.Based on the method that the similarity between PCA or PLS models carries out modal identification, need first to build each sample
Formwork erection type, the similarity between computation model, then clustered using clustering algorithm, then the sample of each cluster is combined again
Modeling, step is complicated, and still needs the intervention of clustering algorithm.
Both methods needs that clusters number is manually set, calculating is complicated, and cluster result is not all there is certain defect
It is enough accurate.In actual industrial process, we tend not to be known in advance several mode in alternate run, and when one
When new sample occurs, it would be desirable to the operating status belonging to it is judged with the shorter time, so need to design a nothing
Clustering algorithm supervise, that time complexity is smaller.
Invention content
For the disadvantages described above or Improvement requirement of the prior art, the present invention provides a kind of multi-modal industrial process mode to distinguish
Knowledge and Fault Classification, thus solve currently based on the clustering algorithm of pure mathematics and based on the phase between PCA or PLS models
It is existing like degree progress modal identification to need that clusters number is manually set, calculate complicated, the inaccurate technology of cluster result
Problem.
To achieve the above object, the present invention provides a kind of multi-modal industrial process modal identification and Fault Classification,
Including：
(1) sample data to be detected is collected as sample point to be detected, is calculated the sample point to be detected and is trained with history
Standard Euclidean distance between sample set；
(2) it is calculated based on the standard Euclidean distance between the sample point to be detected and history training sample set described to be checked
Survey the local density of sample point and the minimum range to high local density's point；
(3) local density of the sample point to be detected is put into the part of the descending arrangement of the history training sample set
In density set, re-starts descending and arrange to obtain new local density's set；
(4) gathered according to the new local density, if the sample point to be detected is to the office than the sample point to be detected
The minimum range of the higher target histories sample point of portion's density is equal to the sample point to be detected to the target histories sample point
Standard Euclidean distance, then the sample point to be detected and the target histories sample point belong to same class cluster, then by institute
The class cluster of category judges the mode belonging to the sample point to be detected and fault type.
Preferably, before step (1), the method further includes：
(11) history training sample set is acquired, wherein the history training sample set is by the normal number from different modalities
It is formed according to fault data；
(12) it calculates the history training sample and concentrates the standard Euclidean distance between each historical sample point；
(13) the standard Euclidean distance and adjusting ginseng between each historical sample point are concentrated according to the history training sample
Number calculates and blocks distance, and obtains the part that the history training sample concentrates each historical sample point by the distance of blocking
Density and minimum range to high local density's point, and each local density of the history training sample set is arranged according to descending
Row；
(14) selection historical sample point local density and historical sample point to high local density's point minimum range all phases
To larger point as cluster centre；
(15) for the remaining historical sample point without the cluster centre, if the first remaining historical sample point is to than institute
The minimum range for stating first the higher second remaining historical sample point of remaining historical sample point local density is surplus equal to described first
Remaining historical sample point to the standard Euclidean distance of the described second remaining historical sample point, then the described first remaining historical sample point with
Described second remaining historical sample point belongs to same class cluster, to obtain the classification results of different modalities and different faults.
Preferably, step (12) includes：
To the history training sample set X ∈ RN×JIn every a line, byObtain history sample
Standard Euclidean distance between this i and historical sample point j, wherein N is the total sample number that the history training sample is concentrated, J
For variable number, skIt is standard deviation, xikIndicate the sample data of the i-th row kth row, xjkIndicate the sample data of jth row kth row.
Preferably, step (13) includes：
(13.1) the history training sample is concentrated to the standard Euclidean distance d between each historical sample pointij(i ＜ j) is pressed
It is arranged in sequence sda (d according to descending1,d2,...,dM), wherein
(13.2) by dcDistance is blocked in=sda [round (N × (N-1) × p)] calculating, wherein round indicates four houses five
Enter, p is adjustment parameter；
(13.3) determine that target blocks distance dc', so that neighbours' number of each historical sample point is all historical sample points
1%-2%；
(13.4) distance d is blocked according to the targetc', by βi=∑jχ(dij-dc') calculate each historical sample point i
Local density βi, and descending arrangement { β is carried out to the local density of each historical sample point1,β2,...,βN, wherein
(13.5) byHistorical sample point i is calculated to than going through
The minimum range of the higher point of history sample point i local densities.
Preferably, step (15) includes：
For the arbitrary remaining historical sample point k without cluster centre, if δk=djk, then historical sample point k belong to history
Class belonging to sample point j is indicated in the local density δ than historical sample point kkIn all historical sample points of bigger, history
Sample point j and historical sample point k distances are recently.
Preferably, step (4) includes：
Gather (β according to new local density1,β2,…,βi,βnew,βj..., βN), than βnewSample point (the β of bigger1,
β2,…,βi) in, find δnew=dmnew, then the sample point to be detected belong to the class cluster belonging to historical sample point m, wherein βnew
Indicate the local density of the sample point to be detected, δnewIndicate the sample point to be detected to high local density's point most narrow spacing
From dmnewIndicate the standard European distance between the sample point to be detected and historical sample point m.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect：
(1) off-line mode division can be effectively performed in the present invention and failure divides, and on-line checking goes out the mode of sample
And fault type；
(2) method in the present invention is a kind of unsupervised method, need not preset cluster number, need not be initial
Change cluster centre；
(3) method calculation amount of the invention is small, easy to implement.
Description of the drawings
Fig. 1 is the flow signal of a kind of multi-modal industrial process modal identification and Fault Classification provided by the invention
Figure；
Fig. 2 is the flow signal of the multi-modal industrial process modal identification of another kind provided by the invention and Fault Classification
Figure；
Fig. 3 is the schematic diagram of Tennessee Eastman processes；
Fig. 4 is the corresponding mode of 6 kinds of operation modes of Tennessee Eastman processes；
Fig. 5 is the verification sample sequence divided for off-line mode；
Fig. 6 (a) and Fig. 6 (b) is the result schematic diagram of the method for the present invention and k-means method modal identifications respectively；
Fig. 7 is the verification sample sequence 1 for offline failure modes；
Fig. 8 (a) and Fig. 8 (b) is that the result that the method for the present invention and k-means methods classify to the failure 1 of mode 1 is shown respectively
It is intended to；
Fig. 9 is the verification sample sequence 2 for offline failure modes；
Figure 10 (a) and Figure 10 (b) is the result that the method for the present invention and k-means methods classify to the failure 2 of mode 1 respectively
Schematic diagram；
Figure 11 is the verification sample sequence 3 for offline failure modes；
Figure 12 (a) and Figure 12 (b) is the result that the method for the present invention and k-means methods classify to the failure 1 of mode 3 respectively
Schematic diagram；
Figure 13 is the verification sample sequence for online modal identification；
Figure 14 (a) and Figure 14 (b) is using the method for the present invention respectively to the online data mode from mode 1 and mode 3
Identification result schematic diagram；
Figure 15 is the verification sample sequence for online failure modes；
Figure 16 (a) and Figure 16 (b) is failure 1 of the data distributing method to mode 1 of the method for the present invention and k-means respectively
Result schematic diagram；
Figure 17 (a) and Figure 17 (b) is failure 2 of the data distributing method to mode 1 of the method for the present invention and k-means respectively
Result schematic diagram；
Figure 18 (a) and Figure 18 (b) is failure 1 of the data distributing method to mode 3 of the method for the present invention and k-means respectively
Result schematic diagram.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
It does not constitute a conflict with each other and can be combined with each other.
The present invention provides a kind of multi-modal industrial process modal identification and Fault Classification, this method need not be prior
Default clusters number calculates the local density of each data point and arrives high office by the standard Euclidean distance between data point
The minimum range of portion's density points, it is cluster centre to choose the big and big point of local density, and remaining point is also according to the two amounts
Distributed from top to bottom, rather than it is traditional distributed according to the distance to cluster centre, cluster result is more accurate, and
Calculation amount is greatly reduced, there is good applicability for multi-modal industrial process.
It is the flow of a kind of multi-modal industrial process modal identification and Fault Classification provided by the invention as shown in Figure 1
Schematic diagram includes the following steps in method shown in Fig. 1：
(1) sample data to be detected is collected as sample point to be detected, calculates sample point to be detected and history training sample
Standard Euclidean distance between collection；
(2) sample point to be detected is calculated based on the standard Euclidean distance between sample point to be detected and history training sample set
Local density and minimum range to high local density's point；
Wherein, high local density's point is that numerical value compares higher local density's point in all local densities, can basis
Actual needs is determined.
(3) local density that the local density of sample point to be detected is put into the descending arrangement of history training sample set gathers
In, it re-starts descending and arranges to obtain new local density's set；
(4) gathered according to new local density, if sample point to be detected is higher to the local density than sample point to be detected
The minimum range of target histories sample point be equal to sample point to be detected to target histories sample point standard Euclidean distance, then it is to be checked
It surveys sample point and belongs to same class cluster with target histories sample point, then judged belonging to sample point to be detected by affiliated class cluster
Mode and fault type.
It is the stream of another kind provided by the invention multi-modal industrial process modal identification and Fault Classification as shown in Figure 2
Journey schematic diagram, include in fig. 2 the present invention the off-line identification stage and the on-line identification stage, specifically, step (1) it
Before, it needs to carry out off-line identification, specifically includes following steps：
(11) acquire history training sample set, wherein history training sample set by from different modalities normal data and
Fault data forms；
(12) it calculates history training sample and concentrates the standard Euclidean distance between each historical sample point；
(13) the standard Euclidean distance and adjustment parameter meter between each historical sample point are concentrated according to history training sample
Distance is blocked in calculation, and is obtained history training sample by blocking distance and concentrated the local density of each historical sample point and to high office
The minimum range of portion's density points, and each local density of history training sample set is arranged according to descending；
(14) selection historical sample point local density and historical sample point to high local density's point minimum range all phases
To larger point as cluster centre；
Wherein, the big point for indicating a large amount of low local densities of local density is around the point, most to high local density's point
The big cluster centre for indicating each class cluster of small distance is at a distance of distant, and then local density is big and minimum to high local density's point
It is more likely reasonable cluster centre apart from big point.
And local density is small and means that it is relatively more isolated to the big point of the minimum range of high local density's point, and far from poly-
Therefore class center, can choose several local densities and arrive height according to actual needs so such point is considered as outlier
The all relatively large point of the minimum range of local density's point is used as cluster centre.
(15) for the remaining historical sample point without cluster centre, if the first remaining historical sample point is remained to than first
The minimum range of the higher second remaining historical sample point of remaining historical sample point local density is equal to the first remaining historical sample point
To the standard Euclidean distance of the second remaining historical sample point, then the first remaining historical sample point and the second residue historical sample point category
In same class cluster, to obtain the classification results of different modalities and different faults.
It according to above-mentioned cluster principle, can classify first to training data, obtain different modalities and different faults
Then classification results distribute sample to be detected online, find mode and fault type belonging to sample to be detected.
In an optional embodiment, step (12) includes：
To history training sample set X ∈ RN×JIn every a line, byObtain historical sample point i
With the standard Euclidean distance between historical sample point j, wherein N is the total sample number that history training sample is concentrated, and J is variable
Number, skIt is standard deviation, xikIndicate the sample data of the i-th row kth row, xjkIndicate the sample data of jth row kth row.
In an optional embodiment, step (13) includes：
(13.1) history training sample is concentrated to the standard Euclidean distance d between each historical sample pointij(i ＜ j) is according to drop
Sequence is arranged in sequence sda (d1,d2,...,dM), whereinIndicate that history training sample concentrates each history sample
The number of standard Euclidean distance between this point；
(13.2) by dcDistance is blocked in=sda [round (N × (N-1) × p)] calculating, wherein round indicates four houses five
Enter, p is adjustment parameter；
(13.3) determine that target blocks distance dc', so that neighbours' number of each historical sample point is all historical sample points
1%-2%, namely selection adjustment parameter p value between 1%-2%；
(13.4) distance d is blocked according to the targetc', by βi=∑jχ(dij-dc') calculate each historical sample point i
Local density βi, and descending arrangement { β is carried out to the local density of each historical sample point1,β2,...,βN, wherein
Wherein, local density βiIt indicates to be less than d at a distance from historical sample point ic' point number.
(13.5) byHistorical sample point i is calculated to than going through
The minimum range of the higher point of history sample point i local densities.
In an optional embodiment, step (15) includes：
For the arbitrary remaining historical sample point k without cluster centre, if δk=djk, then historical sample point k belong to history
Class belonging to sample point j is indicated in the local density δ than historical sample point kkIn all historical sample points of bigger, history
Sample point j and historical sample point k distances are recently.
Since the maximum point of density is cluster centre, so after finding cluster centre, it is big just to distribute local density time
Point, and so on, until all points distribute.
In an optional embodiment, step (4) includes：
Gather (β according to new local density1,β2,…,βi,βnew,βj..., βN), than βnewSample point (the β of bigger1,
β2,…,βi) in, find δnew=dmnew, then sample point to be detected belong to the class cluster belonging to historical sample point m, wherein βnewIt indicates
The local density of sample point to be detected, δnewIndicate sample point to be detected to the minimum range of high local density's point, dmnewExpression waits for
Detect the standard European distance between sample point and historical sample point m.
The method of the present invention is illustrated below in conjunction with specific embodiment.
Embodiment uses multi-modal industrial process modal identification provided by the invention and Fault Classification, passes through
Tennessee Eastman (TE) process is verified.
TE processes are an emulation platforms based on real industrial process, and neck is studied in the fault detect based on data-driven
Domain is widely used in the performance evaluation of various monitoring methods, including 5 parts：Reactor, condenser, recycle compressor,
Gas-liquid separator, stripper, schematic diagram are as shown in Figure 3.The process includes a kind of normal condition data set and 20 kinds of failure shapes
State data set, including 41 measurands and 12 control variables.According to the difference of product G/H mass ratioes, there are six TE processes
Operation mode, each mode parameter as shown in figure 4, make system be operated in corresponding pattern according to different needs in industrial process
Under.
This verification carries out in the case of mode 1 and mode 3, and Simulink programs can be from website http://
Depts.washington.edu/control/LARRY/TE/download.html is obtained.
Off-line verification data sequence for modal identification is as shown in figure 5, all normal samples comprising mode 1 and mode 3
Sheet and fault sample.
Fig. 6 (a) and Fig. 6 (b) compare the method for the present invention and k-means methods, as can be seen from the figure the method for the present invention
Different mode can more effectively be picked out.
Off-line verification data sequence 1 for failure modes is as shown in fig. 7, the normal sample comprising mode 1 and mode 3 is each
100,1 sample of failure of 500 and mode 1.
Fig. 8 (a) and Fig. 8 (b) compare the method for the present invention and k-means methods, as can be seen from the figure the method for the present invention
The failure 1 of mode 1 can more effectively be picked out.
Off-line verification data sequence 2 for failure modes is as shown in figure 9, the normal sample comprising mode 1 and mode 3 is each
100,2 sample of failure of 500 and mode 1.
Figure 10 (a) and Figure 10 (b) compare the method for the present invention and k-means methods, as can be seen from the figure present invention side
Method can more effectively pick out the failure 2 of mode 1.
Off-line verification data sequence 3 for failure modes is as shown in figure 11, includes the normal sample of mode 1 and mode 3
100,1 sample of failure of each 500 and mode 3.
Figure 12 (a) and Figure 12 (b) compare the method for the present invention and k-means methods, as can be seen from the figure present invention side
Method can effectively pick out the failure 1 of mode 3.
In order to which on-line checking goes out the mode belonging to new samples, 100 data from mode 1 or mode 3 are acquired as survey
Sample sheet, verify data sequence are as shown in figure 13.
Figure 14 (a) and Figure 14 (b) is the method for the present invention respectively to carrying out online mould from the new data of mode 1 and mode 3
State identification result schematic diagram, as can be seen from the figure the method for the present invention effectively on-line identification can go out the mould belonging to new samples
State.
In order to which on-line checking goes out the fault type belonging to new samples, 500 normal samples from mode 1 and mode 3 are chosen
This, 2 each 100 samples of failure 1 and failure of mode 1, the failure 1 of mode 3 is 100 samples as training sample, sample sequence
Row are as shown in figure 15.
Figure 16 (a) and Figure 16 (b) compares the data distributing method of the present invention and the data distributing method of k-means to mould
1 identification result of failure of state 1, as can be seen from the figure method of the invention is more effective.
Figure 17 (a) and Figure 17 (b) compares the data distributing method of the present invention and the data distributing method of k-means to mould
2 identification result of failure of state 1, as can be seen from the figure method of the invention is more effective.
Figure 18 (a) and Figure 18 (b) compares the data distributing method of the present invention and the data distributing method of k-means to mould
1 identification result of failure of state 3, as can be seen from the figure method of the invention is more effective.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, all within the spirits and principles of the present invention made by all any modification, equivalent and improvement etc., should all include
Within protection scope of the present invention.
Claims (6)
1. a kind of multi-modal industrial process modal identification and Fault Classification, which is characterized in that including：
(1) sample data to be detected is collected as sample point to be detected, calculates the sample point to be detected and history training sample
Standard Euclidean distance between collection；
(2) test sample to be checked is calculated based on the standard Euclidean distance between the sample point to be detected and history training sample set
The local density of this point and minimum range to high local density's point；
(3) local density of the sample point to be detected is put into the local density of the descending arrangement of the history training sample set
In set, re-starts descending and arrange to obtain new local density's set；
(4) gathered according to the new local density, if the sample point to be detected is close to the part than the sample point to be detected
Spend mark of the minimum range equal to the sample point to be detected to the target histories sample point of higher target histories sample point
The test of notation, then the sample point to be detected and the target histories sample point belong to same class cluster, then by affiliated
Class cluster judges the mode belonging to the sample point to be detected and fault type.
2. according to the method described in claim 1, it is characterized in that, before step (1), the method further includes：
(11) acquire history training sample set, wherein the history training sample set by from different modalities normal data and
Fault data forms；
(12) it calculates the history training sample and concentrates the standard Euclidean distance between each historical sample point；
(13) the standard Euclidean distance and adjustment parameter meter between each historical sample point are concentrated according to the history training sample
Distance is blocked in calculation, and obtains the local density that the history training sample concentrates each historical sample point by the distance of blocking
With the minimum range to high local density's point, and each local density of the history training sample set is arranged according to descending；
(14) selection historical sample point local density and historical sample point to high local density's point minimum range all relatively
Big point is as cluster centre；
(15) for the remaining historical sample point without the cluster centre, if the first remaining historical sample point is to than described the
The minimum range of 1 the higher second remaining historical sample point of remaining historical sample point local density is gone through equal to first residue
History sample point to the standard Euclidean distance of the described second remaining historical sample point, then the described first remaining historical sample point with it is described
Second remaining historical sample point belongs to same class cluster, to obtain the classification results of different modalities and different faults.
3. according to the method described in claim 2, it is characterized in that, step (12) includes：
To the history training sample set X ∈ RN×JIn every a line, byObtain historical sample point i
With the standard Euclidean distance between historical sample point j, wherein N is the total sample number that the history training sample is concentrated, and J is to become
Measure number, skIt is standard deviation, xikIndicate the sample data of the i-th row kth row, xjkIndicate the sample data of jth row kth row.
4. according to the method described in claim 3, it is characterized in that, step (13) includes：
(13.1) the history training sample is concentrated to the standard Euclidean distance d between each historical sample pointij(i ＜ j) is according to drop
Sequence is arranged in sequence sda (d1,d2,...,dM), wherein
(13.2) by dcDistance is blocked in=sda [round (N × (N-1) × p)] calculating, wherein round expressions round up, and p is
Adjustment parameter；
(13.3) determine that target blocks distance dc', so that neighbours' number of each historical sample point is the 1%- of all historical sample points
2%；
(13.4) distance d is blocked according to the targetc', by βi=Σjχ(dij-dc') calculate the office of each historical sample point i
Portion density βi, and descending arrangement { β is carried out to the local density of each historical sample point1,β2,...,βN, wherein
(13.5) byHistorical sample point i is calculated to than historical sample
The minimum range of the higher point of point i local densities.
5. according to the method described in claim 4, it is characterized in that, step (15) includes：
For the arbitrary remaining historical sample point k without cluster centre, if δk=djk, then historical sample point k belong to historical sample
Class belonging to point j is indicated in the local density δ than historical sample point kkIn all historical sample points of bigger, historical sample
Point j and historical sample point k distances are recently.
6. method according to claim 4 or 5, which is characterized in that step (4) includes：
Gather (β according to new local density1,β2,…,βi,βnew,βj..., βN), than βnewSample point (the β of bigger1,β2,…,
βi) in, find δnew=dmnew, then the sample point to be detected belong to the class cluster belonging to historical sample point m, wherein βnewIndicate institute
State the local density of sample point to be detected, δnewIndicate the sample point to be detected to the minimum range of high local density's point, dmnew
Indicate the standard European distance between the sample point to be detected and historical sample point m.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810062893.0A CN108334898A (en) | 2018-01-23 | 2018-01-23 | A kind of multi-modal industrial process modal identification and Fault Classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810062893.0A CN108334898A (en) | 2018-01-23 | 2018-01-23 | A kind of multi-modal industrial process modal identification and Fault Classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108334898A true CN108334898A (en) | 2018-07-27 |
Family
ID=62925639
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810062893.0A Pending CN108334898A (en) | 2018-01-23 | 2018-01-23 | A kind of multi-modal industrial process modal identification and Fault Classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108334898A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111464022A (en) * | 2020-03-25 | 2020-07-28 | 哈尔滨工业大学 | Multi-feature-based parameter identification method for key devices of DC-DC converter |
| CN112329191A (en) * | 2019-08-05 | 2021-02-05 | 华中科技大学 | Modal identification method and transition mode fault detection method in multi-mode process |
| CN113392912A (en) * | 2021-06-18 | 2021-09-14 | 大唐环境产业集团股份有限公司 | Multi-mode operation fault diagnosis and early warning method, system and equipment for slurry circulating pump |
| CN114525372A (en) * | 2022-01-05 | 2022-05-24 | 浙江大学 | Blast furnace state monitoring method and device based on multi-mode fusion |
| CN115186772A (en) * | 2022-09-13 | 2022-10-14 | 云智慧（北京）科技有限公司 | Method, device and equipment for detecting partial discharge of power equipment |
| CN116124218A (en) * | 2023-02-13 | 2023-05-16 | 正泰电气股份有限公司 | Transformer fault diagnosis method and device, storage medium and electronic equipment |
| WO2023130234A1 (en) * | 2022-01-05 | 2023-07-13 | 浙江大学 | Blast furnace state monitoring method and apparatus based on multi-modality fusion |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8170306B2 (en) * | 2007-04-25 | 2012-05-01 | Siemens Aktiengesellschaft | Automatic partitioning and recognition of human body regions from an arbitrary scan coverage image |
| CN106777984A (en) * | 2016-12-19 | 2017-05-31 | 福州大学 | A kind of method that photovoltaic array Working state analysis and fault diagnosis are realized based on density clustering algorithm |
| CN107103336A (en) * | 2017-04-28 | 2017-08-29 | 温州职业技术学院 | A kind of mixed attributes data clustering method based on density peaks |
| CN107491783A (en) * | 2017-07-31 | 2017-12-19 | 广东电网有限责任公司惠州供电局 | Based on the transformer fault genre classification methods for improving density peaks clustering algorithm |
-
2018
- 2018-01-23 CN CN201810062893.0A patent/CN108334898A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8170306B2 (en) * | 2007-04-25 | 2012-05-01 | Siemens Aktiengesellschaft | Automatic partitioning and recognition of human body regions from an arbitrary scan coverage image |
| CN106777984A (en) * | 2016-12-19 | 2017-05-31 | 福州大学 | A kind of method that photovoltaic array Working state analysis and fault diagnosis are realized based on density clustering algorithm |
| CN107103336A (en) * | 2017-04-28 | 2017-08-29 | 温州职业技术学院 | A kind of mixed attributes data clustering method based on density peaks |
| CN107491783A (en) * | 2017-07-31 | 2017-12-19 | 广东电网有限责任公司惠州供电局 | Based on the transformer fault genre classification methods for improving density peaks clustering algorithm |
Non-Patent Citations (2)
| Title |
|---|
| ALEX RODRIGUEZ ET AL.: "Clustering by fast search and find of density peaks", 《SCIENCE》 * |
| HAOLAN YAN ET AL.: "Density peaks clustering based sub-phase partition and", 《2017 IEEE 6TH DATA DRIVEN CONTROL AND LEARNING SYSTEMS (DDCLS)》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112329191A (en) * | 2019-08-05 | 2021-02-05 | 华中科技大学 | Modal identification method and transition mode fault detection method in multi-mode process |
| CN111464022A (en) * | 2020-03-25 | 2020-07-28 | 哈尔滨工业大学 | Multi-feature-based parameter identification method for key devices of DC-DC converter |
| CN113392912A (en) * | 2021-06-18 | 2021-09-14 | 大唐环境产业集团股份有限公司 | Multi-mode operation fault diagnosis and early warning method, system and equipment for slurry circulating pump |
| CN114525372A (en) * | 2022-01-05 | 2022-05-24 | 浙江大学 | Blast furnace state monitoring method and device based on multi-mode fusion |
| CN114525372B (en) * | 2022-01-05 | 2022-10-28 | 浙江大学 | Multi-mode fusion based blast furnace state monitoring method and device |
| WO2023130234A1 (en) * | 2022-01-05 | 2023-07-13 | 浙江大学 | Blast furnace state monitoring method and apparatus based on multi-modality fusion |
| US11898212B1 (en) | 2022-01-05 | 2024-02-13 | Zhejiang University | Method and device for monitoring blast furnace state based on multi-modes fusion |
| CN115186772A (en) * | 2022-09-13 | 2022-10-14 | 云智慧（北京）科技有限公司 | Method, device and equipment for detecting partial discharge of power equipment |
| CN115186772B (en) * | 2022-09-13 | 2023-02-07 | 云智慧（北京）科技有限公司 | Method, device and equipment for detecting partial discharge of power equipment |
| CN116124218A (en) * | 2023-02-13 | 2023-05-16 | 正泰电气股份有限公司 | Transformer fault diagnosis method and device, storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334898A (en) | A kind of multi-modal industrial process modal identification and Fault Classification | |
| CN109740175B (en) | Outlier discrimination method for power curve data of wind turbine generator | |
| CN106371427B (en) | Industrial process Fault Classification based on analytic hierarchy process (AHP) and fuzzy Fusion | |
| CN106355030B (en) | A kind of fault detection method based on analytic hierarchy process (AHP) and Nearest Neighbor with Weighted Voting Decision fusion | |
| CN106845526B (en) | A kind of relevant parameter Fault Classification based on the analysis of big data Fusion of Clustering | |
| CN109614997A (en) | A kind of stealing Risk Forecast Method and device based on deep learning | |
| CN108985380B (en) | Point switch fault identification method based on cluster integration | |
| CN105425150B (en) | A kind of Method of Motor Fault Diagnosis based on RBF and PCA-SVDD | |
| CN106055918A (en) | Power system load data identification and recovery method | |
| CN103473540B (en) | The modeling of intelligent transportation system track of vehicle increment type and online method for detecting abnormality | |
| CN110400231B (en) | Failure rate estimation method for electric energy metering equipment based on weighted nonlinear Bayes | |
| CN111369070A (en) | Envelope clustering-based multimode fusion photovoltaic power prediction method | |
| CN109409425B (en) | Fault type identification method based on neighbor component analysis | |
| CN106482967A (en) | A kind of Cost Sensitive Support Vector Machines locomotive wheel detecting system and method | |
| CN105955214B (en) | Batch process fault detection method based on sample time-series and neighbour's affinity information | |
| CN108920863A (en) | A kind of method for building up of robot servo system energy consumption estimation model | |
| CN108508865A (en) | A kind of fault detection method based on distributing OSC-PLS regression models | |
| KR20170078252A (en) | Method and apparatus for time series data monitoring | |
| CN109298225B (en) | Automatic identification model system and method for abnormal state of voltage measurement data | |
| CN115096627B (en) | Method and system for fault diagnosis and operation and maintenance in manufacturing process of hydraulic forming intelligent equipment | |
| JP2006148070A (en) | Correcting method of sensor data and interlock evaluating method of interlock system | |
| CN108345284A (en) | A kind of quality dependent failure detection method becoming gauge block based on two | |
| CN109240276B (en) | Multi-block PCA fault monitoring method based on fault sensitive principal component selection | |
| CN115877806A (en) | LCP film product production control method and system | |
| CN110782546A (en) | Resistivity virtual measurement method of semiconductor PVD (physical vapor deposition) process based on combined tree model |
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 |
Application publication date: 20180727 |
|
| RJ01 | Rejection of invention patent application after publication |