CN109161931A - Aluminium electrolysis anode electric current classification method based on shapelet conversion - Google Patents
Aluminium electrolysis anode electric current classification method based on shapelet conversion Download PDFInfo
- Publication number
- CN109161931A CN109161931A CN201811231278.4A CN201811231278A CN109161931A CN 109161931 A CN109161931 A CN 109161931A CN 201811231278 A CN201811231278 A CN 201811231278A CN 109161931 A CN109161931 A CN 109161931A
- Authority
- CN
- China
- Prior art keywords
- shapelet
- shapelets
- distance
- high quality
- core feature
- 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
-
- C—CHEMISTRY; METALLURGY
- C25—ELECTROLYTIC OR ELECTROPHORETIC PROCESSES; APPARATUS THEREFOR
- C25C—PROCESSES FOR THE ELECTROLYTIC PRODUCTION, RECOVERY OR REFINING OF METALS; APPARATUS THEREFOR
- C25C3/00—Electrolytic production, recovery or refining of metals by electrolysis of melts
- C25C3/06—Electrolytic production, recovery or refining of metals by electrolysis of melts of aluminium
- C25C3/20—Automatic control or regulation of cells
Abstract
The invention discloses a kind of aluminium electrolysis anode electric current classification methods based on shapelet conversion, choose high quality shapelets based on shapelet conversion method；Core feature is extracted from the high quality shapelets；Increase similar amt feature into the core feature, and change data.The present invention finds core feature from all dimensions of a classification, classification accuracy will not be influenced by signal also not dependent on dimension, it solves the problems, such as that anode current signal is difficult to be classified by general multivariate classification method, and improves the accuracy of classification.
Description
Technical field
The present invention relates to industrial control fields, and in particular to a kind of aluminium electrolysis anode electric current point based on shapelet conversion
Class method.
Background technique
Aluminium cell is the industrial process system of a non-linear, multiple coupled, time-varying and large dead time.Aluminium cell is by slot
What voltage and other ancillary measures were controlled.Automatic process control in aluminium cell is the standard configuration of modern production line.
As electrolytic cell is increasing, aluminum electrolysis process is a challenging production process.Anode current is by voltage
Control, it reflects the variation of resistance components.Therefore, anode current can reflect the local groove condition around anode.Anode electricity
Stream provides important information for aluminium electrolysis process production control.
Anode reaction discharges bubble formation bubble layer.The resistance of bubble layer reflects anode life, anode direction and inclines
Angle.Simultaneously, the resistance of electrolytic cell is by alumina concentration, the influence of bath temperature and electrolyte ingredient.Influence the imbalance of resistance
Condition will will lead to different Current distribution in anode.For example, the corresponding anode current of anode during change poles will be reduced to
0 because the loss of each anode be it is different, the anode needed replacing be also it is unfixed, therefore, in whole slot, bust
Anode current for 0 is unfixed, so the otherness of anode distribution can be generated, same category of anode current signal has
The feature of discrimination property is fixed on certain dimension.
Multi-source time series is a kind of important time series.They are widely used in many fields, such as voice is known
Not, multimedia application, medicine, economics, science and engineering.Multi-source time series classification rises in the data mining of time series
Important role.However, the multivariable of multi-source time series and the inconsistent of length allow with traditional machine learning method come
Classification multi-source time series is difficult to.Therefore, for multi-source time series of classifying, a large amount of research and method are suggested, greatly
Partial research is defined in the method for extracting feature.In these methods, many spies can be extracted from multi-source time series
Sign goes to indicate original time series.Its advantages are mainly that can reduce the dimension either time series of multi-source time series
Length avoid the dimension from exploding.Core feature is a kind of feature for expressing multi-source time series, it is by clustering same class
What the mode of the candidate shapelet of other same dimension obtained, general same category of multi-source time series has with dimension
One or more core features.Two methods of MCFEC-QBC and MCFEC-rule are exactly with core feature come time series of classifying,
They classify to each dimension of sample according to the similitude with core feature, finally according to the primary categories of all dimensions
To determine the classification of sample.Multi-source shapelets discovery (MSD) method is to extract multi-source shapelets, wherein multi-source
Each shapelet of shapelets is extracted from each dimension, they are by calculating shapelets and time sequence
Euclidean distance between column compares multi-source time series.
Most multi-source time series classification method is based on more with same characteristic features in each dimension of same category
Source time sequence.However, it is not to be fixed on certain dimension that same category of anode current signal, which has the feature of discrimination property, institute
It may not apply to anode current signal to be mostly based on the classification method of multi-source time series.
Summary of the invention
The purpose of the present invention is to provide a kind of aluminium electrolysis anode electric current classification methods based on shapelet conversion, can be with
Solve the problems, such as that anode current signal cannot be classified because of special dimension by common multi-source time series, to aluminium cell
Local groove condition analysis and aluminium electroloysis intelligence production provide important support.
The invention adopts the following technical scheme:
A kind of aluminium electrolysis anode electric current classification method based on shapelet conversion, which comprises
S1, high quality shapelets is chosen based on shapelet conversion method；
S2, core feature is extracted from the high quality shapelets；
S3, increase similar amt feature into the core feature, and change data.
Further, the S1 includes:
S11, candidate shapelets is always extracted from primordial time series data；
S12, similarity assessment is carried out to the shapelet in the candidate shapelets, deletes self similarity
Shapelets obtains the shapelets of high quality.
Further, the similarity assessment includes:
It calculates candidate shapelets each multi-source time series distance into data set and obtains distance set；
Calculate the information gain of each distance set.
Further, the calculation formula of the information gain are as follows:
Wherein, O is one group of orderly distance set, O can be divided into two set, set O by split pointLMiddle element
Value be both less than the value of split point, set ORThe value of middle element is both greater than the value of split point, H (O), H (OL)、H(OR) respectively indicate
Raw data set O is split off and is a little divided into two-part data set OLWith data set ORComentropy；
A highest shapelet of information gain is only chosen in the same source time sequence as high quality
shapelet。
Further, the S2 includes:
The core spy that the high quality shapelets extracts the aluminium electrolysis anode current signal is clustered based on distance matrix
Sign.
Further, each element in the distance matrix is the distance between the high quality shapelets.
Further, detailed process is as follows for the cluster:
The second small element value in the distance matrix is found, is deleted corresponding to the second small element value
Shapelets adds the cluster set of deleted shapelets into the distance matrix；
Calculate the cluster set of each high quality shapelet in the distance matrix and deleted shapelets away from
From updating the distance matrix；
Above-mentioned cluster process is repeated until all high quality shapelets are gathered；
The shapelet of high information gain in each cluster is selected as core feature.
Further, the distance between each high quality shapelet and the deleted cluster set of shapelets
It is the average value of each high quality shapelet shapelet distance each deleted into cluster set.
Further, the S3 includes:
It calculates the core feature and data concentrates the distance between each multi-source time series；
Calculate core feature sequence number similar with data set；
Obtain to represent the quantative attribute of the core feature frequency of occurrences according to the distance and number；
It the core feature and the quantative attribute is combined into new data set is put into inside classifier and go to be trained.
The advantages and beneficial effects of the present invention are:
The present invention provides a kind of aluminium electrolysis anode electric current classification method based on shapelet conversion, from the institute of a classification
Have and find core feature in dimension, classification accuracy will not be influenced by signal also not dependent on dimension, and anode current letter is solved
It number is difficult to by general multivariate classification method come the problem of classification, and improves the accuracy of classification.
Detailed description of the invention
Fig. 1 is that the present invention is based on the aluminium electrolysis anode current signal classification method flow diagrams that shapelet is converted；
Fig. 2 is the process schematic that the present invention is clustered with distance matrix.
Specific embodiment
With reference to the accompanying drawings and examples, further description of the specific embodiments of the present invention.Following embodiment is only
For clearly illustrating technical solution of the present invention, and not intended to limit the protection scope of the present invention.
As shown in Figure 1, the present invention provides the aluminium electrolysis anode current signal classification method converted based on shapelet, packet
It includes:
S1, high quality shapelets is chosen based on shapelet conversion method；
S2, core feature is extracted from the high quality shapelets；
S3, increase similar amt feature into the core feature, and change data.
The aluminium electrolysis anode current signal classification method process of the invention based on shapelet conversion is done into one below
Walk detailed elaboration.
One, high quality shapelets is chosen with shapelet conversion method.
The shapelets for obtaining high quality mainly includes two steps: the first step is always mentioned from primordial time series data
Take candidate shapelet.Second step is that the shapelet of self similarity in deletion candidate shapelet obtains high quality
shapelet。
Shapelet is a cross-talk sequence of time series, can be from time series by the way that shapelet length is arranged
Obtain many subsequences.All subsequences obtained in all time serieses are referred to as candidate shapelet.Obtain all times
After selecting shapelet, quality evaluation is carried out to candidate shapelet.
Firstly, the method for definition two sub- sequence similarities of assessment, that is, calculate the Euclidean distance between two subsequences, away from
From smaller, similarity is higher.Shown in the following formula 1 of range formula.Wherein A and B is the son of the time series in two sources respectively
Sequence, they are all one group of orderly sequence of real numbers, wherein A=(a1, a2 ..., am), B=(b1, b2..., bm)。
Then the similitude of assessment candidate shapelet and time series, i.e. calculating candidate shapelet and time series
Minimum range.Shown in the following formula 2 of calculation formula.
Dist (A, T)=min (dist (A, w)) (2)
In formula 2, w indicates any subsequence in time series T, and formula 2 indicates candidate shapeletA and time series
The distance of T is exactly the minimum range of candidate shapelet A Yu all subsequences of time series T.It is candidate that assessment is defined simultaneously
The method of shapelet and multi-source Time Series Similarity, shown in the following formula 3 of calculation formula.
Distance (A, M)=min (Dist (A, Ti)) (3)
Wherein multi-source time series M can be by multiple source time TiComposition.Candidate shapelet A and multi-source time
The distance between sequence M is exactly the minimum range of a source time sequence in A and multi-source time series.For each candidate
Shapelet A and data set Mw, it is by the shapelet distance set that each multi-source time series distance forms into data set
Shown in formula 4.
D={ distance (A, M1), distance (A, M2) ..., distance (A, Mw)} (4)
By calculating the information gain of each distance set, to assess the quality of each shapelet, the meter of information gain
It is as shown in formula 5 to calculate formula.
Wherein O is one group of orderly distance set, O can be divided into two set, set O by split pointLMiddle element
Value be both less than the value of split point, set ORThe value of middle element is both greater than the value of split point.In the set of candidate classification point,
The available highest split point of information gain, the set of split point are generated by formula 6.
Wherein diIt is the element in distance set D.
Compare the shapelet with interpretation in order to obtain, deletes the candidate shapelet of self similarity.Same one
A highest shapelet of information gain is only chosen in source time sequence.Finally, available many effective
shapelets。
Two, core feature is extracted from effective shapelet.
If shapelets has high quality in a class, they are always more more like than the shape of other classes.Therefore, have
Necessity extracts the shapelet that part can be distinguished very well in class.In addition, anode current signal (ACS) can in a class
There can be different forms, i.e., can not find identical form in one dimension.Therefore, it is impossible to the method for shapelets vector
Multi-source time series is compared.
In order to classify to ACS, the central characteristics of ACS are extracted by cluster shapelets in a class.Cluster
Process be based on distance matrix.The coordinate of distance matrix is made of effective shapelet in previous step, in matrix
Each element is the distance between effective shapelet of respective coordinates.Fig. 2 Far Left is a distance matrix, wherein S1-SnFor
Effective shapelet.Numerical value is the distance between shapelet value in figure.
As shown in Figure 2.It is then, the second small element value institute is right firstly, find the second small element value in distance matrix
The shapelets answered is deleted, the cluster set (S of deleted shapelet1, S2) be added in distance matrix.Each have
The distance between effect shapelet and cluster are the average value of shapelet each shapelet distance into cluster.It is having updated
Distance matrix and then process above is come again until all shapelets are gathered.Finally, each poly-
The shapelet of high information gain in class is chosen as core feature.
Three, increase similar amt feature and change data.
Since core feature is extracted from all dimensions of multi-source time series, core feature therefore cannot generation
The dimension number difference of table feature.For example, it cannot distinguish between the time series containing core feature and multiple dimensions all contain
There is the time series of core feature, it may not be possible to distinguish noise signal and normal signal.Therefore, it is necessary to increase a similar amt
It is gone inside feature to core feature, it can represent the frequency that core feature occurs in multi-source time series.
For change data, first calculating core feature and data set MwIn the distance between each multi-source time series,
It calculates core feature sequence number similar with data set again later, is configured to represent core spy according to the distance and number
The data characteristics is added to as a kind of new feature and synthesizes new data in core feature by the quantative attribute for levying the frequency of occurrences
Collection obtains the matrix that new data are a 2*u if the number of core feature is u.This data set matrix can be put into biography
It goes to be trained inside the classifier of system such as XGBoost, Svm etc. devices, so as to well divide anode current signal
Class.
The anode current Modulation recognition method based on shapelet conversion method that the invention proposes a kind of.By extracting energy
It is special to increase a kind of quantity that can represent the core feature frequency of occurrences for the core feature for enough identifying the classification of anode current signal
Sign is put into the feature extracted inside common classifier and goes to be trained, so as to believe well anode current
Number classify.This method can embody the special dimension of anode current signal relative to traditional source signal classification method
Feature, so as to anode current signal of classifying well.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (9)
1. a kind of aluminium electrolysis anode electric current classification method based on shapelet conversion, which is characterized in that the described method includes:
S1, high quality shapelets is chosen based on shapelet conversion method；
S2, core feature is extracted from the high quality shapelets；
S3, increase similar amt feature into the core feature, and change data.
2. the method as described in claim 1, which is characterized in that the S1 includes:
S11, candidate shapelets is always extracted from primordial time series data；
S12, similarity assessment is carried out to the shapelet in the candidate shapelets, the shapelets for deleting self similarity is obtained
To the shapelets of high quality.
3. method according to claim 2, which is characterized in that the similarity assessment includes:
It calculates candidate shapelets each multi-source time series distance into data set and obtains distance set；
Calculate the information gain of each distance set.
4. method as claimed in claim 3, which is characterized in that the calculation formula of the information gain are as follows:
Wherein, O is one group of orderly distance set, O can be divided into two set, set O by split pointLThe value of middle element
The both less than value of split point, set ORThe value of middle element is both greater than the value of split point, H (O), H (OL)、H(OR) respectively indicate it is original
Data set O is split off and is a little divided into two-part data set OLWith data set ORComentropy；
Shapelet of the highest shapelet of information gain as high quality is only chosen in the same source time sequence.
5. the method as described in claim 1, which is characterized in that the S2 includes:
The core feature that the high quality shapelets extracts the aluminium electrolysis anode current signal is clustered based on distance matrix.
6. method as claimed in claim 5, which is characterized in that each element in the distance matrix is the high quality
Distance between shapelets.
7. method as claimed in claim 6, which is characterized in that detailed process is as follows for the cluster:
The second small element value in the distance matrix is found, shapelets corresponding to the second small element value is deleted, adds
Add the cluster set of deleted shapelets into the distance matrix；
The distance of the cluster set of each high quality shapelet and deleted shapelets in the distance matrix is calculated,
Update the distance matrix；
Above-mentioned cluster process is repeated until all high quality shapelets are gathered；
The shapelet of high information gain in each cluster is selected as core feature.
8. the method for claim 7, which is characterized in that each high quality shapelet and be deleted
The distance between cluster set of shapelets is that each high quality shapelet is each deleted into cluster set
The average value of shapelet distance.
9. the method as described in claim 1, which is characterized in that the S3 includes:
It calculates the core feature and data concentrates the distance between each multi-source time series；
Calculate core feature sequence number similar with data set；
Obtain to represent the quantative attribute of the core feature frequency of occurrences according to the distance and number；
It the core feature and the quantative attribute is combined into new data set is put into inside classifier and go to be trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811231278.4A CN109161931B (en) | 2018-10-22 | 2018-10-22 | Aluminium electrolysis anode electric current classification method based on shapelet conversion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811231278.4A CN109161931B (en) | 2018-10-22 | 2018-10-22 | Aluminium electrolysis anode electric current classification method based on shapelet conversion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109161931A true CN109161931A (en) | 2019-01-08 |
| CN109161931B CN109161931B (en) | 2019-11-08 |
Family
ID=64878841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811231278.4A Active CN109161931B (en) | 2018-10-22 | 2018-10-22 | Aluminium electrolysis anode electric current classification method based on shapelet conversion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109161931B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113608146A (en) * | 2021-08-06 | 2021-11-05 | 云南电网有限责任公司昆明供电局 | Fault line selection method suitable for forest fire high-resistance grounding condition |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104020845A (en) * | 2014-03-27 | 2014-09-03 | 浙江大学 | Acceleration transducer placement-unrelated movement recognition method based on shapelet characteristic |
| CN104809226A (en) * | 2015-05-07 | 2015-07-29 | 武汉大学 | Method for early classifying imbalance multi-variable time sequence data |
| CN105203922A (en) * | 2015-09-17 | 2015-12-30 | 国家电网公司 | Electric transmission line fault positioning method based on multiple classifiers |
| CN105975932A (en) * | 2016-05-04 | 2016-09-28 | 广东工业大学 | Gait recognition and classification method based on time sequence shapelet |
| CN107016354A (en) * | 2017-03-16 | 2017-08-04 | 中南大学 | The feature mode extracting method and its system of aluminium electrolysis anode current sequence |
| US20170227584A1 (en) * | 2016-02-05 | 2017-08-10 | Kabushiki Kaisha Toshiba | Time-series data waveform analysis device, method therefor and non-transitory computer readable medium |
| CN108277506A (en) * | 2018-02-07 | 2018-07-13 | 中南大学 | A kind of adaptive oversampler method of aluminium electroloysis degree of superheat unbalanced data |
-
2018
- 2018-10-22 CN CN201811231278.4A patent/CN109161931B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104020845A (en) * | 2014-03-27 | 2014-09-03 | 浙江大学 | Acceleration transducer placement-unrelated movement recognition method based on shapelet characteristic |
| CN104809226A (en) * | 2015-05-07 | 2015-07-29 | 武汉大学 | Method for early classifying imbalance multi-variable time sequence data |
| CN105203922A (en) * | 2015-09-17 | 2015-12-30 | 国家电网公司 | Electric transmission line fault positioning method based on multiple classifiers |
| US20170227584A1 (en) * | 2016-02-05 | 2017-08-10 | Kabushiki Kaisha Toshiba | Time-series data waveform analysis device, method therefor and non-transitory computer readable medium |
| CN105975932A (en) * | 2016-05-04 | 2016-09-28 | 广东工业大学 | Gait recognition and classification method based on time sequence shapelet |
| CN107016354A (en) * | 2017-03-16 | 2017-08-04 | 中南大学 | The feature mode extracting method and its system of aluminium electrolysis anode current sequence |
| CN108277506A (en) * | 2018-02-07 | 2018-07-13 | 中南大学 | A kind of adaptive oversampler method of aluminium electroloysis degree of superheat unbalanced data |
Non-Patent Citations (2)
| Title |
|---|
| 原继东等: "基于Shapelet剪枝和覆盖的时间序列分类算法", 《软件学报》 * |
| 赵慧赟等: "基于shapelets学习的多元时间序列分类_", 《计算机科学》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113608146A (en) * | 2021-08-06 | 2021-11-05 | 云南电网有限责任公司昆明供电局 | Fault line selection method suitable for forest fire high-resistance grounding condition |
| CN113608146B (en) * | 2021-08-06 | 2023-12-19 | 云南电网有限责任公司昆明供电局 | Fault line selection method suitable for forest fire under high-resistance grounding condition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109161931B (en) | 2019-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ding et al. | Towards better validity: Dispersion based clustering for unsupervised person re-identification | |
| CN104992191B (en) | The image classification method of feature and maximum confidence path based on deep learning | |
| US11804069B2 (en) | Image clustering method and apparatus, and storage medium | |
| CN107944410B (en) | Cross-domain facial feature analysis method based on convolutional neural network | |
| CN107066555B (en) | On-line theme detection method for professional field | |
| CN109492673A (en) | A kind of unbalanced data prediction technique based on spectral clustering sampling | |
| CN106250925B (en) | A kind of zero Sample video classification method based on improved canonical correlation analysis | |
| CN112668544A (en) | Pedestrian re-identification method based on hard sample confusion and feature robustness enhancement | |
| CN109161931B (en) | Aluminium electrolysis anode electric current classification method based on shapelet conversion | |
| CN110008828A (en) | Pairs of constraint ingredient assay measures optimization method based on difference regularization | |
| CN111191033B (en) | Open set classification method based on classification utility | |
| CN113779283B (en) | Fine-grained cross-media retrieval method with deep supervision and feature fusion | |
| Wan et al. | A novel shapelet transformation method for classification of multivariate time series with dynamic discriminative subsequence and application in anode current signals | |
| JP2002183171A (en) | Document data clustering system | |
| CN109857892B (en) | Semi-supervised cross-modal Hash retrieval method based on class label transfer | |
| CN109597901B (en) | Data analysis method based on biological data | |
| CN112668633B (en) | Adaptive graph migration learning method based on fine granularity field | |
| CN110647897A (en) | Zero sample image classification and identification method based on multi-part attention mechanism | |
| CN111738039A (en) | Pedestrian re-identification method, terminal and storage medium | |
| CN113177521B (en) | Intelligent radiation source identification method based on combined twin network | |
| CN114155403A (en) | Image segmentation Hash sorting method based on deep learning | |
| Wang et al. | A novel distance learning for elastic cross-modal audio-visual matching | |
| CN111259176B (en) | Cross-modal Hash retrieval method based on matrix decomposition and integrated with supervision information | |
| Hsu et al. | Automatic facial expression recognition for affective computing based on bag of distances | |
| Alam et al. | Automatic clustering for selection of optimal number of clusters by K-means integrated with enhanced firefly algorithms |
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 |