CN103136327A - Time series signifying method based on local feature cluster - Google Patents
Time series signifying method based on local feature cluster Download PDFInfo
- Publication number
- CN103136327A CN103136327A CN2012105838393A CN201210583839A CN103136327A CN 103136327 A CN103136327 A CN 103136327A CN 2012105838393 A CN2012105838393 A CN 2012105838393A CN 201210583839 A CN201210583839 A CN 201210583839A CN 103136327 A CN103136327 A CN 103136327A
- Authority
- CN
- China
- Prior art keywords
- series
- time series
- time
- sub
- local 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.)
- Pending
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a time series signifying method based on a local feature cluster. The time series signifying method based on the local feature cluster includes that original time series is read; a sliding window procedure is called, original time series is divided into multiple sub time series by utilizing of a sliding window; multiple slopes are adopted to show each sub time series of the original time series; a K mean value clustering algorithm is adopted to achieve clustering of the sub time series; and a corresponding sign identification is given to each clustering result. The time series signifying method based on the local feature cluster can well reduce dimensionality and keep form features of time series, is beneficial to further studying of the times series, and further solves the problems that similarity query, classification, clustering, mode digging and the like are directly conducted on the original time series in the prior, low storage and computational efficiency are caused, accuracy and reliability of an algorithm are effected and the like.
Description
Technical field
The present invention relates to Data Mining, particularly relate to a kind of time series symbolism method based on the local feature cluster.
Background technology
At Data Mining, time series data is the important data object of a class, ubiquity in scientific research, business application, traffic control, commercial production.By time series is analyzed and researched, can disclose the inherent law of thing movement, change and progress, the decision-making of the correct understanding things of people also being made accordingly science has important practical significance.Because time series data has high-dimensional, the characteristics such as much noise interference and unstable state, therefore directly carry out the work such as similarity query, classification, cluster, mode excavation on original time series, thus not only can cause storage and counting yield low but also can affect the accuracy of algorithm and reliability and be difficult to obtain the result of being satisfied with.Therefore, before carrying out Time Series Data Mining, need to process time series data.Many researchers have proposed the seasonal effect in time series modal representation, and wherein the time series Symbolic Representation is a kind of important method.
The basic thought of seasonal effect in time series symbolism is mapped to the continuous real number value of seasonal effect in time series or the time series waveform in a period of time on limited symbol table by some discretization methods, time series is converted to the ordered set of limited symbol, namely time series glossary of symbols S={s
1, s
2..., s
| S|Show, wherein | the size (symbol numbers) of S| is-symbol collection.
Summary of the invention
The shortcoming of prior art in view of the above, the object of the present invention is to provide a kind of time series symbolism method based on the local feature cluster, be used for solving prior art and directly carry out the work such as similarity query, classification, cluster, mode excavation on original time series, can cause storage and counting yield low, and can affect the accuracy of algorithm and the problem of reliability.
Reach for achieving the above object other relevant purposes, the invention provides a kind of time series symbolism method based on the local feature cluster, described time series symbolism method comprises: the step that reads original time series; Calling the moving window program utilizes moving window described original time series to be divided into the step of a plurality of Time Sub-series; The step that adopts a plurality of slopes to represent each Time Sub-series of described original time series; Adopt the K means clustering algorithm to realize the step of described Time Sub-series cluster; And the step of giving corresponding symbol logo for each cluster result.
Preferably, the described moving window program of calling utilizes moving window to cut apart pattern M with what described original time series was divided into a plurality of Time Sub-series
segBe defined as follows:
M
seg＝{(b
0,e
0),(b
1,e
1),...(b
k,e
k)}；
Wherein, b
iBe subsequence zero-time, e
iBe the subsequence concluding time.
Preferably, in utilizing moving window described original time series to be divided in the step of a plurality of Time Sub-series, cut apart and be specially based on the above-mentioned pattern of cutting apart:
Setting size is the moving window of W, and the setting original time series is X;
With described original time series X=＜x
0=(x
0, t
0), x
1=(x
1, t
1) ..., x
n=(x
n, t
n) be segmented into a plurality of Time Sub-series; Described a plurality of Time Sub-series of segmentation are expressed as
Individual Time Sub-series S
0, S
1,
Preferably, described each Time Sub-series with described original time series adopts in the step that a plurality of slopes represent, with a plurality of Time Sub-series S
0, S
1,
Represent with SlopeNum slope respectively, respectively this Time Sub-series is:
S
0＝(sp
0,0,sp
0,1,...,sp
0,SlopeNum)；……
Preferably, described each Time Sub-series with original time series adopts in the step that a plurality of slopes represent, the modal representation of described Time Sub-series is t ', is expressed as follows:
t′＝{(sp
0,0,sp
0,1，...,sp
0,SlopeNum),...,
(sp
k,0,sp
k,1,...,sp
k,SlopeNum)}；
Wherein, (sp
i,0, sp
i,1..., sp
I, SlopeNum) be i sub-seasonal effect in time series modal representation, sp
i,jI j slope of sub-seasonal effect in time series.
Preferably, adopt the K means clustering algorithm realize the step of described Time Sub-series cluster and be specially segmentation Time Sub-series S to described original time series for the step that each cluster result is given corresponding symbol logo
0, S
1...,
Adopt the K mean algorithm to carry out cluster, the glossary of symbols size is | S|, and fragment sequence is divided into | the S| class, give corresponding symbolic representation with each class, with the Symbolic Representation of deadline sequence.
As mentioned above, time series symbolism method based on the local feature cluster of the present invention, having following beneficial effect is can be good at dimensionality reduction and has kept the seasonal effect in time series morphological character, be conducive to seasonal effect in time series is further studied, and then solved directly carried out the work such as similarity query, classification, cluster, mode excavation in prior art on original time series, can cause storage and counting yield low, and can affect the problems such as the accuracy of algorithm and reliability.
Description of drawings
Fig. 1 is shown as the time series symbolism method process flow diagram based on the local feature cluster of the present invention.
Fig. 2 is shown as seasonal effect in time series symbolic representation schematic diagram.
Fig. 3 a and Fig. 3 b are shown as the result schematic diagram of distinct symbols collection size.
Fig. 4 a to Fig. 4 c is shown as the result schematic diagram of different iterationses.
Fig. 5 a and Fig. 5 b are shown as the result schematic diagram of different moving windows.
Fig. 6 a and Fig. 6 b are shown as the result schematic diagram of Different Slope number.
Fig. 7 a is shown as the original time series schematic diagram.
Fig. 7 b is shown as and utilizes LFSA method of the present invention to carry out the schematic diagram of time series symbolism.
Fig. 7 c is shown as the schematic diagram after the time series standardization.
Fig. 7 d is shown as the schematic diagram that classical time series symbolism method SAX method is carried out the time series symbolism.
Embodiment
Below by specific instantiation explanation embodiments of the present invention, those skilled in the art can understand other advantages of the present invention and effect easily by the disclosed content of this instructions.The present invention can also be implemented or be used by other different embodiment, and the every details in this instructions also can be based on different viewpoints and application, carries out various modifications or change under spirit of the present invention not deviating from.
For the seasonal effect in time series characteristic, the present invention is in conjunction with the thought of moving window and slope, a kind of algorithm of new time series symbolism has been proposed, namely based on the symbolic algorithm (Symbolic Algorithm Based On Local FeaturesLFSA) of local feature.At first this algorithm utilizes moving window that time series is cut apart expression, then utilizes slope to represent each section of cutting apart, and the recycling clustering algorithm is realized the seasonal effect in time series cluster, at last to symbol logo of each classification.Experiment shows, the method can be good at dimensionality reduction and kept the seasonal effect in time series morphological character, is conducive to seasonal effect in time series is further studied, and specifically details are as follows:
The invention provides a kind of time series symbolism method based on the local feature cluster, described time series symbolism method comprises the following steps:
At first execution in step S1, read original time series; In specific embodiment, computing machine reads and is pre-stored in the original time series of storing in this locality or far-end server.
Then execution in step S2, computing machine calls its slides within window writing routine and utilizes moving window that described original time series is divided into a plurality of Time Sub-series; In specific embodiment, described call the moving window program utilize moving window with described original time series be divided into a plurality of Time Sub-series to cut apart mode-definition as follows:
M
seg＝{(b
0,e
0),(b
1,e
1)，...(b
k,e
k)}；
Wherein, b
iBe subsequence zero-time, e
iBe the subsequence concluding time.
Particularly, described original time series is divided in the step of a plurality of Time Sub-series in the above-mentioned moving window that utilizes, cuts apart and be specially based on the above-mentioned pattern of cutting apart:
Setting size is the moving window of W, and the setting original time series is X;
With described original time series X=＜x
0=(x
0, t
0), x
1=(x
1, t
1) ..., x
n=(x
n, t
n) be segmented into a plurality of Time Sub-series; Described a plurality of Time Sub-series of segmentation are expressed as
Individual Time Sub-series S
0, S
1,
Then execution in step S3 adopts a plurality of slopes to represent each Time Sub-series of described original time series; In specific embodiment, preferably, described each Time Sub-series with described original time series adopts in the step that a plurality of slopes represent, with a plurality of Time Sub-series S
0, S
1,
Represent with SlopeNum slope respectively, respectively this Time Sub-series is:
S
0＝(sp
0,0,sp
0,1,...,sp
0,SlopeNum)；……
Particularly, described each Time Sub-series with original time series adopts in the step that a plurality of slopes represent, the modal representation of described Time Sub-series is t ', is expressed as follows:
t′＝{(SP
0,0,SP
0,1,...,SP
0,SlopeNum),...,
(sp
k,0,sp
k,1,...,sp
k,SlopeNum)}；
Wherein, (sp
i,0, sp
i,1..., sp
I, SlopeNum) be i sub-seasonal effect in time series modal representation, sp
i,jI j slope of sub-seasonal effect in time series.
Then execution in step S4, adopt the K means clustering algorithm to realize described Time Sub-series cluster; In specific embodiment, adopt the K means clustering algorithm realize the step of described Time Sub-series cluster and be specially segmentation Time Sub-series S to described original time series for the step that each cluster result is given corresponding symbol logo
0, S
1...,
Adopt the K mean algorithm to carry out cluster.
Last execution in step S5 gives corresponding symbol logo for each cluster result.In specific embodiment, to the segmentation Time Sub-series S of described original time series
0, S
1...,
Adopt the K mean algorithm to carry out cluster, the glossary of symbols size is | S|, and fragment sequence is divided into | the S| class, give corresponding symbolic representation with each class, with the Symbolic Representation of deadline sequence.
For further setting forth principle of the present invention and effect, see also Fig. 2 to Fig. 7 d, at first, time series of our arbitrary extractings is utilized our symbolism method, this time series is expressed as symbol sebolic addressing, getting moving window size W is 4, and iterations I is 20 times, and each segmentation adopts 2 slopes to represent, the character set size is 4, obtains result as shown in Figure 2.Fig. 2 is shown as seasonal effect in time series symbolic representation schematic diagram, and in Fig. 2, original time series is expressed such character string sequence for " abcadcccacbcddd ".
Utilize symbolic algorithm (LFSA) method, original time series is expressed such character string sequence for " abcadcccacbcddd ", so the method can realize the seasonal effect in time series dimensionality reduction.By observing Fig. 2, we can roughly find out each symbology a kind of local feature, so the SFSA method that we propose not only can effectively reduce the dimension of original time series, and can keep the shape mode information of original time series.
By the elaboration of front, the parameter of LFSA method mainly contains big or small W, iterations I, slope number SlopeNum and the glossary of symbols size of moving window | S|.We concentrate to appoint from experimental data and get time series, and in different parameter situations, result is as shown in Fig. 3 a to 6b respectively.
See also Fig. 3 a and Fig. 3 b, be shown as the result schematic diagram of distinct symbols collection size, in Fig. 3 a, W=4, I=40, SlopeNum=2, | S|=4; In Fig. 3 b, W=4, I=40, SlopeNum=2, | S|=6.
As shown in the figure, comparing us by Fig. 3 a and Fig. 3 b can find, in the situation that other parameters are identical, the size of glossary of symbols has affected the careful program of the represented original time series of symbol sebolic addressing, glossary of symbols is larger, the original time series local feature that symbol sebolic addressing can characterize is more careful, and the segmentation of same symbolic representation is more accurate; Vice versa.But have a problem here, the excessive effect that affects dimensionality reduction of glossary of symbols is so the size of glossary of symbols should be set according to the situation of reality.
See also Fig. 4 a to Fig. 4 c, be shown as the result schematic diagram of different iterationses, in Fig. 4 a, W=4, I=1, SlopeNum=2, | S|=4; In Fig. 4 b, W=4, I=5, SlopeNum=2, | S|=4; In Fig. 4 c, W=4, I=40, SlopeNum=2, | S|=4.
As shown in the figure, by Fig. 4 a, Fig. 4 b, Fig. 4 c three figure as can be known, along with the increase of iterations, the accuracy rate that symbol sebolic addressing represents improves gradually, and each symbolic representation piecewise approximation degree uprises gradually, but, when iterations be increased to a certain degree after, symbol sebolic addressing will can not change, this is owing to having adopted the K means clustering algorithm when cluster is carried out in segmentation, and clustering algorithm exists result to the close characteristics of local optimum, so such phenomenon can occur.In addition, iterations increase the operational efficiency that affects simultaneously algorithm, after we can know that from above iterations reaches certain value, result will no longer change, so iterations is not to be the bigger the better.
See also Fig. 5 a and Fig. 5 b, be shown as the result schematic diagram of different moving windows, in Fig. 5 a, W=4, I=40, SlopeNum=2, | S|=4; In Fig. 5 b, W=6, I=40, SlopeNum=2, | S|=4.
As shown in the figure, we can find by comparison diagram 5a and Fig. 5 b, in the situation that other parameter constants, the size of moving window has affected accuracy and the compressibility of algorithm.Moving window increases, and compressibility uprises, but corresponding accuracy rate descends.Otherwise moving window reduces, and compressibility reduces, and accuracy rate uprises.
See also Fig. 6 a and Fig. 6 b, be shown as the result schematic diagram of Different Slope number, in Fig. 6 a, W=6, I=40, SlopeNum=2, | S|=4; In Fig. 6 b, W=6, I=40, SlopeNum=3, | S|=4.
As shown in the figure, analyzed as can be known by Fig. 6 a and Fig. 6 b, each segmentation represents with more slope, and the accuracy rate of this algorithm has improved, and namely has the enough symbolic representations more accurately of same or analogous original segmentation energy.But the too much slope of this algorithm represents that segmentation will cause algorithm complexity to improve, and operational efficiency reduces, so between accuracy rate and efficient, should be according to actual this parameter that arranges.
The analysis-by-synthesis foregoing has adopted the clustering algorithm of K average to carry out cluster to the segmentation of original time series due to the SFSA algorithm, so iterations I and glossary of symbols size | and S| is the leading indicator that affects algorithm performance.Moving window size W and slope number SlopeNum are the leading indicators that affects algorithm accuracy and compressibility.
For advantage of the present invention is described, we choose classical time series symbolism method (Symbolic AggregateApproximation, SAX) method and method in this paper compare experiment, at first our relatively optional time series.The SFSA method arranges W=4, I=40, and SlopeNum=2, | S|=6 and SAX method arrange W=4, | S=6|, result is as shown in Fig. 7 a to Fig. 7 d.Fig. 7 a is shown as the original time series schematic diagram; Fig. 7 b is shown as and utilizes LFSA method of the present invention to carry out the schematic diagram of time series symbolism; Fig. 7 c is shown as the schematic diagram after the time series standardization; Fig. 7 d is shown as the schematic diagram that classical time series symbolism method SAX method is carried out the time series symbolism.
By analysis chart 7c and Fig. 7 d as can be known, the SAX method adopts the approximate representation method of PAA, adopts mean value to represent each segmentation, has ignored original local feature information of original time series, the description time series that can only be similar to roughly feature with; Method in this paper is dimensionality reduction effectively, considered simultaneously the local feature of original time series, utilize slope to represent segmentation, the information that can keep as much as possible original time series, a shape mode of original time series that the method has made each symbology makes result more accurate.Therefore, SFSA method in this paper can comprise the information of more original time series, and the method provides descriptor more comprehensively than SAX method, for follow-up further Time Series Data Mining provides reliable assurance.
In sum, time series symbolism method based on the local feature cluster of the present invention, having following beneficial effect is can be good at dimensionality reduction and has kept the seasonal effect in time series morphological character, be conducive to seasonal effect in time series is further studied, and then solved directly carried out the work such as similarity query, classification, cluster, mode excavation in prior art on original time series, can cause storage and counting yield low, and can affect the problems such as the accuracy of algorithm and reliability.So the present invention has effectively overcome various shortcoming of the prior art and the tool high industrial utilization.
Above-described embodiment is illustrative principle of the present invention and effect thereof only, but not is used for restriction the present invention.Any person skilled in the art scholar all can under spirit of the present invention and category, modify or change above-described embodiment.Therefore, have in technical field under such as and know that usually the knowledgeable modifies or changes not breaking away from all equivalences of completing under disclosed spirit and technological thought, must be contained by claim of the present invention.
Claims (6)
1. the time series symbolism method based on the local feature cluster, is characterized in that, described time series symbolism method comprises:
Read the step of original time series;
Calling the moving window program utilizes moving window described original time series to be divided into the step of a plurality of Time Sub-series;
The step that adopts a plurality of slopes to represent each Time Sub-series of described original time series;
Adopt the K means clustering algorithm to realize the step of described Time Sub-series cluster; And
Give the step of corresponding symbol logo for each cluster result.
2. the time series symbolism method based on the local feature cluster according to claim 1 is characterized in that: the described moving window program of calling utilizes moving window to cut apart pattern M with what described original time series was divided into a plurality of Time Sub-series
segBe defined as follows:
M
seg＝{(b
0,e
0),(b
1,e
1),...(b
k,e
k)}；
Wherein, b
iBe subsequence zero-time, e
iBe the subsequence concluding time.
3. the time series symbolism method based on the local feature cluster according to claim 2, it is characterized in that: in utilizing moving window described original time series to be divided in the step of a plurality of Time Sub-series, cut apart and be specially based on the above-mentioned pattern of cutting apart:
Setting size is the moving window of W, and the setting original time series is X;
With described original time series X=＜x
0=(x
0, t
0), x
1=(x
1, t
1) ..., x
n=(x
n, t
n) be segmented into a plurality of Time Sub-series;
Described a plurality of Time Sub-series of segmentation are expressed as
Individual Time Sub-series S
0, S
1,
4. the time series symbolism method based on the local feature cluster according to claim 3 is characterized in that: described each Time Sub-series with described original time series adopts in the step that a plurality of slopes represent, with a plurality of Time Sub-series S
0, S
1,
Represent with SlopeNum slope respectively, respectively this Time Sub-series is: S
0=(sp
0,0, sp
0,1..., sp
0, SlopeNum);
5. the time series symbolism method based on the local feature cluster according to claim 4, it is characterized in that: described each Time Sub-series with original time series adopts in the step that a plurality of slopes represent, the modal representation of described Time Sub-series is t ', is expressed as follows:
t′＝{(sp
0,0,sp
0,1，...,sp
0,SlopeNum),...,
(sp
k,0,sp
k,1，...,sp
k,SlopeNum)}；
Wherein, (sp
i,0, sp
i,1..., sp
I, SlopeNum) be i sub-seasonal effect in time series modal representation, sp
i,jI j slope of sub-seasonal effect in time series.
6. the time series symbolism method based on the local feature cluster according to claim 5 is characterized in that: adopt the K means clustering algorithm realize the step of described Time Sub-series cluster and be specially segmentation Time Sub-series S to described original time series for the step that each cluster result is given corresponding symbol logo
0, S
1...,
Adopt the K mean algorithm to carry out cluster, the glossary of symbols size is | S|, and fragment sequence is divided into | the S| class, give corresponding symbolic representation with each class, with the Symbolic Representation of deadline sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105838393A CN103136327A (en) | 2012-12-28 | 2012-12-28 | Time series signifying method based on local feature cluster |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105838393A CN103136327A (en) | 2012-12-28 | 2012-12-28 | Time series signifying method based on local feature cluster |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103136327A true CN103136327A (en) | 2013-06-05 |
Family
ID=48496154
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012105838393A Pending CN103136327A (en) | 2012-12-28 | 2012-12-28 | Time series signifying method based on local feature cluster |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103136327A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104714953A (en) * | 2013-12-12 | 2015-06-17 | 日本电气株式会社 | Time series data motif identification method and device |
| CN104820673A (en) * | 2015-03-27 | 2015-08-05 | 浙江大学 | Time sequence similarity measurement method based on self-adaptive piecewise statistical approximation |
| CN105091732A (en) * | 2015-08-13 | 2015-11-25 | 国家电网公司 | Method and system for detecting deformation of transformer winding |
| CN105302867A (en) * | 2015-09-28 | 2016-02-03 | 浙江宇视科技有限公司 | Search engine check method and apparatus |
| CN105786823A (en) * | 2014-12-19 | 2016-07-20 | 日本电气株式会社 | System and method for multidimensional timing sequence data analysis |
| CN106095787A (en) * | 2016-05-30 | 2016-11-09 | 重庆大学 | A kind of Symbolic Representation method of time series data |
| CN106909664A (en) * | 2017-02-28 | 2017-06-30 | 国网福建省电力有限公司 | A kind of power equipment data stream failure recognition methods |
| CN107392979A (en) * | 2017-06-29 | 2017-11-24 | 天津大学 | The two dimensional visible state composition and quantitative analysis index method of time series |
| CN110032585A (en) * | 2019-04-02 | 2019-07-19 | 北京科技大学 | A kind of time series bilayer symbolism method and device |
| CN110986915A (en) * | 2019-12-13 | 2020-04-10 | 西安航天精密机电研究所 | Real-time compensation method for temperature drift of fiber-optic gyroscope |
| CN112131322A (en) * | 2020-09-22 | 2020-12-25 | 腾讯科技（深圳）有限公司 | Time series classification method and device |
| CN113157768A (en) * | 2021-04-09 | 2021-07-23 | 天津大学 | Heating ventilation air conditioner operation data association attribute mining method and system |
| CN115952443A (en) * | 2023-03-09 | 2023-04-11 | 湖南纽帕科技有限公司 | Energy storage application wind power plant sliding window clustering method based on SOM network |
-
2012
- 2012-12-28 CN CN2012105838393A patent/CN103136327A/en active Pending
Non-Patent Citations (6)
| Title |
|---|
| DAS G等: "Rule discovery from time series", 《FOURTH ANNUAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING》 * |
| 李宏等: "基于时序模式关联的股票走势研究", 《计算机工程与应用》 * |
| 王波等: "一种基于云模型的时间序列特征表示方法", 《2005中国控制与决策学术年会论文集》 * |
| 蒋嵘等: "基于形态表示的时间序列相似性搜索", 《计算机研究与发展》 * |
| 钟清流等: "基于统计特征的时序数据符号化算法", 《计算机学报》 * |
| 陈湘涛等: "基于分割模式的时间序列矢量符号化算法", 《计算机工程》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104714953A (en) * | 2013-12-12 | 2015-06-17 | 日本电气株式会社 | Time series data motif identification method and device |
| CN105786823A (en) * | 2014-12-19 | 2016-07-20 | 日本电气株式会社 | System and method for multidimensional timing sequence data analysis |
| CN105786823B (en) * | 2014-12-19 | 2019-06-28 | 日本电气株式会社 | System and method for the analysis of multi-dimensional time sequence data |
| CN104820673B (en) * | 2015-03-27 | 2018-03-06 | 浙江大学 | Time Series Similarity measure based on adaptivity segmentation statistical approximation |
| CN104820673A (en) * | 2015-03-27 | 2015-08-05 | 浙江大学 | Time sequence similarity measurement method based on self-adaptive piecewise statistical approximation |
| CN105091732A (en) * | 2015-08-13 | 2015-11-25 | 国家电网公司 | Method and system for detecting deformation of transformer winding |
| CN105302867A (en) * | 2015-09-28 | 2016-02-03 | 浙江宇视科技有限公司 | Search engine check method and apparatus |
| CN105302867B (en) * | 2015-09-28 | 2019-06-11 | 浙江宇视科技有限公司 | A kind of search engine inquiry method and device |
| CN106095787A (en) * | 2016-05-30 | 2016-11-09 | 重庆大学 | A kind of Symbolic Representation method of time series data |
| CN106909664A (en) * | 2017-02-28 | 2017-06-30 | 国网福建省电力有限公司 | A kind of power equipment data stream failure recognition methods |
| CN107392979A (en) * | 2017-06-29 | 2017-11-24 | 天津大学 | The two dimensional visible state composition and quantitative analysis index method of time series |
| CN107392979B (en) * | 2017-06-29 | 2019-10-18 | 天津大学 | The two dimensional visible state composition and quantitative analysis index method of time series |
| CN110032585A (en) * | 2019-04-02 | 2019-07-19 | 北京科技大学 | A kind of time series bilayer symbolism method and device |
| CN110986915A (en) * | 2019-12-13 | 2020-04-10 | 西安航天精密机电研究所 | Real-time compensation method for temperature drift of fiber-optic gyroscope |
| CN112131322A (en) * | 2020-09-22 | 2020-12-25 | 腾讯科技（深圳）有限公司 | Time series classification method and device |
| CN112131322B (en) * | 2020-09-22 | 2023-10-10 | 腾讯科技（深圳）有限公司 | Time sequence classification method and device |
| CN113157768A (en) * | 2021-04-09 | 2021-07-23 | 天津大学 | Heating ventilation air conditioner operation data association attribute mining method and system |
| CN115952443A (en) * | 2023-03-09 | 2023-04-11 | 湖南纽帕科技有限公司 | Energy storage application wind power plant sliding window clustering method based on SOM network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103136327A (en) | Time series signifying method based on local feature cluster | |
| CN107992982B (en) | A kind of Default Probability prediction technique of the unstructured data based on deep learning | |
| Awal et al. | A global learning approach for an online handwritten mathematical expression recognition system | |
| CN104391942B (en) | Short essay eigen extended method based on semantic collection of illustrative plates | |
| CN109753660B (en) | LSTM-based winning bid web page named entity extraction method | |
| CN110232114A (en) | Sentence intension recognizing method, device and computer readable storage medium | |
| CN105117054B (en) | A kind of recognition methods of handwriting input and system | |
| EP4009219A1 (en) | Analysis of natural language text in document using hierarchical graph | |
| CN101127042A (en) | Sensibility classification method based on language model | |
| CN105205124B (en) | A kind of semi-supervised text sentiment classification method based on random character subspace | |
| CN102902826B (en) | A kind of image method for quickly retrieving based on reference picture index | |
| CN104517112A (en) | Table recognition method and system | |
| CN102591612B (en) | General webpage text extraction method based on punctuation continuity and system thereof | |
| CN106250372A (en) | A kind of Chinese electric power data text mining method for power system | |
| CN110222184A (en) | A kind of emotion information recognition methods of text and relevant apparatus | |
| Gajawada et al. | Missing value imputation method based on clustering and nearest neighbours | |
| CN102750552B (en) | Handwriting recognition method and system as well as handwriting recognition terminal | |
| CN104899601A (en) | Identification method of handwritten Uyghur words | |
| CN114647741A (en) | Process automatic decision and reasoning method, device, computer equipment and storage medium | |
| CN110222328A (en) | Participle and part-of-speech tagging method, apparatus, equipment and storage medium neural network based | |
| CN103106211B (en) | Emotion recognition method and emotion recognition device for customer consultation texts | |
| CN101583147A (en) | Segmental online support vector regression method applied in traffic prediction | |
| CN104809105A (en) | Method and system for identifying event argument and argument role based on maximum entropy | |
| CN102314252B (en) | Character segmentation method and device for handwritten character string | |
| CN110347841A (en) | A kind of method, apparatus, storage medium and the electronic equipment of document content classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130605 |