KR20190116032A - A novel join technique for similar-trend searches supporting normalization on time-series databases - Google Patents

Abstract

The present invention relates to a subsequence matching method of a time series database, which matches a query subsequence from an index generated for a data sequence. The subsequence matching method of a time series database comprises the steps of: dividing a given query sequence into a plurality of query windows, and determining the number of query windows corresponding to the query processing from the query sequence among the divided query windows; selecting a query window by calculating a query cost with respect to the determined number of query windows; recalculating a similarity tolerance range by using the selected query window and a preset similarity tolerance value; and retrieving an answer candidate subsequence matching the query sequence from the index using a range query constructed based on the recalculated similarity tolerance range.

Description

Subsequence matching system and method thereof of time series database {A NOVEL JOIN TECHNIQUE FOR SIMILAR-TREND SEARCHES SUPPORTING NORMALIZATION ON TIME-SERIES DATABASES}

The subsequence matching method in the time-series database of the present invention normalizes a given data sequence and a given query sequence, improves the accuracy of the search by measuring similarity using Euclidean distance, and provides various correct answers in post-processing. The present invention relates to a subsequence matching system of a time series database and a method thereof that can reduce an error solution by using an intersection of candidate sets.

The searching method for an answer sequence for a query sequence is a method for searching for an answer sequence to be matched with a query sequence among data sequences according to the similarity between the query sequence and the data sequence.

In a conventional method for searching an answer sequence for a query sequence, the answer sequence is searched by calculating a similarity measure between the data sequence and the query sequence.

However, since the length of the query sequence may change according to a user's input, the database system cannot predict the length of the query sequence. Therefore, the conventional method for retrieving the answer sequence is not able to construct and search the index for all the various lengths of the query sequence.

Therefore, the prior art has used sequential search optimized in various ways for subsequence matching that supports normalization transformation and time warping.

However, the sequential search has a problem in that it is much slower than a search using an index because all possible subsequences of the data sequence and the query sequence must be compared.

Korean Patent Publication No. 10-2004-0095802

The present invention improves the accuracy of the search by normalizing and converting a data sequence constituting an index and a given query sequence, measuring similarity using Euclidean distance, and using an intersection of several correct candidate sets in a post-processing process. The present invention provides a subsequence matching system and a method of a time-series database that can significantly reduce performance and improve performance.

The problem to be solved by the present invention is not limited to the problem (s) mentioned above, and other object (s) not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above problems, the subsequence matching method of the time-series database provided in the present invention is a method for matching a query subsequence from an index generated for a data sequence, wherein the given query sequence is divided into a plurality of query windows. Determining a number of query windows corresponding to a query process from a query sequence among the divided query windows; Selecting a query window by calculating a query cost with respect to the determined number of query windows; Recalculating a similar tolerance range using the selected query window and a preset similarity tolerance value; And retrieving an answer candidate subsequence matching the query sequence from the index using a range query constructed based on the recalculated similar allowable range.

Advantageously, after the step of selecting the query window, further comprising normalizing the selected query window.

Preferably, the method further comprises the step of constructing a range query using the normalized query window and the recalculated similar tolerance after the step of recalculating the similar tolerance.

Advantageously, in the step of selecting the query window, the query cost is calculated by reflecting the query window similarity and the search range calculated by the query window based on the query cost function.

Preferably, after the searching of the answer candidate subsequence, the method further includes verifying the answer candidate subsequence using the query sequence.

Advantageously, verifying the answer candidate subsequence comprises: extracting a correct candidate set according to an intersection between the sets of answer candidate subsequences; And verifying the correct candidate set using the query sequence.

Preferably, verifying the correct candidate set comprises: calculating a Euclidean distance between the query sequence and a data sequence included in the correct candidate set, and when the calculated distance is equal to or less than the similar allowable range, the data sequence Is included in the correct candidate set of answer sequences to match the query sequence.

Preferably, after the step of verifying the answer candidate subsequence, it is characterized in that each step after the step of selecting the query window by determining whether it is the last query window among the selected query windows. .

Advantageously, the step of indexing the generated index comprises: splitting a data sequence into a plurality of data windows using a sliding window; Normalizing the data windows such that the scales of the data windows are the same; Reducing the dimension of the normalized data windows; Creating a record of the reduced data windows; Inserting the record into a multidimensional index structure; Its features are to include.

Advantageously, said multidimensional index structure comprises said dimensional reduced data windows as Minimum Bounding Rectangles (MBRs), wherein a minimum bounding rectangle of a higher level among the minimum bounding rectangles is a minimum of a plurality of lower levels. Its feature is that it is a hierarchical tree structure constructed by including bounding rectangles.

In addition, in order to solve the above problems, the subsequence matching system of the time-series database provided in the present invention is a system for matching a query subsequence from an indexing means for generating an index of a data sequence. A sequence divider for dividing the query window into a plurality of query windows; A window calculator for calculating an optimal number of query windows for query processing from the query sequence among the divided query windows; A window selector for selecting a query window having a minimum query cost with respect to the calculated number of query windows; A property retrieval unit for recalculating a similar tolerance range using the normalized query window and a preset similarity tolerance value; A query constructing unit configured to construct a range query using the normalized query window and the recalculated similar allowable range; And a query search unit for searching for an answer candidate subsequence to match the query sequence in the index using the configured range query.

Preferably, the method further includes a window normalization unit for normalizing the query window selected by the window selection unit.

Preferably, the query cost is characterized in that it is calculated by reflecting the query window similarity and the search range calculated by the query window based on the query cost function.

Preferably, after the query includes an extractor for extracting a correct candidate set according to the intersection between the sets of the answer candidate subsequences retrieved by the query search unit and a verification unit for verifying the correct candidate set using the query sequence. Its characteristics are that it further includes a processing unit.

Preferably, the verification unit calculates a Euclidean distance between the query sequence and the data sequence included in the correct candidate set, and if the calculated distance is equal to or less than the similar allowable range, matching the data sequence to the query sequence. The characteristic is that it is included in the set of correct candidates for the answer sequence to be answered.

According to the subsequence matching system and method of the time-series database of the present invention, the data sequence constituting the index and the given query sequence are normalized and converted, and the similarity is measured using Euclidean distance to improve the accuracy of the search and post-processing. In the process, the intersection of a set of correct answer candidates can be used to significantly reduce error and improve performance.

1 is a diagram schematically illustrating a configuration of a subsequence matching system according to an embodiment of the present invention.
FIG. 2 is a diagram schematically showing the configuration of the data sequence indexing means of FIG.
3 is a diagram schematically showing the configuration of the query sequence matching means of FIG.
4 is a diagram schematically illustrating a configuration of a query post-processing unit of FIG. 2.
5 is a flowchart illustrating an indexing process of generating an index with respect to a data sequence according to an embodiment of the present invention.
6 is a flowchart illustrating a matching process of deriving a result by performing a query from an index according to an embodiment of the present invention.
7A and 7B are views for explaining a sliding window and a disjoint window according to an embodiment of the present invention.

As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the drawings, similar reference numerals are used for similar elements.

Terms such as first, second, A, and B may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. The term and / or includes a combination of a plurality of related items or any item of a plurality of related items.

When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in between. Should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically showing the configuration of a subsequence matching system according to an embodiment of the present invention, FIG. 2 is a diagram schematically showing the configuration of the data sequence indexing means of FIG. 1, and FIG. 1 is a diagram schematically showing the configuration of the query sequence matching means of FIG. 1, and FIG. 4 is a diagram schematically showing the configuration of the query post-processing unit of FIG.

In the present invention, "sequence" means an array consisting of n entries, "time series data" means a sequence of entries having a real value measured at each time, and "time series database" means a database storing time series data. it means.

In addition, "data sequence" means time series data stored in a time series database, and "query sequence" means a sequence given by a user. "Subsequence" refers to a portion of a data sequence, and "subsequence matching" refers to a method of locating a position on a data sequence of subsequences similar to a query sequence. In this case, if the similarity (distance) between the two sequences is equal to or smaller than ε, which is the "allowed value" suggested by the user, it is defined as ε-match, which is defined as "similarity.

As shown in FIG. 1, the subsequence matching system according to the present invention comprises a data sequence indexing means 110, an index database 120, and a query sequence matching means 130. Here, the data sequence indexing means 110 may generate an index of a data sequence for retrieving the data sequence by dividing the data sequence into a data sequence window using a sliding window and normalizing and converting the data sequence. ). In this case, the sequence refers to an array composed of n entries, and the time series data may be a sequence of entries having a real value measured for each time.

The index database 120 may store and manage the data sequence and the index of the data sequence generated by the data sequence indexing means 110. In this case, the data sequence may be time series data. For example, the index database 120 may be a time series database that stores time series data.

The query sequence matching means 130 may search for and output an answer sequence matching the query sequence using an index of the data sequence stored in the index database 120. In this case, the query sequence may be a sequence input from the user, and the subsequence may be a part of the data sequence.

The query sequence matching means 130 may search for a subsequence similar to the query sequence.

For example, trend data of stocks may be input in a query sequence. In this case, the query sequence matching means 130 may search a data sequence having a trend similar to the trend data of the stock input as the query sequence among the data sequences which are the trend data of the conventional stock as the answer sequence. In addition, the user may predict a trend change of the stock based on the trend data of the stock output as the answer sequence.

In addition, when the similarity or distance between the query sequence and the answer sequence of the data sequence is equal to or less than ε, which is a user-set allowable value, the query sequence matching means 130 may determine that the subsequence is similar to the query sequence. Then, the matching between the query sequence and the data sequence including the subsequence searched by the query sequence matching means 130 may be defined as ε-matching. In addition, the query sequence matching means 130 may find the location of the corresponding subsequence on the data sequence.

In this case, the similarity is a unit for measuring the degree of similarity between the two sequences, and the value of the similarity may be represented by a distance, and the query sequence matching means 130 may use the Euclidean distance to determine the similarity. Can be calculated

For example, the query sequence matching means is a sequence of length n Wow The Euclidean distance of may be calculated using Equation 1 below.

[Equation 1]

Detailed configurations and operations of the data sequence indexing means and the query sequence matching means will be described in detail with reference to FIGS. 2 and 3.

As illustrated in FIG. 2, the data sequence indexing unit 110 may include a data sequence division unit 210, a data sequence window normalization unit 220, and an index generation unit 230.

The data sequence divider 210 may divide a data sequence into a plurality of data windows by using a sliding window.

For example, if a data sequence In this case, the data sequence divider 210 converts the data sequence into a data window of length w using a sliding window technique. Can be divided into

The data sequence window normalizer 220 may normalize the data windows so that the scales are the same while maintaining the characteristics of the data windows divided by the data sequence divider 210. At this time, the data sequence window normalization unit 220 is a data window Using normalization techniques on each of the elements of Can be converted to

For example, the data sequence window normalization unit 220 may normalize the data windows by a Z-Score technique defined as in Equation 2 below.

[Equation 2]

However, the normalization technique used by the data sequence window normalization unit 220 is not limited to the Z-Score technique and may normalize the data windows by using one of various normalization techniques.

The index generator 230 may generate an index of the data sequence by using the data windows normalized by the data sequence window normalizer 220.

More specifically, the index generator 230 may reduce the dimension of the normalized data windows. At this time, the index generator 230 reduces the dimension of the normalized data windows, so that when the dimension of the data indexed in the multi-dimensional index structure is more than a certain dimension, a curse of dimensionality occurs that rapidly decreases in performance. Can be avoided.

Next, the index generator 230 may generate a record of data windows whose dimensions are reduced. In this case, the index generator 230 may generate a record including at least one of a reduced data window, a data sequence identifier, and a window start position for each of the data windows.

Finally, the index generator 230 may insert a record of the data windows into the multidimensional index structure to generate an index of the data sequence. In this case, the index generator 230 may insert a record into the multidimensional index structure by setting the reduced data window as a key for each of the data windows.

In this case, the multi-dimensional index structure used by the index generator 230 configures the dimension-reduced data windows as minimum boundary rectangles (MBRs), and the minimum boundary rectangles of the upper level among the minimum boundary rectangles are plural. It may be a hierarchical tree structure generated by configuring to include the minimum boundary rectangles of the lower level of.

The data sequence indexing means 110 may repeat the operations until an index of all data sequences included in the index database 120 is generated.

In addition, as shown in FIG. 3, the query sequence matching means 130 includes a sequence divider 310, a window calculator 320, a window selector 330, a window normalizer 340, and an asset retrieval unit. 350, the query constructing unit 360, the query searching unit 370, and the query post-processing unit 380 may be configured.

The sequence dividing unit 310 divides a given query sequence into a plurality of query windows using a sliding window. For example, if the query sequence If, the sequence divider 310 uses a sliding window technique to window the length w used to index the query sequence. Can be divided into

The window calculator 320 calculates an optimal number of query windows for query processing from the query sequence among the divided query windows. Here, an optimal number of windows n for query processing is calculated from a given query sequence using Equation 3 below.

[Equation 3]

The window selector 330 selects a query window having a minimum query cost with respect to the calculated number of query windows.

More specifically, for the calculated number of query windows, n windows are selected by predicting optimal performance based on a query cost function. In this case, the query cost function calculates the query cost by the following Equation 4 in consideration of the similarity of the window and the search range calculated by the window.

[Equation 4]

The window selector 330 may select a query window including the feature of the query sequence most among the query windows divided by the sequence divider 310.

In this case, the window selector 330 may calculate a standard deviation of each of the query windows divided by the sequence divider 310. The window selector 330 may select n query windows that are disjoint from each other in the order of the calculated standard deviation. In this case, n may be a positive integer defined by a user.

In addition, the window selector 330 may select the window in the order of the magnitude of the standard deviation, thereby selecting the window that includes the most feature of the query sequence to reduce the error answer of the search result.

The window normalizer 340 normalizes the query window selected by the window selector 330.

More specifically, the window normalization unit 340 may normalize the query window selected by the window selector 330. The window normalization unit 340 may have n query windows selected by the window selector 330. Query window using the same normalization technique as the window normalization unit 330 Can be converted to

The property retrieval unit 350 normalizes the selected query window by a window normalization unit and uses a similar similarity value to be preset. To be recalculated using Equation 5 below. Similar tolerance May be a range of similarity values that may be determined to be similar to an answer window to be matched with a query window.

[Equation 5]

The query constructing unit 360 includes the normalized query window and the recalculated similar tolerance range. You will construct a range query using.

Here, the query constructing unit 360 may reduce the dimension of the n multidimensional query windows, and generate a range query using the n reduced query windows and similar allowable ranges.

In this case, the range query may be a query method that searches for a data window within the similar tolerance range recalculated from the query window by using an index, using the similar tolerance range recalculated using the selected query window and the user-defined similarity tolerance. . For example, the range query generated by the query configuration unit 360 may be in the form of <selected query window, recalculated similar tolerance> generated by matching the selected query window with the recalculated similar tolerance in pairs. . In addition, the range query generated by the query configuration unit 360 may be used as an input variable for searching through the index.

The query retrieval unit 370 retrieves an answer candidate subsequence to match the query sequence from the index by using the range query configured in the query constructer 360.

Here, the query retrieval unit 370 may search for an answer candidate subsequence to match the query sequence in the index using a normalized query window and a similar allowable range, and generate a set of retrieved answer candidate subsequences. In this case, the query retrieval unit 370 may repeat the above process until all of the set of answer candidate subsequences for the n windows selected by the window selection unit 330 are generated.

The query postprocessor 380 may include an extractor 381 and a verifyer 382 as shown in FIG. 4.

The extractor 381 extracts a correct answer candidate set according to an intersection between sets of the answer candidate subsequences searched by the query search unit 370. The verification unit 382 may include a verification unit that verifies the set of correct answer candidates using the query sequence. The verification unit 382 calculates a Euclidean distance between the query sequence and the data sequence included in the correct candidate set, and if the calculated distance is equal to or less than the similar tolerance, match the data sequence to the query sequence. Include in the correct candidate set of answer sequences.

7A and 7B illustrate a sliding window and a disjoint window according to an embodiment of the present invention. As shown in FIGS. 7A and 7B, a window is a unit for dividing a sequence, and may be divided into a sliding window and a disjoint window according to a splitting method. The sliding window may be a window configured with all possible positions of the sequence as a start position, and the disjoint window may be a window configured with a start position of a position that is a multiple of the window length in the sequence. 7A shows an example of dividing a sequence 710 into a sliding window 720 having a length of four. "Disjoint window" means a window configured with a start position of a multiple of the window length in a sequence, and FIG. 7B shows an example of a disjoint window 730 having a length of 4. FIG.

Indexing process

5 is a flowchart illustrating an indexing process of generating an index with respect to a data sequence according to an embodiment of the present invention.

As shown in FIG. 5, the index generator 230 first generates and initializes a multidimensional index structure (S510). In this case, the index generator 230 configures the reduced-dimensional data windows as the minimum boundary rectangles (MBRs), and the minimum boundary rectangles of the upper level among the minimum boundary rectangles include the minimum boundary rectangles of the plurality of lower levels. To create a multidimensional index structure.

In operation S520, the data sequence is divided into a plurality of data windows by using the sliding window.

Subsequently, normalizing the data windows such that the scales of the data windows are the same (S530) is performed. That is, the data windows can be normalized so that the scales are the same while maintaining the characteristics of the divided data windows.

Next, the index generator 230 may reduce the dimension of the normalized data windows (S540), and generate a record of the reduced data windows (S550).

The index generator 230 may insert a record of the data windows into the multidimensional index structure generated and initialized in operation S510 to generate an index of the data sequence. In this case, the index generator 230 inserts a record into the multidimensional index structure by setting a data window having a reduced dimension as a key for each of the data windows (S560).

Finally, the index generator 230 may check whether all the data sequences included in the index database 120 are indexed. When the indexes of all data sequences included in the index database 120 are generated, the index generator 230 may terminate the index generation method (S570).

In addition, when there is a non-indexed data sequence among the data sequences included in the index database 120, the index generator 230 repeatedly selects one of the non-indexed data sequences S520. Can be done.

Matching process

5 is a flowchart illustrating a matching process for deriving a result by performing a query from an index according to an embodiment of the present invention. Here, the same description will be omitted with reference to the description of FIGS. 3 and 4.

As shown in FIG. 5, first, a step (S610) of dividing a given query sequence into a plurality of query windows using a sliding window is performed. Here, the sequence dividing unit 310 divides a given query sequence into a plurality of query windows using a sliding window. For example, if the query sequence If, the sequence divider 310 uses a sliding window technique to window the length w used to index the query sequence. Can be divided into

Subsequently, a step (S620) of calculating the number of query windows for query processing is performed from the query sequence among the divided query windows. Here, the window calculator 320 calculates an optimal number of query windows for query processing from the query sequence among the divided query windows.

In operation S630, a query window having a query cost is selected based on the calculated number of query windows. More specifically, the window selector 330 selects a query window having a minimum query cost with respect to the calculated number of query windows. Here, for the calculated number of query windows, n windows are selected by predicting optimal performance based on a query cost function. In this case, the query cost function calculates the query cost in consideration of the similarity of the window and the search range calculated by the window.

Then, normalizing the selected query window (S640) is performed.

More specifically, the window normalization unit 340 may normalize the query window selected by the window selector 330. The window normalization unit 340 may have n query windows selected by the window selector 330. Query window using the same normalization technique as the window normalization unit 330 Can be converted to

Subsequently, in operation S650, a similar allowable range is recalculated using the normalized query window and a preset similar allowable value. Here, the property retrieval unit 350 normalizes the selected query window by a window normalization unit and uses a similar similarity allowance value to be preset. Will be recalculated. Similar tolerance May be a range of similarity values that may be determined to be similar to an answer window to be matched with a query window.

A step 660 of constructing a range query using the normalized query window and the recalculated similar allowable range is performed. Here, the query constructing unit 360 may reduce the dimension of the n multidimensional query windows, and generate a range query using the n reduced query windows and similar allowable ranges.

In this case, the range query may be a query method that searches for a data window within the similar tolerance range recalculated from the query window by using an index, using the similar tolerance range recalculated using the selected query window and the user-defined similarity tolerance. . For example, the range query generated by the query configuration unit 360 may be in the form of <selected query window, recalculated similar tolerance> generated by matching the selected query window with the recalculated similar tolerance in pairs. . In addition, the range query generated by the query configuration unit 360 may be used as an input variable for searching through the index.

Next, a step 670 of searching for an answer candidate subsequence to match the query sequence in the index using the configured range query is performed.

The query retrieval unit 370 retrieves an answer candidate subsequence to match the query sequence from the index by using the range query configured in the query constructer 360.

Here, the query retrieval unit 370 may search for an answer candidate subsequence to match the query sequence in the index using a normalized query window and a similar allowable range, and generate a set of retrieved answer candidate subsequences. In this case, the query retrieval unit 370 may repeat the above process until all of the set of answer candidate subsequences for the n windows selected by the window selection unit 330 are generated.

Subsequently, a step 680 of verifying the answer candidate subsequence using the query sequence is performed.

More specifically, the extractor 381 extracts a correct answer candidate set according to an intersection between sets of the answer candidate subsequences searched by the query search unit 370. The verification unit 382 may include a verification unit that verifies the set of correct answer candidates using the query sequence. The verification unit 382 calculates a Euclidean distance between the query sequence and the data sequence included in the correct candidate set, and if the calculated distance is equal to or less than the similar tolerance, match the data sequence to the query sequence. Include in the correct candidate set of answer sequences.

On the other hand, if it is determined whether the last query window among the selected query windows (S690), if the remaining n windows remain, it may proceed to each step after the step of repeatedly selecting the query window.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

110 --- Data sequence indexing means 120 --- Index database
130 --- query sequence matching means 210 --- sequence divider
220 --- window calculator 230 --- window selector
240 --- Window Normalization 250 --- Property Retrieval
260 --- Query Constructor 270 --- Query Finder
280 --- Query Post-Processing Unit 310 --- Extraction Unit
320 --- Verification unit 610 --- Sequence
620 --- sliding window 630 --- disjoint window

Claims (15)

1. A method of matching a query subsequence from an index generated over a data sequence, the method comprising:
Dividing a given query sequence into a plurality of query windows, and determining the number of query windows corresponding to the query processing from the query sequence among the divided query windows;
Selecting a query window by calculating a query cost with respect to the determined number of query windows;
Recalculating a similar tolerance range using the selected query window and a preset similarity tolerance value; And
Retrieving an answer candidate subsequence matching a query sequence from the index using a range query constructed based on the recalculated likelihood range;
Subsequence matching method of a time series database comprising a.
The method of claim 1,
After selecting the query window,
And subnormalizing the selected query window.
The method of claim 1,
After recalculating the similar tolerance range,
And constructing a range query using the normalized query window and the recalculated similar allowable ranges.
The method of claim 1,
In the step of selecting the query window,
The query cost is calculated by reflecting the query window similarity and the search range calculated by the query window based on the query cost function.
The method of claim 1,
After retrieving the answer candidate subsequence,
Verifying the answer candidate subsequence using the query sequence.
The method of claim 5,
Verifying the answer candidate subsequence,
Extracting a correct candidate set according to an intersection between the sets of answer candidate subsequences; And
Verifying the set of correct answer candidates using the query sequence; and subsequence matching method of a time series database.
The method of claim 6,
Verifying the correct candidate set,
A Euclidean distance is calculated between the query sequence and the data sequence included in the set of correct candidates, and when the calculated distance is less than or equal to the similar allowable range, the data sequence is included in the correct candidate set of the answer sequence to match the query sequence. Subsequence matching method of a time series database comprising a.
The method of claim 1,
After verifying the answer candidate subsequence,
Determining whether the query window is the last query window among the selected query windows, and repeatedly performing each step after selecting the query window.
The method of claim 1,
The indexing step for the generated index,
Dividing a data sequence into a plurality of data windows using a sliding window;
Normalizing the data windows such that the scales of the data windows are the same;
Reducing the dimension of the normalized data windows;
Creating a record of the reduced data windows; And
Inserting the record into a multidimensional index structure;
Sequence matching method of a time series database comprising a.
The multidimensional index structure,
A layer created by configuring the dimensionally reduced data windows as Minimum Bounding Rectangles (MBR) and configuring a minimum boundary rectangle of a higher level among the minimum boundary rectangles to include a plurality of minimum boundary rectangles of a lower level. A sequence matching method of a time series database, characterized in that the tree structure.
A system for matching a query subsequence from indexing means for generating an index of a data sequence,
A sequence divider for dividing a given query sequence into a plurality of query windows using a sliding window;
A window calculator for calculating an optimal number of query windows for query processing from the query sequence among the divided query windows;
A window selector for selecting a query window having a minimum query cost with respect to the calculated number of query windows;
A property retrieval unit for normalizing the selected query window by a window normalization unit and recalculating a similar tolerance range using a preset similarity tolerance value;
A query constructing unit configured to construct a range query using the normalized query window and the recalculated similar allowable range; And
A query retrieval unit for retrieving an answer candidate subsequence to match a query sequence from an index using the constructed range query;
And a subsequence matching system of a time series database.
The method of claim 11,
And a window normalization unit for normalizing a query window selected by the window selection unit.
The method of claim 11,
The window selector,
The query cost is calculated based on the query cost function to reflect the query window similarity and the search range calculated by the query window.
The method of claim 11,
And a query post-processing unit including an extracting unit for extracting a correct candidate set according to an intersection between the sets of answer candidate subsequences retrieved by the query searching unit, and a verifying unit for verifying the correct candidate set using the query sequence. And a subsequence matching system of a time series database.
The method of claim 14,
The verification unit,
A Euclidean distance is calculated between the query sequence and the data sequence included in the set of correct candidates, and when the calculated distance is less than or equal to the similar allowable range, the data sequence is included in the correct candidate set of the answer sequence to match the query sequence. And a subsequence matching system of a time series database.