JP2007072752A - Similar time series data calculation method, similar time series data calculation device, and similar time series data calculation program - Google Patents

Abstract

An object of the present invention is to prevent a search omission in determination of similarity of time series data and prevent an excessive search. It is also possible to deal with time-series data of queries having an arbitrary length.
In the distance calculation unit 4, it is determined that the lower limit value of the distance is equal to or less than the threshold value to prevent the occurrence of search omission, and the similarity is determined to be similar when the upper limit value of the distance is equal to or less than the threshold value. Prevent over search. Further, the segment management unit 32 in the dimension compression unit 3 divides the time series data into a plurality of segments having exponentially different lengths, and the dimension compression coefficient calculation unit 33 obtains an average value and a standard deviation in each segment. Thus, it is possible to deal with time-series data of queries of arbitrary length.
[Selection] Figure 1

Description

  The present invention relates to a method, an apparatus, and a program for obtaining similar data among a plurality of time series data.

  Processing for obtaining similar time-series data at high speed is used in various fields. For example, a stock online trading system monitors a large amount of stock prices and searches for companies with similar price movements at high speed. The moving body position management system senses a large number of traveling vehicles, and detects those having a similar movement track at high speed. The seismic monitoring system finds a point where the way of shaking at the time of an earthquake resembles at high speed based on information from a large number of seismometers.

The present application deals with a problem of searching for a combination of time series data similar to a query having a predetermined length from a search time among a plurality of time series data. The problem is that the number of data points in the time series data is n, the length of the query is l, the threshold is ε, the time series data X = (x ₁ , x ₂ ,..., X _n ) and the time series data Y = If the Euclidean distance of (y ₁ , y ₂ ,..., Y _n ) is D (X, Y), it can be expressed as a set of combinations of time series data satisfying the following conditions.

  In general, high calculation cost is required to obtain D (X, Y). This is because time series data has a large number of data points. For this reason, a dimension compression method has been used when calculating the distance of time series data. Typical dimensional compression methods include discrete Fourier transform (see Non-Patent Document 1), discrete weblet transform (see Non-Patent Document 2), singular value decomposition (see Non-Patent Document 3), and the like. These dimension compression methods have a feature that there is no search omission (determining that similar time-series data is not similar). This search-free property is guaranteed by lower bounding lemma. The lower bounding lemma means that when L (X ′, Y ′) ≦ D (X, Y) holds when the time-series data after dimension compression is L (X ′, Y ′), no search leakage occurs. This is the lemma. That is, if the distance after dimensional compression is the lower limit value of the distance before dimensional compression, no search omission occurs.

The time series data handled here includes discrete values such as stock prices and continuous values such as movement trajectories. If the time-series data is a discrete value, whether the time-series data is similar may be determined by checking whether the distance between the time-series data is within the threshold using the discrete values. Further, when the time series data is a continuous value, the time series data is sampled at the time of processing and becomes a discrete value, so that a discrete value processing method is eventually used.
R. Agrawl, C. Faloutsos, and ANSwami. Efficient Similarity Search In Sequence Databases. In Proc. FODO, 1993 K. Chan, AWFu. Efficient Time Series Matching by Wavelets. In Proc. ICDE, 1999 F. Korn, HVJagadish, C. Faloutsos. Efficient Supporting Ad Hoc Queries in Large Datasets of Time Sequences. In Proc. SIGMOD, 1997

  The conventional dimension compression method has problems in the following two points.

  The first is that it has only the property of no search omission. Considering the application of the similar search of time series data, it is necessary to have the property that an excessive search (determining a dissimilar sequence as similar) does not occur.

  The second is that it can only deal with time-series data of queries of a specific length. Considering the application of time series data similarity search, it is necessary to be able to deal with query time series data of any length.

  The present invention has been made in view of the above, and the object of the present invention is firstly to prevent search omission and excessive search. Secondly, it is possible to deal with time-series data of queries having an arbitrary length.

  In the similar time series data calculation method according to the first aspect of the present invention, a pair of time series data to be subjected to similarity determination is read from a memory by a dimension compression coefficient calculation means, and an average value and a standard deviation are respectively obtained as dimension compression coefficients. The average value and the standard deviation are read from the memory by the step of calculating and storing in the memory and the upper limit / lower limit calculation means, and using these, the upper and lower limits of the distance of the time series data are obtained and stored in the memory. And a step of reading the upper limit value, the lower limit value, and the threshold value from the memory by the similarity determination means, and determining that the pair of time series data is similar when the lower limit value is equal to or less than the threshold value and the upper limit value is equal to or less than the threshold value. It is characterized by having.

  In the present invention, by determining that the lower limit value of the distance is equal to or less than the threshold value, it is ensured that there is no search omission according to the lower bounding lemma lemma, and the upper limit value of the distance is equal to or less than the threshold value. Judging by the upper bounding lemma lemma, we guarantee that there is no excessive search.

  In the similar time series data calculation method, when the similarity determination unit determines that the lower limit value is less than or equal to the threshold value and the upper limit value is greater than the threshold value, the exact distance calculation unit calculates the pair of times before dimension compression. The sequence data is read from the memory, the distance between the two is obtained and stored in the memory, and the similarity determination means reads the distance and threshold of the pair of time series data before dimension compression from the memory, and the distance is less than the threshold. Determining that the pair of time-series data is similar to each other.

  In the present invention, even if it is determined to be similar in relation to the lower limit value, if it is determined to be dissimilar in relation to the upper limit value, it is accurate from the time-series data after dimension compression. It is assumed that the similarity cannot be determined, the distance between the two is obtained using a pair of time-series data before dimension compression, and the distance is compared with a threshold value, thereby making a more exact similarity determination.

  The similar time series data calculation method includes a step of segmenting time series data into a plurality of segments having exponentially different lengths by a segment management means, and an average for each segment of time series data by the dimension compression coefficient calculation means. Obtaining a value and a standard deviation.

  In the present invention, the time-series data is divided into a plurality of segments having exponentially different lengths, and the average value and the standard deviation in each segment are obtained, whereby the time-series data of a query of an arbitrary length is obtained. Is also available.

  The similar time-series data calculation method includes a step of counting the number of segments having a specific length by the segment counting means, and a segment having the length when the counted number exceeds a predetermined value by the segment integration means. And a step of obtaining an average value and a standard deviation in the segment after the integration by the dimension compression coefficient calculation means.

  In the present invention, when the number of segments of a specific length counted is larger than a certain value, two segments of that length are integrated, and an average value and a standard deviation in the segment after integration are obtained, The feature quantity of dimension compression can be updated.

  The similar time series data calculation method includes a step of segmenting time series data into a plurality of segments of equal length by segment management means, and an average value in each segment of time series data by the dimension compression coefficient calculation means. And obtaining a standard deviation.

  In the present invention, the time series data is divided into a plurality of segments of equal length, and the average value and standard deviation in each segment are obtained, thereby supporting the time series data of a specific length query. Make it possible.

  The similar time-series data calculation apparatus according to the second aspect of the present invention reads a pair of time-series data to be subjected to similarity determination from a memory, obtains an average value and a standard deviation as a dimension compression coefficient for each, and stores them in the memory. Dimensional compression coefficient calculation means, upper limit / lower limit calculation means for reading the average value and standard deviation from the memory, and using these to obtain the upper and lower limit values of the distance of the time series data and storing them in the memory, and the upper limit value And a similarity determination unit that reads the lower limit value and the threshold value from the memory, and determines that the pair of time series data is similar when the lower limit value is equal to or lower than the threshold value and the upper limit value is equal to or lower than the threshold value. .

  A similar time-series data calculation program according to the third aspect of the present invention reads a pair of time-series data to be subjected to similarity determination from a memory, obtains an average value and a standard deviation as a dimension compression coefficient for each, and stores them in the memory. Processing, reading the average value and standard deviation from the memory, using these to obtain the upper limit value and lower limit value of the distance of the time series data, and storing the upper limit value, lower limit value, threshold value from the memory, When the lower limit value is equal to or less than the threshold value and the upper limit value is equal to or less than the threshold value, the computer is caused to execute a process of determining that the pair of time series data is similar.

  According to the present invention, it is possible to prevent occurrence of search omission and excessive search by performing similarity determination using the upper limit value and the lower limit value of the distance. Further, by dividing the time series data into segments of exponential length, it is possible to deal with time series data of an arbitrary length query.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  As shown in the block diagram of FIG. 1, the similar time-series data calculation apparatus 1 according to the present embodiment has a configuration including a data reception unit 2, a dimension compression unit 3, a distance calculation unit 4, and a calculation result transmission unit 5. The similar time-series data calculation device 1 is a computer including hardware such as an arithmetic processing device, a memory, and a storage device, and the processing of each unit is executed by a program that operates in cooperation with these hardware. The processing of each part will be described from the overview.

  The data receiving unit 2 receives from the outside two pieces of time-series data to be judged whether they are similar and the query length of each piece of time-series data, and stores them in the memory. The dimension compressing unit 3 reads each time series data from the memory, compresses the dimension, and stores it in the memory. The distance calculation unit 4 reads each time-series data after dimension compression from the memory, calculates the distance between the time-series data, and determines the similarity between the two based on the distance. If this determination is difficult, each time-series data before dimension compression is read from the memory, a distance between the time-series data is calculated, and similarity is determined based on this distance. The calculation result transmission unit 5 transmits the result of similarity determination to the outside.

  As shown in the block diagram of FIG. 2, the dimension compression unit 3 includes a data reception unit 31, a segment management unit 32, a dimension compression coefficient calculation unit 33, and a calculation result transmission unit 34.

  The data receiving unit 31 receives time-series data before dimension compression. The segment management unit 32 manages the length of each segment, and divides the time series data into a plurality of segments having exponentially different lengths or a plurality of segments having an equal length. The dimension compression coefficient calculation unit 33 performs dimension compression by obtaining an average value and a standard deviation in each segment of the time series data as a dimension compression coefficient. The calculation result transmission unit transmits the time series data after dimension compression to the distance calculation unit 4.

  As shown in the block diagram of FIG. 3, the segment management unit 32 includes a data reception unit 321, a segment counting unit 322, a segment integration unit 323, and a data transmission unit 324. The data receiving unit 321 receives time-series data before dimension compression. The segment counting unit 322 counts the number of segments having a specific length. When the counted number is larger than a certain value, the segment integration unit 323 integrates two segments having the length. The data transmission unit 324 transmits the combined segment combination to the dimension compression coefficient calculation unit 33.

  As shown in the block diagram of FIG. 4, the distance calculation unit 4 includes a data reception unit 41, a use segment determination unit 42, a distance coefficient calculation unit 43, an upper limit value calculation unit 44, a lower limit value calculation unit 45, and an similarity determination unit 46. , A strict distance calculation unit 47 and a determination result transmission unit 48.

  The data receiving unit 41 receives time-series data before dimension compression, query length, and time-series data after dimension compression (that is, dimension compression count). The use segment determination unit 42 determines a segment used when calculating the distance after dimension compression. The distance coefficient calculation unit 43 calculates a distance coefficient between time series data using a dimensional compression coefficient. The upper limit calculator 44 calculates the upper limit of the distance using the distance coefficient and stores it in the memory. The lower limit calculator 45 obtains the lower limit value of the distance using the distance coefficient and stores it in the memory. The similarity determination unit 46 reads the upper limit value, the lower limit value, and the threshold value from the memory, and determines that the pair of time series data is similar when the lower limit value is equal to or less than the threshold value and the upper limit value is equal to or less than the threshold value.

  When the similarity determination unit 46 determines that the lower limit value is less than or equal to the threshold value but the upper limit value is greater than the threshold value, the exact distance calculation unit 47 reads a pair of time series data before dimension compression from the memory, The distance between the time series data is obtained and stored in the memory. In this case, the similarity determination unit 46 reads the distance and the threshold value from the memory, and determines that the time series data is similar when the distance is equal to or less than the threshold value. Thus, an accurate determination is made by using a strict distance. The determination result transmission unit 48 transmits the result of similarity determination to the calculation result transmission unit 5.

  The upper bounding lemma guarantees that no excessive search occurs even if the time series data is dimensionally compressed by the dimension compression unit 3. Upper bounding lemma means that excessive search does not occur if D (X, Y) ≦ U (X ′, Y ′) holds when the distance of the sequence after dimension compression is U (X ′, Y ′). Is an auxiliary theorem. That is, if the distance after dimensional compression is the upper limit value of the distance before dimensional compression, excessive search will not occur.

  Moreover, it is guaranteed by lower bounding lemma that no search omission occurs. The lower bounding lemma means that when L (X ′, Y ′) ≦ D (X, Y) holds when the time-series data after dimension compression is L (X ′, Y ′), no search leakage occurs. This is the lemma. That is, if the distance after dimensional compression is the lower limit value of the distance before dimensional compression, no search omission occurs.

  Next, processing of each unit in the similar time series data calculation apparatus 1 will be described in detail.

First, in the dimension compression unit 3, when the segment management unit 32 divides the sequence X (n data points) into N specific length segments and sets the i-th segment as σ _i , the dimension compression coefficient calculation is performed. In section 33, the length of σ _i is l (σ _i ), the average is av (σ _i ), the standard deviation is sd (σ _i ), and the sequence X ′ after dimension compression is a tuple of three coefficients as follows: It expresses with.

Here, the following equation holds for l (σ _i ).

If x _{i: j} is the j-th data point of σ _i , the dimension compression coefficient calculation unit 33 calculates av (σ _i ) and sd (σ _i ) as follows.

Next, processing in the distance calculation unit 4 will be described with reference to the flowchart in FIG. In this process, a segment used for distance calculation after dimension compression differs depending on whether a lower limit value or an upper limit value is calculated for a query having an arbitrary length. That is, in step 1 (shown as “S1” in FIG. 5; the same applies to the following), as shown in FIG. 6, the use segment determination unit 42 determines the segment in the longest sequence among sequences shorter than the query when obtaining the lower limit value. Is determined as a utilization segment, and when the upper limit value is obtained, the segment in the shortest sequence among the sequences longer than the query is determined as the utilization segment. Each sequence is configured by aggregating a plurality of segments. Assuming that the segment with the smallest number among the segments used for the calculation of the lower limit value is N _l and the segment with the smallest number among the segments used for the calculation of the upper limit value are N _u , N _l and N _u can be expressed as The largest and smallest segments to satisfy.

In step 2, the distance coefficient calculation unit 43 calculates distance coefficients D1, D2, and D3 by the following equations.

In step 3, the lower limit calculator 45 calculates the lower limit L (X ′, Y ′) of the distance by the following formula.

Here, L (X ′, Y ′) can be rewritten as follows, and by calculating D1, D2, and D3 in advance, they can be used for calculating L (X ′, Y ′).

  In step 4, the similarity determination unit 46 compares the lower limit value with the threshold value, and if the lower limit value is larger than the threshold value, the time series data is determined not to be similar, and the process is terminated (step 5). On the other hand, if the lower limit value is less than or equal to the threshold value, the process proceeds to step 6.

In step 6, the upper limit calculation unit 44 calculates the upper limit value U (X ′, Y ′) of the distance by the following equation.

Here, since N _u ≦ N _l, U (X ′, Y ′) can be rewritten as follows, and by calculating D1, D2, D3 in advance, they can be rewritten as U (X ′, Y ′). ).

  In step 7, the similarity determination unit 46 compares the upper limit value with the threshold value. If the upper limit value is less than or equal to the threshold value, the time series data is determined to be similar, and the process ends (step 8). On the other hand, when the upper limit value is larger than the threshold value, it is determined that the time series data is not similar, and the process proceeds to step 9.

  In step 9, the exact distance calculation unit 47 calculates the Euclidean distance between the two using time-series data before dimension compression, and sends the distance to the similarity determination unit 46.

  In step 10, the similarity determination unit 46 compares the distance calculated by the strict distance calculation unit 47 with a threshold value, and determines that the time series data is similar if the distance is equal to or less than the threshold value (step 11). Otherwise, it is determined that they are dissimilar and the process is terminated (step 12).

  Next, by using the lower limit value L (X ′, Y ′) and the upper limit value U (X ′, Y ′) obtained as described above, L (X ′, Y ′) ≦ D (X, Y) It proves that ≦ U (X ′, Y ′). First, it is shown that L (X ′, Y ′) ≦ D (X, Y).

Since D (X, Y) is an Euclidean distance, the following is clearly true if k ≧ l.

Assume that the following equation holds for k.

Where Δx _{i: j} = av (σ _i ) −x _{i: j}

here

Because

It becomes. Here, when Δx _{i: j} is regarded as a vector, ‖Δx _{i: j} ‖ is the size of the vector Δx _{i: j} . Then, from the inner product definition formula

It becomes. Here, θ _i is an angle formed by the vector Δx _{i: j} and the vector Δy _{i: j} . Since cosθ _i ≦ 1,

It becomes. Similarly, if k ≦ 1 and cos θ _i ≧ 1, it can be shown that D (X, Y) ≦ U (X ′, Y ′).

Next, processing for managing segments having exponentially different lengths in the segment management unit 32 will be described in detail. The segment management unit 32 sets the length of the segment of level h to 2 ^h, and increases the level h according to the processing time so that the length of the segment increases exponentially as shown in the schematic diagram of FIG. Perform segmentation. This is because it corresponds to a query having an arbitrary length. As a result, whether the query is short or long, the relative error when the lower limit value and the upper limit value are calculated is the same.

  When the number of segments at each level becomes larger than the capacity c, the segment integration unit 323 integrates the two segments at that level and incrementally updates the dimension compression feature quantity.

  For example, specific update processing when c = 2 will be described with reference to the schematic diagram of FIG. 8 and the flowchart of FIG.

  In step 21 (shown as “S21” in FIG. 9; the same applies hereinafter), the segment counting unit 322 sets the level h to the initial value 0. In step 22, the number of segments at level h = 0 is counted. Here, the number of segments is 2 at level h = 0, 2 at h = 1, and 1 at h = 2. Here, when sequence data flows in, there are three segments with h = 0. Since the number of segments is larger than c (= 2) (step 24), in step 25, the segment integration unit 323 integrates the segments and creates a segment with h = 1. Then, since there are three segments of level h = 1, the level h is incremented by 1 in step 26 and the same processing is performed. This process is repeated until the level h reaches the maximum level of the time series data (step 23). As a result, the number of segments after the update is 1 at level h = 0, 1 at level h = 1, and 2 at level h = 2.

The dimension compression feature quantity when the segments σ _i and σ _{i + 1} are integrated into a new segment σ ′ _i is updated as follows.

  Next, processing for managing segments of equal length in the segment management unit 32 will be described in detail. Here, it is assumed that similar time series data is searched for a query of a fixed length, not a query of an arbitrary length. In this case, the first segment is updated sequentially, and the last segment is updated for each capacity c. The dimension compression coefficient calculation unit 33 calculates the average value and standard deviation of the time series data in each segment of equal length collected for each value of the capacity c, and stores the average value and the standard deviation in the memory.

  A specific update process when c = 2 will be described with reference to the schematic diagram of FIG. There are four dimension compression feature quantities, and the segment length is 4 for the first, 4 for the second, 4 for the third, 4 for the fourth, and 0 before the update (t = 1). Here, when the sequence data flows in (t = 2), the length of the segment is 4 for the first, 4 for the second, 4 for the third, and 1 for the fourth. When the next sequence data flows in (t = 3), the length of the segment is 2 for the first, 4 for the second, 4 for the third, and 2 for the fourth.

Segment is sequentially updated or per capacity c, but the coefficient of dimensional compression at this time and length l _i + l i _{+ 1} segment sigma _'i, a segment of length l _i and sigma _i, the length When the segment of l _{i + 1} is σ _{i + 1} , it is updated using the following formula.

  As described above, according to the present embodiment, the distance calculation unit 4 determines that the lower limit value of the distance is similar to the threshold value or less so that there is no search omission due to the lower bounding lemma theorem. And guaranteeing the property that there is no excessive search by the upper bounding lemma theorem by determining that the upper limit of the distance is similar to the threshold value. Thereby, the occurrence of search omission and the occurrence of excessive search can be prevented.

  According to the present embodiment, the segment management unit 32 divides the time series data into a plurality of segments having exponentially different lengths, and the dimension compression coefficient calculation unit 33 obtains an average value and a standard deviation in each segment. Thus, it is possible to deal with time-series data of queries having an arbitrary length.

  According to the present embodiment, even if it is determined that the similarity determination unit 46 is similar in relation to the lower limit value, if it is determined to be dissimilar in relation to the upper limit value, after dimension compression Therefore, the exact distance calculation unit 47 obtains the Euclidean distance between the pair of time series data before the dimension compression, and the similarity determination unit 46 again. Thus, by comparing the Euclidean distance with the threshold value, it is possible to make a stricter similarity determination.

  According to the present embodiment, when the number of segments of a specific length counted by the segment counting unit 322 becomes larger than a certain value, the segment integration unit 323 integrates two segments of that length, By obtaining the average value and the standard deviation in the segment after integration by the dimension compression coefficient calculation unit 33, it is possible to update the feature quantity of dimension compression.

  According to the present embodiment, the segment management unit 32 divides the time series data into a plurality of segments of equal length, and the dimension compression coefficient calculation unit 33 obtains the average value and standard deviation in each segment. Thus, it is possible to deal with time-series data of a query having a specific length.

  According to the present embodiment, by dividing the time series data into segments of exponential or equal length, it is possible to determine the similarity of the time series data with a relatively small error.

It is a block diagram which shows the structure of the similar time series data calculation apparatus in one embodiment. It is a block diagram which shows the structure of the dimension compression part in the said similar time series data calculation apparatus. It is a block diagram which shows the structure of the segment management part in the said dimension compression part. It is a block diagram which shows the structure of the distance calculation part in the said similar time series data calculation apparatus. It is a flowchart which shows the process in a distance calculation part. It is a schematic diagram for demonstrating the segment used for calculation of the lower limit, and the segment used for calculation of an upper limit. It is a schematic diagram for demonstrating that a relative error does not change with respect to a query of an arbitrary length. It is a schematic diagram for demonstrating the process at the time of updating a feature-value. It is a flowchart which shows the update process of the feature-value. It is a schematic diagram for demonstrating the process at the time of updating the feature-value in case a query is fixed length.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Similar time series data calculation apparatus 2 ... Data reception part 3 ... Dimension compression part 4 ... Distance calculation part 5 ... Calculation result transmission part 31 ... Data reception part 32 ... Segment management part 33 ... Dimension compression coefficient calculation part 34 ... Calculation result Transmission unit 41 ... Data reception unit 42 ... Use segment determination unit 43 ... Distance coefficient calculation unit 44 ... Upper limit value calculation unit 45 ... Lower limit value calculation unit 46 ... Similarity determination unit 47 ... Strict distance calculation unit 48 ... Determination result transmission unit 321 ... Data reception unit 322 ... Segment counting unit 323 ... Segment integration unit 324 ... Data transmission unit

Claims (7)

A step of reading out a pair of time series data subject to similarity determination from the memory by the dimension compression coefficient calculating means, obtaining an average value and a standard deviation as a dimension compression coefficient for each, and storing them in the memory;
The average value and the standard deviation are read from the memory by the upper limit / lower limit calculation means, and the upper limit value and the lower limit value of the distance of the time series data are obtained and stored in the memory using these,
A step of reading the upper limit value, the lower limit value, and the threshold value from the memory by the similarity determination means, and determining that the pair of time series data is similar when the lower limit value is equal to or less than the threshold value and the upper limit value is equal to or less than the threshold value;
A similar time series data calculation method characterized by comprising: When the similarity determination means determines that the lower limit value is less than or equal to the threshold value and the upper limit value is greater than the threshold value, the exact distance calculation means reads a pair of time series data before dimension compression from the memory, Determining the distance and storing it in memory;
A step of reading the distance and threshold of the pair of time series data before dimension compression from the memory by the similarity determination means, and determining that the pair of time series data is similar when the distance is equal to or less than the threshold;
The similar time series data calculation method according to claim 1, wherein: Dividing the time-series data into a plurality of segments exponentially different in length by the segment management means;
Obtaining a mean value and a standard deviation in each segment of the time series data by the dimension compression coefficient calculating means;
The similar time series data calculation method according to claim 1, wherein: Counting the number of segments of a specific length by the segment counting means;
A step of integrating two segments of the length when the counted number becomes larger than a certain value by the segment integration means;
Obtaining a mean value and a standard deviation in the segment after integration by the dimension compression coefficient calculating means;
The similar time series data calculation method according to claim 3, wherein: Dividing the time series data into a plurality of segments of equal length by segment management means;
Obtaining a mean value and a standard deviation in each segment of the time series data by the dimension compression coefficient calculating means;
The similar time series data calculation method according to claim 1, wherein: A dimensional compression coefficient calculation means for reading a pair of time series data to be subjected to similarity determination from the memory, obtaining an average value and a standard deviation as a dimensional compression coefficient for each, and storing them in the memory;
An upper limit / lower limit calculation means for reading an average value and a standard deviation from the memory, and using these to obtain an upper limit value and a lower limit value of the distance of the time series data and storing them in the memory,
An similarity determination unit that reads the upper limit value, the lower limit value, and the threshold value from the memory, and determines that the pair of time series data is similar when the lower limit value is equal to or less than the threshold value and the upper limit value is equal to or less than the threshold value;
A similar time-series data calculation device comprising: A process of reading a pair of time series data to be subjected to similarity determination from the memory, obtaining an average value and a standard deviation as a dimensional compression coefficient for each, and storing them in the memory;
A process of reading an average value and a standard deviation from the memory, and using these to obtain an upper limit value and a lower limit value of the distance of the time series data and storing them in the memory,
A process of reading the upper limit value, the lower limit value, and the threshold value from the memory, and determining that the pair of time series data is similar when the lower limit value is equal to or less than the threshold value and the upper limit value is equal to or less than the threshold value;
A similar time series data calculation program characterized by causing a computer to execute.

Images