JP2013089112A - Analysis device and analysis program of time series data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To analyze not only a singular point but also increase and decrease tendency in time series data.SOLUTION: A time series data analysis device comprises: a regression line generation unit 2 which, for time series data of a predetermined period which is specified by a user within time series data, generates a plurality of patterns of regression lines from the time series data of the predetermined period; an index calculation unit 3 which calculates predetermined indexes for the respective patterns; a separation point extraction unit 4 which extracts separation points of a regression line from a pattern having the best index among the plurality of patterns; and a tendency line generation unit 5 which generates a tendency line of the time series data by calculating a regression line of the time series data using the plurality of separation points extracted from the time series data as a boundary. By specifying tendency of time series data with a line, it is possible to analyze increase and decrease tendency of the time series data by the inclination of the line, and analyze a point at which inclination of line significantly changes as a singular point.

Description

  The present invention relates to a time-series data analysis apparatus and analysis program, and is particularly suitable for use in an analysis apparatus and analysis program that performs data mining on time-series data.

  In general, a technique for extracting some knowledge from a large amount of data by performing data analysis based on statistics or pattern recognition is known. This is an analysis technique called data mining. For example, a method for detecting statistical outliers and change points appearing in time-series data by applying an AR model (Auto-Regressive) curve to time-series data and analyzing it has also been proposed. (For example, refer to Patent Document 1).

JP 2004-54370 A

  However, in the conventional technique disclosed in Patent Document 1, since a curve is applied to time series data for analysis, it is possible to detect singular points such as outliers and change points of time series data. There was a problem that the trend of increase / decrease in series data could not be grasped.

  The present invention has been made to solve such a problem, and an object thereof is to analyze not only the singular points of time series data but also the increase / decrease tendency of the time series data.

  In order to solve the above-described problems, in the present invention, a regression line is applied to time series data for analysis. Specifically, the present invention generates a plurality of patterns of regression lines from time-series data within a period, which is time-series data within the predetermined period, for a predetermined period set from time-series data. The dividing point of the regression line is extracted from the pattern having the best predetermined index value among the patterns. Then, the trend line of the time-series data is generated by moving the predetermined period sequentially from the beginning to the end of the time-series data and performing the same process, and obtaining the regression line with the multiple division points extracted as a boundary. Like to do.

  According to the present invention configured as described above, since the tendency of the time series data is specified by a straight line, the increase / decrease tendency of the time series data can be analyzed by the slope of the straight line. In addition, a point where the slope of the straight line changes greatly can be analyzed as a singular point. Thereby, according to this invention, in addition to the singular point which appears in time series data, the increase / decrease tendency of time series data can also be analyzed.

It is a block diagram which shows the function structural example of the analysis apparatus of the time series data by this embodiment. It is a figure for demonstrating the processing content of the regression line production | generation part by this embodiment, and is a figure which shows the state which specified the predetermined period from the time series data. It is a figure for demonstrating the processing content of the regression line production | generation part by this embodiment, and is a figure which shows the state which produces | generates the regression line of a some pattern from the time series data in a period. It is a figure for demonstrating the processing content of the tendency straight line generation part by this embodiment. It is a flowchart which shows the operation example of the analysis apparatus of the time series data by this embodiment.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram illustrating a functional configuration example of a time-series data analysis apparatus according to the present embodiment. As shown in FIG. 1, the time-series data analysis apparatus according to the present embodiment has, as its functional configuration, a period designation receiving unit 1, a regression line generation unit 2, an index calculation unit 3, a dividing point extraction unit 4, and a trend line generation. Unit 5, second index calculation unit 6, optimum trend straight line specifying unit 7, increase / decrease tendency specifying unit 8, singular point specifying unit 9, similar data searching unit 10, time series data storage unit 20, division point storage unit 21, trend A straight line storage unit 22 and a search target data storage unit 23 are provided.

  Note that the functional configurations listed above can be realized by any of a hardware configuration, a DSP, and software. For example, when realized by software, the time-series data analysis apparatus according to the present embodiment is actually configured with a computer CPU or MPU, RAM, ROM, etc., and an analysis program stored in the RAM or ROM operates. It can be realized by doing.

  Therefore, the above functional configuration can be realized by recording the analysis program on a recording medium such as a CD-ROM and causing the computer to read the analysis program. As a recording medium for recording the analysis program, a flexible disk, a hard disk, a magnetic tape, an optical disk, a magneto-optical disk, a DVD, a nonvolatile memory card, and the like can be used in addition to the CD-ROM. The analysis program may be downloaded to a computer via a network such as the Internet.

  The period designation accepting unit 1 accepts designation of the size of a predetermined period as a parameter w through an operation by a user of an operation unit (not shown) such as a keyboard or a mouse. The parameter w received here is a value indicating the size of the predetermined period, and does not indicate the position of the predetermined period. The size of the predetermined period indicates, for example, how many pieces of data among time-series data composed of a plurality of data are to be analyzed. One or a plurality of parameters w can be designated.

  The regression line generation unit 2 performs regression analysis on the time series data stored in the time series data storage unit 20 to generate a regression line of the time series data. Here, the regression line generation unit 2 targets a predetermined period determined based on the parameter w specified by the period specification receiving unit 1 and uses a plurality of patterns from the time-series data in the period that is the time-series data in the predetermined period. Generate a regression line for Specifically, the regression line generation unit 2 generates two regression lines before and after a time point within a predetermined period as a plurality of patterns of regression lines, and 1 without setting a boundary within the predetermined period. Generate a regression line of books.

  2 and 3 are diagrams for explaining the processing contents of the regression line generation unit 2. FIG. 2 is a diagram illustrating a state in which a predetermined period having a size specified by the parameter w is specified from time-series data. FIG. 3 is a diagram illustrating a state in which a plurality of patterns of regression lines are generated from time-series data within a predetermined period for the specified predetermined period. 2A shows a state in which a predetermined period is specified in the first part of the time-series data, and FIG. 3 generates a plurality of patterns of regression lines in the predetermined period specified in FIG. 2A. Indicates the state.

  In the example of FIGS. 2 and 3, the value of the parameter w is set to “5”. In the case of w = 5, the regression line generation unit 2 generates three patterns of regression lines as shown in FIGS. FIG. 3A shows a predetermined period (time point t1 to t5) set in the first part of the time series data composed of 17 pieces of data from time points t1 to t17, with the second time point t2 as a boundary. 2 shows a state in which two regression lines are generated. That is, one regression line is generated between time points t1 and t2, and one regression line is generated between time points t3 and t5.

  FIG. 3B shows a state in which two regression lines are generated before and after the third time point t3 as a boundary. That is, one regression line is generated between the time points t1 and t3 and one between the time points t4 and t5. FIG. 3C shows a state in which one regression line is generated between time points t1 and t5 without setting a boundary in a predetermined period (time points t1 to t5) set in the first part. Yes.

The index calculation unit 3 shows an index for evaluating the magnitude of the error between the regression line generated by the regression line generation unit 2 and the actual value of the time-series data within the period, and the complexity of the linear model in FIG. Calculation is performed for each of a plurality of patterns shown in a) to (c). As this index, for example, a known information criterion can be used as an index representing the goodness of a statistical model. In this embodiment, the Akaike information criterion AIC _sum and AIC _term as shown in the following (Expression 1) and (Expression 2) are used.

  Note that (Equation 1) is an index used when there are two regression lines (in the case of FIGS. 3 (a) and (b)), and (Equation 2) is when there is only one regression line (FIG. 3 ( This is an index used in the case of c). In the above (Expression 1), t is a numerical value indicating the number of the boundary point when viewed from the start point of the predetermined period. In the case of FIG. 3A, t = 2, and in the case of FIG. 3B, t = 3.

Further, S _e1 in the first term of the right side of (Equation 1) indicates the residual sum of squares of the errors between the actual value of the time series data regression in the linear and duration before the boundary point, also the right-hand side 2 _{Se 2} in the term represents the residual sum of squares of errors between the regression line after the boundary point and the actual value of the time-series data within the period. S _e on the right side of equation (2) shows the residual sum of squares of the errors between the actual value of the time series data within a single regression line and duration.

The dividing point extraction unit 4 identifies the pattern with the best index calculated by the index calculation unit 3 from the plurality of regression line patterns generated by the regression line generation unit 2, and determines the regression line from the identified pattern. Extract dividing points. The best index means that the values of the Akaike information criterion AIC _sum and AIC _term calculated by (Equation 1) and (Equation 2) are the smallest.

  The division point extracted by the division point extraction unit 4 corresponds to the boundary point in the case of a pattern including two regression lines as shown in FIG. That is, when the pattern index shown in FIG. 3A is the best, the division point is t2. When the index of the pattern shown in FIG. 3B is the best, the division point is t3. On the other hand, when the index of the pattern including one regression line is the best as shown in FIG.

  When the division point extraction unit 4 extracts the division point, the division point extraction unit 4 stores the division point in the division point storage unit 21. In addition, the dividing point extraction unit 4 notifies the regression line generation unit 2 of the processing result. That is, when a division point is extracted, the division point is notified to the regression line generation unit 2, and when a division point is not extracted, that fact is notified to the regression line generation unit 2. The regression line generation unit 2 that has received the notification from the dividing point extraction unit 4 moves the predetermined period by an amount corresponding to the notified processing result, and executes the above-described processing in the predetermined period after the movement.

  Here, when the regression line generation unit 2 receives notification of the extracted division point, the regression line generation unit 2 moves the start point of the predetermined period to the division point and performs the above-described processing. For example, when the index of the pattern shown in FIG. 3B is the best and the dividing point t3 is extracted, the regression line generation unit 2 starts the predetermined period until the dividing point t3 as shown in FIG. 2B. And the above-described processing is performed. On the other hand, when receiving a notification that the dividing points have not been extracted, the regression line generation unit 2 performs the above-described processing by moving the start point of the predetermined period to the next time point after the start point of the immediately preceding predetermined period.

  If a regression line is generated by the regression line generation unit 2 in a predetermined period after movement, the index calculation unit 3 and the dividing point extraction unit 4 also perform the above-described processing in the predetermined period after movement. Such processing is sequentially repeated. That is, the processes of the regression line generation unit 2, the index calculation unit 3, and the dividing point extraction unit 4 are performed a plurality of times by sequentially moving the predetermined period from the beginning to the end of the time series data. When the predetermined period cannot be taken by the length of the parameter w at the end of the time series data, the predetermined period is set to the end of the time series data. By performing such processing a plurality of times, the dividing point storage unit 21 stores a plurality of dividing points.

  The trend line generation unit 5 obtains a regression line of the time series data stored in the time series data storage unit 20 with a plurality of division points stored in the division point storage unit 21 as boundaries, thereby obtaining time series data Generate a trend line. FIG. 4 is a diagram for explaining the processing content of the trend line generator 5. In the example of FIG. 4, three division points t3, t9, and t13 are extracted by the division point extraction unit 4 and stored in the division point storage unit 21. In this case, the trend line generation unit 5 uses three division points t3, t9, and t13 as boundaries, and obtains a regression line of time series data before and after the boundaries.

  That is, the trend line generator 5 performs the regression line 1 between the time points t1 and t3, the regression line 2 between the time points t4 and t9, the regression line 3 between the time points t10 and t13, and the regression between the time points t14 and t17. A straight line 4 and a total of four regression lines are obtained. These four regression lines or the entire straight line connecting them is a trend line of time series data composed of 17 data from time points t1 to t17. The trend line generation unit 5 stores the trend line generated in this way in the trend line storage unit 22.

  When the period specification receiving unit 1 receives specification of a plurality of parameters w (w1, w2,...), The regression line generation unit 2, the index calculation unit 3, the dividing point extraction unit 4, and the trend line generation unit 5 Performs the above-described processing for each parameter w1, w2,. Thereby, a plurality of trend lines corresponding to the parameters w1, w2,... Are generated and stored in the trend line storage unit 22.

The second index calculation unit 6 stores in the trend line storage unit 22 a second index for evaluating the magnitude of error between the trend line and the actual value of the time series data and the complexity of the line model. Calculation is performed for each of a plurality of trend lines. For this second index, for example, a known information criterion can be used as an index representing the goodness of the statistical model. In this embodiment, the Akaike information criterion AIC _all as shown in the following (formula 3) is used.

In the above (Formula 3), i is a numerical value indicating the number of the plurality of regression lines included in the trend line from the beginning. n _i is a numerical value indicating the number of division points when viewed from the start point of each regression line. In the example of FIG. 4, n ₁ = 3 for the regression line ₁ , n ₂ = 6 for the regression line ₂ , n ₃ = 4 for the regression line 3, and n ₄ = 4 for the regression line 4. is there. S _{ei on} the right side of (Expression 3) indicates the residual sum of squares of errors between the i-th regression line and the actual value of the time series data.

The optimum trend straight line identifying unit 7 identifies the trend straight line having the best second index calculated by the second index calculating unit 6 among the plurality of trend straight lines stored in the trend straight line storage unit 22. The best index means that the value of the Akaike information criterion AIC _all calculated by (Equation 3) is the smallest. The optimum trend straight line specifying unit 7 stores the identified optimum trend straight line in the trend straight line storage unit 22 in a state where it can be identified from other trend straight lines.

  Further, the optimum trend straight line specifying unit 7 reads time-series data corresponding to the identified optimum trend straight line from the time-series data storage unit 20, and associates the optimum trend straight line with the corresponding time-series data to store search target data. Store in the unit 23. Each time the optimum trend line is obtained by performing the above-described processing on different time series data, the search target data storage unit 23 stores the optimum trend line and the corresponding time series data. A plurality of sets of time-series data and its trend line stored in the search target data storage unit 23 in this way are used as search target data for pattern recognition as will be described later.

  When the period designation receiving unit 1 receives designation of only one parameter w, only one trend line is generated and stored in the trend line storage unit 22. In this case, the second index calculation process by the second index calculation unit 6 and the optimum trend line identification process by the optimum trend line identification unit 7 are unnecessary. The optimum trend line specifying unit 7 associates one trend line stored in the trend line storage unit 22 with time series data corresponding to the trend line and stores it in the search target data storage unit 23.

  The increase / decrease trend specifying unit 8 determines the inclination of the trend line (the optimum trend line when a plurality of parameters w1, w2,... Are specified) generated by the trend line generation unit 5 and stored in the trend line storage unit 22. Based on this, the increase / decrease tendency of the time series data is specified. For example, the increase / decrease trend specifying unit 8 accepts designation of an arbitrary time t as a parameter, and specifies the increase / decrease trend of the time-series data at the specified time t. Specifically, it is determined that if the slope of the trend line is positive at a specified time t, it is an increasing trend, and if it is negative, it is a decreasing trend.

  The singular point specifying unit 9 specifies a sudden increase point or a sudden decrease point of the time series data as a singular point. Candidates for sudden increase / decrease points are a plurality of division points recorded in the division point storage unit 21 as boundaries between regression lines. That is, the singular point specifying unit 9 calculates, for a plurality of division points extracted by the division point extraction unit 4 and stored in the division point storage unit 21, differences in actual values of time-series data at points before and after the division point. Then, the division point where the difference is equal to or greater than a predetermined value is specified as a sudden increase point or a sudden decrease point. For example, the singularity specifying unit 9 obtains the standard deviation σ of the error between the trend line and the actual value of the time series data, and if the difference between the actual values of the time series data at the points before and after the dividing point is an increase of 2σ or more. If it is a sudden increase or a decrease of -2σ or less, it is judged as a rapid decrease.

  The similar data search unit 10 searches for a trend line similar to the trend line generated by the trend line generation unit 5 (an optimal trend line when a plurality of parameters w1, w2,... Are specified) by pattern recognition. A search is performed from the data storage unit 23, and time series data corresponding to the searched trend line is extracted from the search target data storage unit 23. That is, the similar data search unit 10 searches the search target data storage unit 23 for another time series data having a tendency similar to the time series data to be analyzed this time.

  FIG. 5 is a flowchart illustrating an operation example of the time-series data analysis apparatus according to the present embodiment. FIG. 5 shows an operation example when generating a trend line from time series data. The flowchart shown in FIG. 5 starts when the time-series data to be analyzed is read from the time-series data storage unit 20 and the user specifies the value of the parameter w by operating the operation unit (not shown).

  In FIG. 5, the regression line generation unit 2 acquires the designated parameter w (step S1), and sets a predetermined period having the width of the parameter w in the time series data (step S2). Initially, a predetermined period is set at the beginning of the time series data. Then, the regression line generation unit 2 generates a plurality of patterns of regression lines by performing regression analysis of the time series data in the period among the time series data read from the time series data storage unit 20 for the set predetermined period. (Step S3).

The index calculation unit 3 calculates an index based on the Akaike information criterion AIC _sum and AIC _term shown in (Equation 1) and (Equation 2) for each of the plurality of patterns of regression lines generated by the regression line generation unit 2 ( Step S4). Next, the dividing point extraction unit 4 identifies a pattern with the best index calculated by the index calculation unit 3 from the plurality of regression line patterns generated by the regression line generation unit 2, and uses the identified pattern. A division point of the regression line is extracted (step S5).

  Here, the division point extraction unit 4 determines whether or not the division point can be extracted from the predetermined period to be analyzed (step S6). If the division point cannot be extracted, the process proceeds to step S8. On the other hand, when the dividing point can be extracted, the dividing point extracting unit 4 stores the dividing point in the dividing point storage unit 21 (step S7). The dividing point extracting unit 4 notifies the regression line generating unit 2 of the fact that the dividing point cannot be extracted, and when it can be extracted.

  Receiving this notification, the regression line generator 2 determines whether or not the analysis has been completed by setting a predetermined period until the end of the time series data (step S8). If the analysis has not been completed yet until the end of the time series data, the process returns to step S2 to move a predetermined period according to the notification result. And the process after step S3 is similarly performed about the predetermined period after a movement.

  On the other hand, when the analysis is completed until the end of the time series data, the trend line generation unit 5 obtains a regression line of the time series data by using a plurality of division points stored in the division point storage unit 21 as boundaries, thereby obtaining a time series. A trend line of data is generated (step S9). And the tendency straight line is memorize | stored in the tendency straight line memory | storage part 22 (step S10).

  Subsequently, the regression line generation unit 2 determines whether or not all parameters w designated by the user have been processed (step S11). If only one parameter w is designated by the user, all designated parameters w have been processed. On the other hand, if a plurality of parameters w are designated by the user, it is determined whether or not there are any unprocessed ones. If not, all designated parameters w have been processed.

  If an unprocessed parameter w remains, the process returns to step S1, and the process after step S2 is similarly performed on the newly acquired parameter w. On the other hand, when all the designated parameters w have been processed, the second index calculation unit 6 determines whether there are a plurality of designations of the parameter w (step S12). When a plurality of parameters w are specified, a plurality of trend lines are stored in the trend line storage unit 22, and therefore the second index calculation unit 6 performs the operation for each of the plurality of trend lines. 2 is calculated (step S13).

  Subsequently, the optimum trend line specifying unit 7 specifies the trend line having the best second index calculated by the second index calculating unit 6 among the plurality of trend lines stored in the trend line storage unit 22. (Step S14). Then, the optimum trend straight line identifying unit 7 stores the identified optimum trend straight line in the trend straight line storage unit 22 in a state where it can be identified from other trend straight lines. Further, the optimum trend straight line specifying unit 7 reads time-series data corresponding to the identified optimum trend straight line from the time-series data storage unit 20, and associates the optimum trend straight line with the corresponding time-series data to store search target data. The data is stored in the unit 23 (step S15).

  When only one parameter w is specified, the trend line storage unit 22 stores only one trend line. In this case, the processes of steps S13 and S14 are not performed, and the optimum trend line specifying unit 7 obtains time series data corresponding to one trend line stored in the trend line storage unit 22 from the time series data storage unit 20. The data is read out, and the trend line and the time-series data corresponding thereto are associated with each other and stored in the search target data storage unit 23 (step S15). Thereby, the process of the flowchart shown in FIG.

  As described above in detail, according to the present embodiment, the trend of time-series data is specified by a straight line. Therefore, the increase / decrease tendency of the time-series data is analyzed by the slope of the straight line, or the slope of the straight line changes greatly. Etc. can be analyzed as singular points. For example, if the trend line generated by the trend line generation unit 5 is displayed as a graph on the display, the user can intuitively grasp the increase / decrease tendency and singularity of the time series data. Further, by performing processing by the increase / decrease trend specifying unit 8 or the singular point specifying unit 9 for the trend line generated by the trend line generating unit 5, it is possible to specify the increase / decrease trend or singular point of the time series data by a computer. it can.

  In the above embodiment, the user can designate the width of the predetermined period as the parameter w, but it may be a fixed value. However, by making the width of the predetermined period variable as the parameter w, there is an advantage that time series data can be analyzed flexibly. That is, if the value of the parameter w is increased, the general tendency of the time series data can be analyzed. If the value of the parameter w is decreased, the tendency of the fine part of the time series data can be analyzed.

  In the above embodiment, the parameter w can be specified by one or a plurality of parameters. However, only one parameter w may be specified. In this case, the second index calculation unit 6 and the optimum trend straight line identification unit 7 are not necessary. However, a trend line that more faithfully represents the characteristics of the time-series data can be generated by specifying the plurality of parameters w and performing the processing of the second index calculation unit 6 and the optimum trend line specifying unit 7. It has the merit of becoming

  In the above embodiment, the increase / decrease tendency specifying unit 8, the singular point specifying unit 9, and the similar data searching unit 10 are provided, but these are not essential components and may be omitted. Alternatively, only one or two of them may be provided. For example, even if the increase / decrease trend specifying unit 8 and the singular point specifying unit 9 are omitted, if the trend line generated by the trend line generating unit 5 is displayed as a graph on the display, the increase / decrease trend or singularity of the time series data can be displayed. The user can grasp intuitively.

  In the above embodiment, the trend line generated by the trend line generation unit 5 and the time series data corresponding to the trend line are stored in the search target data storage unit 23, and a search for pattern recognition of another time series data later is performed. Although used as target data, the present invention is not limited to this. For example, several patterns of time-series data to be searched and its trend line may be generated in advance and stored in the search target data storage unit 23.

  In the above embodiment, the Akaike information criterion is used as an example of the index, but the present invention is not limited to this. For example, other information criterion such as Bayesian statistical criterion may be used.

  In addition, each of the above-described embodiments is merely an example of implementation in carrying out the present invention, and the technical scope of the present invention should not be construed in a limited manner. That is, the present invention can be implemented in various forms without departing from the gist or the main features thereof.

DESCRIPTION OF SYMBOLS 1 Period designation reception part 2 Regression line generation part 3 Index calculation part 4 Division point extraction part 5 Trend line generation part 6 2nd index calculation part 7 Optimal tendency straight line specific part 8 Increase / decrease tendency specific part 9 Singular point specific part 10 Similar data Search part

Claims (8)

For a predetermined period set out of time series data composed of a plurality of data, a regression line generation unit that generates a plurality of patterns of regression lines from time series data within a period that is time series data within the predetermined period;
Index calculation for calculating an index for evaluating the magnitude of error between the regression line generated by the regression line generation unit and the actual value of the time-series data within the period and the complexity of the linear model for each of the plurality of patterns And
A dividing point extracting unit that identifies a pattern with the best index calculated by the index calculating unit and extracts a dividing point of a regression line from the identified pattern;
A plurality of division points extracted by performing the processing of the regression line generation unit, the index calculation unit, and the division point extraction unit a plurality of times by sequentially moving the predetermined period from the beginning to the end of the time series data. An apparatus for analyzing time-series data, comprising: a trend line generation unit that generates a trend line of the time-series data by obtaining a regression line of the time-series data as a boundary. The regression line generation unit generates two regression lines before and after the time point within the predetermined period as a boundary of the plurality of patterns, and sets one line without setting the boundary within the predetermined period. The time series data analysis apparatus according to claim 1, wherein a regression line is generated. 2. The time series data according to claim 1, further comprising an increase / decrease tendency specifying unit that specifies an increase / decrease tendency of the time series data based on an inclination of the trend line generated by the trend line generation unit. Analysis device. For a plurality of division points extracted by the division point extraction unit, a difference between actual values of the time series data at points before and after the division point is calculated, and a difference between actual values at points before and after the division point is predetermined. The time-series data analysis apparatus according to claim 1, further comprising a singular point identification unit that identifies a division point that is equal to or greater than a value as a sudden increase point or a sudden decrease point. A search target data storage unit that stores a plurality of sets of arbitrary time-series data and its trend line as search target data;
A trend line similar to the trend line generated by the trend line generation unit is searched from the search target data storage unit by pattern recognition, and time series data corresponding to the searched trend line is extracted from the search target data storage unit The time-series data analysis apparatus according to claim 1, further comprising a similar data search unit. A period designation accepting unit that accepts designation of the size of the predetermined period as a parameter;
The regression line generator generates a plurality of patterns of regression from time-series data within a period, which is time-series data within the predetermined period, for the predetermined period determined based on the parameter received by the period specification receiving unit. The time series data analysis apparatus according to claim 1, wherein a straight line is generated. When the period specification receiving unit receives specification of a plurality of parameters, the regression line generation unit, the index calculation unit, the division point extraction unit, and the trend line generation unit perform processing for each of the plurality of parameters. To generate multiple trend lines,
A second index calculation unit that calculates a second index for evaluating the magnitude of error between the trend line and the actual value of the time-series data and the complexity of the line model for each of the plurality of trend lines; ,
7. An optimum trend straight line specifying unit for specifying a best trend straight line for the second index calculated by the second index calculating unit among the plurality of trend straight lines. Analyzing device for time series data described in 1. Regression line generating means for generating a plurality of patterns of regression lines from time-series data within a period, which is time-series data within the predetermined period, for a predetermined period set from time-series data composed of a plurality of data,
Index calculation for calculating each of the plurality of patterns as an index for evaluating the magnitude of error between the regression line generated by the regression line generation means and the actual value of the time-series data within the period and the complexity of the linear model means,
The index calculated by the index calculation means identifies a pattern having the best index, and a dividing point extracting means for extracting a dividing point of a regression line from the identified pattern, the regression line generating means, the index calculating means, and the division By calculating a regression line of the time series data with a plurality of division points extracted as a boundary by sequentially moving the predetermined period from the beginning to the end of the time series data and performing a plurality of times. A program for analyzing time series data for causing a computer to function as a trend line generating means for generating a trend line of the time series data.

Images