JPH0812659B2 - Database search method for time-series data - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method for retrieving data showing similar fluctuations in time series data collected at equal time intervals.

[Conventional technology]

In a process such as a sewage treatment process whose internal structure is not clear and has many disturbances, past operating experience is often referred to when the state of the process changes.
Therefore, the state of the process is usually measured by various detectors and recorded on a recording paper of a recorder, a daily operation report, a storage device using a computer, or the like. When the state of the process changes, the past measurement record is referred to and the operation for stabilizing the process is performed.

However, in order to retrieve a situation similar to the situation of the current process from past measurement records, it is necessary to manually investigate the above-mentioned recording medium based on the memory of a person concerned such as an operation manager. . Therefore, the survey takes a long time, and the survey may be omitted. In addition, there is a problem in that the judgment of similarity largely depends on the subjectivity of the person in charge of the survey, and cannot be said to be an objective evaluation.

Therefore, for example, in the invention proposed in Japanese Examined Patent Publication No. 60-53340, the measured time-series data is Fourier-transformed at regular intervals, and the Fourier coefficient is stored in a storage device. The time series data was also Fourier transformed and the coefficients were accumulated.

[Problems to be solved by the invention]

However, in this conventional method, when the correlation coefficient between the Fourier coefficient of the time-series data of interest and the Fourier coefficient of the past data is obtained and the similarity is determined based on the magnitude of the correlation coefficient, when performing the Fourier transform. The degree of similarity may differ depending on how the starting point of the series data is set. This point will be described below using a specific example.

Fig. 3 shows the measurement results of the difference in UV-visible light absorbance of the aeration tank inflow water of a certain sewage treatment plant. FIG. 3 (a) is time series data in units of calendar months for June 1985, (b) July 1985, and (c) April 1986. The data length of such time-series data is set to one month, and the starting point thereof is the start of the calendar month, Fourier transform is performed, and Fourier coefficients are accumulated. Then, in order to retrieve data similar to the data for April 1986 shown in FIG. 3 (c), the data for April 1986 is Fourier-transformed, and the obtained Fourier coefficient and the accumulated Fourier coefficient are stored. The correlation coefficient with the coefficient was obtained by the equation (2) described later.

As a result, the correlation coefficient between the data for June 1985 and the data for April 1986 shown in Figure 3 (a) was 0.79, and that for July 1985 in Figure 3 (b) was 0.60. became. However, if the starting point of the time series data is changed and the one month from June 10, 1985 to July 10, 1985 shown in Fig. 3 (d) is treated as the data to be searched, The correlation coefficient is 0.98, indicating that the degree of similarity is high.

As described above, in the conventional method, the degree of similarity varies depending on how the starting point of the time-series data is set, and there is a risk that similar data may be missed.

The present invention has been made in view of such a conventional problem, and a portion similar to a specific operation status from a past operation record of a process, based on an objective evaluation, is provided with a dark circle in a short time. The purpose is to be able to search without.

[Means for solving problems]

In order to achieve this object, the database search method for time-series data of the present invention is a Fourier transform for each fixed interval obtained by equally dividing a period from the start point to the end point of a plurality of time series data from a start point to a end point. Then, the Fourier series obtained as a result of this Fourier transform is stored, and the time series data to be searched is also Fourier-transformed at the same constant intervals as above, and the Fourier coefficient obtained by the Fourier transform is , A correlation coefficient with a Fourier coefficient of the Fourier series that has already been stored is calculated for each section, and a portion in which a section with a large correlation coefficient is continuous is detected as similar time series data. And

When storing the Fourier series obtained as a result of the Fourier transform, it is not always necessary to store all the number of original data, and an order sufficient to express the behavior of the original time series data, for example, 1 Memory capacity can be saved by thinning out the number such as / 3 or more.

[Action]

The flow of the method of the present invention is shown in FIG. The present invention, in order to grasp the status of a process, various sensors and measuring instruments 1 installed at various points in the process, and a time-series data sampling device 2 for sampling outputs from the measuring instruments 1 at equal time intervals. , A time series data storage device 3 for storing time series data input from the time series data collection device 2,
A Fourier transform device 4 for performing a Fourier transform of the time series data stored in the time series data storage device 3, a Fourier coefficient storage device 5 for storing the Fourier coefficient obtained as a result of the Fourier transform, and a similar data search specification. According to the output from the search command device 6 and the search command device 6
Correlation coefficient calculation device 7 for obtaining the correlation coefficient between the stored Fourier coefficients, compares the obtained correlation coefficient with a preset reference value of the correlation coefficient, and binarizes the result. Correlation coefficient comparison device 8, comparison result storage device 9 for storing binarized comparison results, continuity determination device 10 for examining continuity of comparison results, and search result output device for outputting search results Composed of 11.

In the present invention, first, measurement signals from various sensors and measuring instruments 1 installed at various points in the process are sampled at equal time intervals by the time series data sampling device 2 and stored in the time series data storage device 3. Keep it. When the time-series data for a fixed time set to be relatively short is stored, this is output to the Fourier transform device 4 to perform the Fourier transform. That is, the time series data X (t) is Fourier-transformed according to the following equation (1) to obtain Fourier coefficients A _n and B _n .

However, Δt is the sampling interval of the time series data, and N is the number of data of the time series data.

The Fourier coefficient thus obtained is stored in the Fourier coefficient storage device 5. Therefore, past operation records are stored and accumulated in the Fourier coefficient storage device 5 in the form of Fourier coefficients.

When searching for similar data, the period of the data of interest, the frequency band of the data of interest, and the first and last data to be searched, such as from XX year to YY year Is input from the search command device 6. Next, the Fourier coefficient storage device 5 outputs the Fourier coefficient of the data of interest and the Fourier coefficient of the data to be searched to the correlation coefficient calculation device 7 in accordance with the search specifications from the search command device 6 and compares them. The correlation coefficient divided into small pieces is calculated by the correlation coefficient calculation device 7 for each data section.

The method of obtaining the correlation coefficient is as described below. That is, if the Fourier coefficients of the time-series data of interest are α _n and β _n, and the time-series data to be searched are a _n and b _n , the correlation coefficient r is calculated by the following equation. be able to.

However, β _o = 0 and b _o = 0.

The correlation coefficient r calculated by the equation (2) is output to the correlation coefficient comparison device 8. The correlation coefficient comparison device 8 compares the magnitude of the preset correlation coefficient with the reference value,
The result is binarized. That is, if it is larger than the reference value, "1" is stored in the comparison result storage device 9 as having similarity, and if it is smaller than the reference value, "0" is stored as not having similarity.
Of course, as the logic of binarization, the logic reverse to the above may be applied.

Next, the continuity determination device 10 determines whether or not “1” is continuous in the output from the comparison result storage device 9 over a period corresponding to the data length of the time series data of interest, that is, there is similarity. See if the data is continuous. That is, in the time-series data in which the similarity is examined by dividing the data into relatively small pieces, if the data determined to have the similarity is continuous with the data of interest for the same number of days, it may be determined to be the similar data.

The search result of the similar data thus obtained is output from the search result output device 11.

〔Example〕

Hereinafter, as an example of the present invention, a time-series data retrieval result of an ultraviolet-visible-light absorbance difference in aeration tank inflow water in a sewage treatment process will be described. Here, let us consider a case of searching for data showing a variation similar to the data of April 1986 shown in FIG. 3 (c).

A process with slow fluctuations such as a sewage treatment process needs to have similar fluctuations for about one month. Therefore, the time-series data is divided into three units of 10 days each at the beginning, middle, and end of the calendar month, and the Fourier coefficient is accumulated for each section. Next, we are paying attention
The data for April 1986 is also 3 in early, mid, and late.
The division was performed to obtain the Fourier coefficient, and the correlation coefficient with the accumulated Fourier coefficient was calculated. Some of the results are shown in Table 1. And, of the correlation coefficient, the reference value for comparison is 0.95,
Table 2 shows the comparison results binarized.

By the way, when you search for the part where the section judged to be “similar” to the data of the beginning, middle and end of April 1986 is continuous, the mark of ○ is added in Table 2 to the middle of June 1985. The section from the beginning of July corresponds to that, and it was detected as similar data. It can be understood that the time-series data shown in FIG. 3D is detected in this way, and is similar data when compared with FIG. 3C.

The present invention has been described above with respect to the configuration for retrieving similar data from past operation records of the same plant. But,
As shown in FIG. 2, a time series data transfer device 12 is installed in another plant, and a time series data receiving device 13 is added,
If the Fourier coefficients of the data of other plants are also stored, it is possible to construct an operation record reference database over a wide area.

Further, in the above description, when calculating the correlation coefficient,
As shown in the equation (2), all the obtained Fourier coefficients are used. However, if the correlation coefficient is calculated by the equation (3), the similarity can be examined for the specific frequency band of the time series data.

And

〔The invention's effect〕

As described above, in the present invention, time-series data is divided into relatively small sections and Fourier-transformed, the correlation coefficient is obtained for the Fourier coefficient, and the reference value of the preset correlation coefficient is set. In comparison, the portion in which the sections determined to have similarity are continuous is detected as similar data. Since such a search process is performed by a computer, the time required for the search is significantly shortened, and the data stored in the storage device is searched sequentially, so that the search omission is eliminated. Further, since the correlation coefficient is calculated according to the defined formula and the similarity is evaluated based on the magnitude thereof, it is possible to make an objective judgment. The data is divided into relatively small sections, the similarity is examined, and portions with a high degree of similarity are continuously detected as similar data.
It is not affected by the data division method. Since data of other plants can be input as the data to be stored in the storage device, the survey target can be expanded in a wider range. When calculating the correlation coefficient, if the number of terms of the Fourier coefficient to be used is designated, similar data can be searched for in a specific frequency band of the time series data.

[Brief description of drawings]

FIG. 1 is a flow chart of a similar time series data retrieval system according to the present invention, FIG. 2 is a block diagram of a wide area retrieval system according to the present invention, and FIG. 3 is a graph of ultraviolet-visible-light absorbance difference of aeration tank inflow water. .

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location 9194-5L G06F 15/40 370 Z

Claims (1)

[Claims] 1. A plurality of time-series data from a certain starting point to a certain ending point are Fourier-transformed for each constant section obtained by equally dividing a period from the starting point to the ending point, and a Fourier series obtained as a result of this Fourier transformation is stored. Also, with respect to the time series data to be searched, Fourier transform is performed for each of the same constant intervals as described above, and the Fourier coefficient obtained by the Fourier transform and the Fourier coefficient of the previously stored Fourier series are stored. A database search method for time-series data, wherein a correlation coefficient is calculated for each of the sections, and a portion in which sections having a large correlation coefficient are continuous is detected as similar time-series data.