JP6365145B2 - Time series data analysis method and time series data abnormality monitoring apparatus - Google Patents

Description

  The present invention relates to a time-series data analysis method and time-series data abnormality monitoring apparatus for analyzing time-series data using a chaos analysis technique.

  In order to detect anomalies and signs of monitoring, obtain time-series data from monitoring objects, and analyze this time-series data using the Trajectory Parallel Measure Method (TPM method) based on chaos theory. Devices have been proposed (for example, Patent Documents 1 and 2 and Non-Patent Document 1). The trajectory parallel measure method is useful as a means for detecting a stochastic process factor mixed in time series data based on determinism. For example, in Patent Document 1, the trajectory parallel measure (TPM) of time series data is sequentially determined. To detect stochastic factors in time-series data.

  The orbital parallel measure method is not only a probabilistic process factor included in such time series data, but also when the dynamic system behind the time series data transitions to another dynamic system, the transition is regarded as a change in TPM. Can be caught. Therefore, it is expected to capture the transition of the dynamic system of time series data using the orbit parallel measure method.

JP2013-211002A Japanese Patent Laid-Open No. 10-134034

Yasunari Fujimoto, Tadashi Izumi Hata, Takayoshi Tanimura, “Orbital Parallel Measure Method for Measuring Deterministic Properties of Observed Time Series Data”, Journal of Japan Fuzzy Society, 1997, Vol. 9, no. 4, pp580-588

  However, since the orbital parallel measure method captures stochastic factors sharply, it is difficult to distinguish between a TPM that captures stochastic factors and a TPM that captures transitions of dynamical systems. In other words, the TPM that captures the transition of the dynamic system is buried in the TPM that captures the stochastic process factor, and it may be difficult to detect the transition of the dynamic system.

  In view of the above circumstances, an object of the present invention is to provide a technique for detecting a transition of a dynamic system by an orbital parallel measure method.

  One aspect of the time-series data analysis method of the present invention that achieves the above object is to calculate a basic period of time-series data, calculate a discrete moving average of the time-series data based on the basic period, and The time-series data is analyzed based on the processed time-series data.

  Another aspect of the time-series data analysis method of the present invention that achieves the above object is to calculate a basic period of the time-series data, calculate a discrete moving average of the time-series data based on the basic period, A trajectory parallel measure of the time series data subjected to the discrete moving average process is calculated, and the time series data is analyzed based on the calculated trajectory parallel measure.

  According to another aspect of the time-series data analysis method of the present invention that achieves the above object, in the time-series data analysis method, a moving average of the time-series data of the orbital parallel measure is calculated, and a moving average process is performed. The time-series data is analyzed based on the time-series data.

  One aspect of the time series data abnormality monitoring device of the present invention that achieves the above object calculates a basic period of time series data, calculates a discrete moving average of the time series data based on the basic period, The trajectory parallel measure of the time series data is calculated based on the time series data subjected to the discrete moving average process.

  According to another aspect of the time-series data abnormality monitoring apparatus of the present invention for achieving the above object, the time-series data abnormality monitoring apparatus calculates a moving average of the time-series data of the orbital parallel measure in the time-series data abnormality monitoring apparatus. The time-series data is analyzed based on the processed time-series data.

  According to the above invention, the transition of the dynamic system in the time series data can be detected using the orbit parallel measure method.

It is the schematic of the abnormality monitoring apparatus which concerns on 1st Embodiment of this invention. It is a figure explaining the detail of the data determination process part of the abnormality monitoring apparatus which concerns on 1st Embodiment of this invention. It is a figure which shows an example of the time series data of analysis object. It is a figure which shows the time series data which simulated the water distribution amount when the water leak has occurred, and the time series data which simulated the water distribution amount when the water leak has not occurred. (A) The figure which shows the time series data after performing a discrete moving average process with respect to the time series data of FIG. 3, (b) The figure which shows the calculation result of TPM in the time series data of FIG. 5 (a). . It is a figure which shows the calculation result (water leakage period partial enlarged view) in the time series data after a discrete moving average process. It is a figure which shows the calculation result of TPM in the time series data of FIG. (A) The figure which shows the time series data after performing the moving average process (for 6 time) with respect to the time series data of FIG. 3, (b) The calculation result of TPM in the time series data of FIG. 8 (a) FIG. (A) The figure which shows the time series data after performing the moving average process (for 12 time) with respect to the time series data of FIG. 3, (b) The calculation result of TPM in the time series data of FIG. 9 (a) FIG. It is a figure which shows the time series data (5th) of electric power supply amount. It is a figure which shows the time series data (1 day) of electric power supply amount. It is a figure which shows the result of having embedded the time series data of FIG. 10 in the attractor. It is a figure which shows the result embedded in the attractor after performing a discrete moving average process (for 5 time) with respect to the time series data of FIG. It is a figure which shows the result embedded in the attractor after performing a moving average process (for 5 time) with respect to the time series data of FIG. 10A is a diagram showing time-series data in which noise is superimposed on the time-series data in FIG. 10, FIG. 10B is a time after performing a moving average process (for five times) on the time-series data in FIG. The figure which shows series data, (c) It is a figure which shows the time series data after performing a discrete moving average process (for 5 time) with respect to the time series data of Fig.15 (a). It is a figure explaining the detail of the data determination process part of the abnormality monitoring apparatus which concerns on 2nd Embodiment of this invention. (A) The figure which shows the time series data (after discrete moving average processing) which is a measured value of the amount of water distribution at the time of water leakage, (b) The transition of TPM of the time series data shown to Fig.17 (a) is shown. FIG. It is a figure which shows the time series data which performed the moving average process with respect to the time series data of FIG.17 (b).

  A time series data analysis method, time series data abnormality monitoring apparatus, and a program for causing a computer to function as each means of the abnormality monitoring apparatus according to an embodiment of the present invention will be described in detail with reference to the drawings.

[First Embodiment]
As shown in FIG. 1, the abnormality monitoring apparatus 1 according to the first embodiment of the present invention acquires time-series data from a monitoring / control target 2 and monitors and controls the monitoring / control target 2. In addition, the output device 3 is connected to the abnormality monitoring device 1, and the determination result of the monitoring / control target 2 is output to the output device 3.

  The abnormality monitoring device 1 includes a data collection unit 4, a data storage unit 5, and a data determination processing unit 6. The data collection unit 4 detects time-series data from the monitoring / control target 2 and captures the detected time-series data. The data storage unit 5 stores the time series data captured by the data collection unit 4. The data determination processing unit 6 determines whether an abnormality has occurred in the time series data stored in the data storage unit 5.

  The monitoring / control target 2 is, for example, a rotating shaft of a rotating machine system, an electric power demand amount, or a water demand amount (water distribution amount). The abnormality monitoring device 1 is transmitted with time series data such as shaft vibration sound data and vibration data, power demand time series data or water demand time series data.

  The output device 3 is, for example, a display or a printer, and outputs the analysis result of the time series data in the abnormality monitoring device 1.

  FIG. 2 is a diagram illustrating details of the data determination processing unit 6 of the abnormality monitoring apparatus 1. With reference to FIG. 2, the function of each processing unit of the data determination processing unit 6 will be described in detail.

  The data acquisition unit 7 acquires the time series data stored in the data storage unit 5 as time series data to be determined.

  The discrete moving average processing unit 8 calculates the discrete moving average of the time series data acquired by the data acquisition unit 7 and outputs the time series data subjected to the discrete moving average. The discrete moving average is an averaging process using discrete points separated by a predetermined interval, and can be calculated based on Expression (1).

  For example, the discrete moving average processing unit 8 calculates the basic period of the time series data based on the autocorrelation of the acquired time series data, and outputs the average of the time series data using the basic period as the discrete interval d. In addition, when the basic period of the time-series data to be analyzed can be predicted in advance, the basic period can be used as the discrete interval d. More specifically, when the time-series data is power demand, the power demand varies depending on weekdays, public holidays, seasons, etc., but basically a periodic waveform (basic period) is used on a daily basis. Therefore, the discrete interval d can be set to 1 day (24 hours). In addition, the number of time-series data (number of samples m) extracted for taking a discrete moving average is appropriately set depending on the time-series data to be analyzed. For example, when the time-series data is power demand, the time-series data on weekdays and the time-series data on holidays are often different so that no difference appears between the time-series data on weekdays and the time-series data on holidays. The discrete moving average is calculated using samples for 1 to 2 weeks (that is, 7 to 14 samples when the basic period is 1 day).

The embedding processing unit 9 performs embedding processing on the time-series data output from the discrete moving average processing unit 8 in an n-dimensional state space (n is a positive integer). That is, from the time series data y (t) subjected to the discrete moving average processing, the vector X _t = (y (t), y (t−τ),..., Y (t− (n−1) τ)). (Τ is the delay time). This vector represents one point in the n-dimensional reconstruction state space R ⁿ . The dimension n and the delay time τ are values set in advance according to the target system.

The data vector selection unit 10 selects the most advanced data vector X _i representing the current time point of the data series at each time from the time series data embedded by the embedding processing unit 9. Note that the data vector selection unit 10 is not limited to selecting the most advanced data vector representing the current time point of the data series at each time described above, but one, two, before,. A data vector immediately before (n is a positive integer) may be selected.

The neighborhood vector detector 11 detects the neighborhood vector X _j in the neighborhood of the data vector X _i selected by the data vector selector 10.

The tangential direction calculation unit 12 calculates unit tangent vectors (tangential directions) T _i and T _j with respect to the trajectory of the selected data vector X _i and neighborhood vector X _j .

Parallelism evaluation unit 13, each data vector X _i calculated by the tangential calculation unit 12, the unit tangent vector T _i in X _j, parallelism based on T _j gamma _i (the unit tangent vector T _i and T _j Parallelism) is calculated. In the actual analysis, the average value of the parallelism is obtained in order to reduce the statistical error. Hereinafter, the calculated parallelism (or average value of parallelism) is referred to as a trajectory parallel measure (TPM).

  The parallelism determination unit 14 detects an abnormality in the time-series data (for example, transition of the dynamic system according to the degree of progression of the abnormality) based on the change in TPM obtained by the parallelism evaluation unit 13.

The current time point update unit 15 updates the current time point as a reference for selecting the data vector X _i each time the parallelism determination unit 14 evaluates time-series data in a data vector at a certain time. Do.

  The accumulated data update unit 16 deletes data with low contribution to abnormality detection from the data accumulation unit 5 (excludes it from the analysis target) in order to maintain the accumulated data in the data accumulation unit 5 at an appropriate amount. For example, the accumulated data update unit 16 deletes data that has passed for a set period or longer from the data accumulation unit 5. This is because old data in time series data has a low contribution to abnormality detection.

[Time-series data analysis method]
Next, a processing procedure of time series data by the abnormality monitoring apparatus 1 will be described.
<Step S1> The data collection unit 4 captures time series data. In the normal abnormality monitoring apparatus 1, the data collection unit 4 detects time-series data from the monitoring / control target 2 and captures the detected time-series data.
<Step S <b>2> The data storage unit 5 stores the time series data captured by the data collection unit 4.
<Step S3> The data acquisition unit 7 acquires time series data stored in the data storage unit 5.
<Step S4> The discrete moving average processing unit 8 calculates the discrete moving average of the time series data acquired by the data acquisition unit 7, and outputs the time series data subjected to the discrete moving average process.
<Step S5> The embedding processing unit 9 performs embedding processing in the n-dimensional state space on the time series data subjected to the discrete moving average processing output from the discrete moving average processing unit 8.
<Step S6> The data vector selection unit 10 selects the most advanced data vector representing the current time point of the data series at each time from the time series data embedded by the embedding processing unit 9.
<Step S <b>7> The neighborhood vector detection unit 11 detects a neighborhood vector in the data vector neighborhood space selected by the data vector selection unit 10.
<Step S8> The tangent direction of the trajectory of the time-series data after being embedded in the data vector selected by the data vector selection unit 10 and the neighborhood vector detected by the neighborhood vector detection unit 11 Is calculated. For the tangent direction of the selected data vector, for example, three consecutive data vectors such as the selected data vector and a data vector adjacent to the selected data vector are selected, and these three data vectors are identified as points. Thus, a circle passing through the three data vectors is calculated as a tangent in the selected data vector.
<Step S9> The parallelism evaluation unit 13 calculates a trajectory parallel measure (TPM) between the tangent direction of the trajectory in the data vector selected by the data vector selection unit 10 and the tangential direction of the trajectory in the neighborhood vector. The TPM becomes 0 as the tangential direction is aligned.
<Step S10> The parallelism determination unit 14 evaluates time-series data based on the TPM calculated by the parallelism evaluation unit 13. The parallelism determination unit 14 may also perform a conventional time series data evaluation method. For example, a threshold value is set for the TPM, and the abnormality of the time series data is determined by comparing the TPM with the threshold value.
<Step S11> Each time the current time point update unit 15 evaluates the time series data in the data vector at a certain time by the parallelism determination unit 14, the current time point of the time series data embedded by the embedding processing unit 9 Update.
<Step S12> The accumulated data updating unit 16 deletes data that has passed a set period from the time-series data accumulated in the data accumulation unit 5.

  In the processing operations of the current time point update unit 15 and the accumulated data update unit 16, in addition to the processing of step S11 and step S12, acquisition of new time series data, definition of a time series data range to be referred to, discrete moving average Processing and time-series data embedding processing are performed.

[Example 1]
The time-series data analysis method and time-series data abnormality monitoring apparatus 1 according to the embodiment of the present invention will be described in more detail with specific examples.

  In Example 1, the time-series data shown in FIG. 3 was analyzed by the orbit parallel measure method (procedure from Step S4 to Step S10). The time-series data in FIG. 3 is data simulating the water distribution amount data of the water treatment plant, and is data in which water leakage occurs at time 721.

  The water distribution amount data of the water treatment plant is time-series data having a basic period of one day as shown in FIG. It is known that when water leaks, there will be a difference in the amount of water distribution in the morning compared to when no water leaks. Therefore, when simulating the amount of water distribution when water leakage occurs, time-series data in which 3% of the maximum load is added as the amount of water leakage from 0 o'clock to 6 o'clock is added to the time-series data where water leakage does not occur. Used as series data.

  That is, the time-series data in FIG. 3 is a probabilistic process factor before time 721 to time-series data in which water leakage has not occurred (time-series data having the basic period of the time-series data shown by the solid line in FIG. 4). This is time series data on which noise is superimposed, and after time 721, it is based on stochastic factors in time series data (time series data having a basic period of time series data indicated by a dotted line in FIG. 4) in which water leakage occurs. This is time-series data with superimposed noise. Note that noise based on stochastic factors is random noise that is uniformly superimposed on time-series data, and simulates fluctuation due to changes in demand trends in actual water distribution data.

  Discrete moving average processing was performed on the time-series data of FIG. 3 (step S4). The discrete interval d was 24 time intervals (that is, the basic period = 24), and the discrete moving average process was performed with the number of samples for the past 6 times. The time series data after the discrete moving average process is shown in FIG.

  The time series data after the discrete moving average processing is embedded in the attractor (embedding dimension n = 4, delay time τ = 6, number of neighbors 5), and TPM is calculated (steps S5 to S9). The TPM calculation result is shown in FIG.

  As shown in FIG. 5 (b), a decrease in TPM was confirmed as a whole as compared with the result of Comparative Example 1 described later (when the discrete averaging process was not performed). And in the time series data of FIG.5 (b), only TPM of the water leakage generation | occurrence | production period showed the high TPM continuously. In other words, by performing discrete moving average processing, it is possible to capture the rise in TPM based on the transition of the dynamic system due to water leakage without being buried in the TPM capturing noise based on the stochastic process factor superimposed on the time series data. It was.

[Estimation of transition time of dynamical system]
As shown in FIG. 5B, the increase in TPM based on the occurrence of water leakage (transition of the dynamic system) appears after some time has elapsed since the occurrence of water leakage. This is due to taking a discrete moving average when analyzing time-series data. That is, since the time series data immediately after the occurrence of water leakage takes a discrete moving average with the time series data before the occurrence of water leakage, it is affected by the characteristics of the time series data before the occurrence of water leakage. As a result, it is considered that the characteristics of the time series data after the occurrence of water leakage cannot be captured immediately after the occurrence of water leakage.

  Therefore, the time when water leakage started was estimated based on the obtained change in TPM. Specifically, water leakage is based on the difference between the time when the TPM starts to rise significantly and the product of the discrete moving average interval (discrete interval d) and the number of data used for averaging (number of samples m). Estimated time when started. Note that the time at which the TPM starts to rise significantly is obtained by, for example, performing an envelope process with a value during the TPM rising period, linearly approximating this, and calculating the intersection of the obtained approximate curve and the time axis as the TPM. Was calculated as the time at which began to rise significantly.

  FIG. 6 shows the result of estimation of the water leakage occurrence time based on the analysis result of the time series data of FIG. Since “the time when the TPM starts to rise” calculated based on the value during the TPM rise period is 907 o'clock, the estimation is performed when the discrete interval d is 24 and the number of samples m used for averaging is 6. The value is obtained as 907− (24 × 6) = 763. Since the actual time when the water leak occurred was 721 o'clock, the time when the water leak occurred can be approximated with some errors.

[Comparative Example 1]
FIG. 7 shows the result of calculating the TPM by the same method as in Example 1 except that the discrete moving average process is not performed.

  As shown in FIG. 7, an increase in TPM was observed throughout the analyzed time series data. Similar to the analysis result of Example 1, it is considered that the TPM increased due to the occurrence of water leakage in the analysis result of Comparative Example 1. It was difficult to separate from the rise of TPM that captured the noise based on it, and it was difficult to detect the occurrence of water leakage.

[Comparative Example 2]
After performing a simple moving average process (for 6 times) on the time series data of FIG. 3, the TPM of the time series data after the simple moving average process was calculated. The time series data after the simple average process is shown in FIG. 8A, and the TPM calculation result in the time series data after the simple average process is shown in FIG. 8B.

  The simple moving average process is a method used for noise removal of general time-series data, and is a method of calculating an average value in time-series data while gradually shifting the period. The simple moving average of the time series data can be calculated by equation (2).

  As shown in FIG. 8B, it can be seen that by performing a simple moving average process for six times, it is possible to reduce an increase in TPM that captures the influence of noise based on a stochastic process factor. However, an extreme increase in TPM was observed even before the occurrence of water leakage. For this reason, it is difficult to separate the rise in TPM that captures the occurrence of water leakage from the rise in TPM that captures noise based on stochastic factors, and it is difficult to detect the occurrence of water leakage.

[Comparative Example 3]
After performing the simple moving average process (for 12 times) on the time series data of FIG. 3, the TPM of the time series data after the simple moving average process was calculated. FIG. 9A shows the time series data after the simple average process, and FIG. 9B shows the TPM calculation result in the time series data after the simple average process.

  As shown in FIG. 9A, by performing the simple moving average process, the characteristics of the original time series data (FIG. 3) have been lost from the time series data after the simple moving average process of Comparative Example 1. . In addition, as shown in FIG.9 (b), it turns out that the raise of TPM which caught the influence of the noise based on a stochastic process factor can be reduced by performing the simple moving average process for 12 time. However, there is no significant difference between the size of the TPM that captures noise based on stochastic factors and the size of the TPM that captures the occurrence of water leakage, and it is difficult to detect the occurrence of water leakage based on this. It was.

  As is clear from the comparison between Comparative Example 2 and Comparative Example 3, in order to reduce the influence of the stochastic process factor included in the time series data, when the simple moving average of the time series data is taken, the stochastic process factor ( Noise) is averaged and canceled out, so that the influence of noise can be reduced. However, if the number of samples used for the moving average process is increased, the influence of noise can be further reduced. However, if the number of samples is increased too much, characteristic information of time-series data may be impaired.

  According to the time-series data analysis method and the time-series data abnormality monitoring apparatus 1 according to the first embodiment of the present invention as described above, the influence of noise based on stochastic factors is suppressed, and the time-series data is behind. Transitions of certain dynamical systems can be detected.

  In other words, by calculating the basic period of the time-series data to be analyzed and taking the discrete moving average of the time-series data based on this basic period, the stochastic process with the basic characteristics of the time-series data retained It is possible to reduce the influence of noise based on the physical factor.

  The effect of taking the discrete moving average will be described with reference to an example of analyzing time series data of power demand shown in FIG. The time-series data in FIG. 10 is time-series data indicating changes in power demand for 5 days, and is time-series data from which noise has been removed for the sake of explanation. This time series data has a basic period (one day) as shown in FIG.

  When the time series data of FIG. 10 is reconstructed (n = 3, τ = 5), the attractor shown in FIG. 12 is drawn. Further, when the time series data obtained by performing discrete moving average processing on the time series data of FIG. 10 is reconfigured (n = 3, τ = 5), the attractor shown in FIG. 13 is drawn. This attractor has almost the same structure as the attractor that does not perform the discrete moving average (FIG. 12), and it can be seen that the basic characteristics of the original time-series data are retained. By performing the discrete moving average process, the blur width on the attractor trajectory is smaller than that of the attractor that is not performing the discrete moving average.

  On the other hand, when the time series data obtained by performing the simple moving average process on the time series data of FIG. 10 is reconfigured (n = 3, τ = 5), the attractor shown in FIG. 14 is drawn. As is apparent from FIG. 14, it can be seen that the simple moving average process results in a shape (simpler shape) different from the attractor (FIG. 12) that has not performed the simple moving average process.

  The trajectory parallel measure method captures the characteristics of time-series data from the direction of the contact vector of the trajectory of the attractor. Therefore, the characteristics of the attractor are easier to capture when the change in the attractor structure before and after the transition of the dynamic system is large. Therefore, the discrete moving average process that stores the basic information of the original dynamic system (basic characteristics of the time-series data) is considered to be a useful method for capturing the transition of the dynamic system of the time-series data. It is done.

  That is, when analyzing data in which noise is superimposed on time-series data as shown in FIG. 15A, noise is reduced as shown in FIG. 15B when simple moving average processing is performed. In addition, the basic characteristics of the original time-series data may be impaired. On the other hand, by performing the discrete moving average process, it is possible to reduce the influence of noise based on the stochastic process factor while retaining the characteristics of the original time series data as shown in FIG. .

[Second Embodiment]
The time-series data analysis method and time-series data abnormality monitoring apparatus according to the second embodiment of the present invention will be described in detail with reference to FIG. In the abnormality monitoring apparatus according to the second embodiment, after the moving average process is performed on the time series data of the TPM calculated by the parallelism evaluation unit 13, the parallelism determination unit 14 evaluates the TPM. This is different from the abnormality monitoring device 1 according to one embodiment. Therefore, the same code | symbol is attached | subjected about the structure similar to the abnormality monitoring apparatus 1 which concerns on 1st Embodiment, and detailed description is abbreviate | omitted.

  FIG. 16 is a diagram showing details of the data determination processing unit 6 ′ of the abnormality monitoring apparatus according to the second embodiment of the present invention.

  The data determination processing unit 6 ′ includes each processing unit (data acquisition unit 7, discrete moving average processing unit 8, embedding processing unit 9, data vector selection) included in the data determination processing unit 6 of the abnormality monitoring apparatus 1 according to the first embodiment. Unit 10, neighborhood vector detection unit 11, tangential direction calculation unit 12, parallelism evaluation unit 13, parallelism determination unit 14, current time point update unit 15, and accumulated data update unit 16), and moving average processing unit 17 Have.

  The moving average processing unit 17 performs a moving average process on the TPM time-series data calculated by the parallelism evaluating unit 13. The number of samples m to be subjected to the moving average process is appropriately set depending on the target time series data. For example, when analyzing data on water distribution, the time series data on weekdays and holidays tend to be different, so the number of samples m for one week (168 points) or two weeks (336 points) is used as a guide. A moving average is calculated. This is because by selecting the sample number m so as to be equal to or greater than the fundamental period, it is possible to expect the removal of the stochastic factors while leaving the characteristics of the fundamental period. That is, in the time series data of the water distribution amount, in addition to the 24-hour period (basic period), there may be a week period or a monthly period. For the purpose of reducing the influence of fluctuations in the characteristics of the week period or the monthly period. The number of samples m is selected. When there is no period other than the basic period as in a rotating machine, for example, the moving average is calculated with the number of samples m that is about five times the basic period. The time-series data of the TPM after the moving average process is transmitted to the parallelism determining unit 14, and the parallelism determining unit 14 determines the time-series data.

[Time-series data analysis method]
Next, a time-series data processing procedure performed by the abnormality monitoring apparatus according to the second embodiment will be described.

  First, the same processing procedure as <Step S1> to <Step S9> of the first embodiment is performed to calculate the TPM of the time series data subjected to the discrete moving average processing.

  Next, the moving average processing unit 17 calculates a moving average based on the obtained time-series data of TPM.

  Then, the same processing as <Step S10> to <Step 12> in the first embodiment is performed to detect the transition of the dynamic system in the time series data.

[Example 2]
An example of analyzing water distribution amount data (time-series data subjected to discrete moving average processing) of a water treatment plant simulating water leakage shown in FIG. 17A is shown, and the abnormality monitoring apparatus and time series according to the second embodiment The data analysis method will be described in more detail.

  The time-series data in FIG. 17A is time-series data in which local noise is superimposed on the time-series data shown in FIG. 3 (that is, when local noise is superimposed at time 482 and time 506). Time series data obtained by performing a discrete moving average process on (series data). That is, the time series data in FIG. 17A is data simulating time series data including noise based on stochastic factors that cannot be removed by the discrete moving average process. When the TPM transition of this time series data was obtained, the time series data of FIG. 17B was obtained.

  As shown in FIG. 17 (b), the TPM that captures the influence of local noise before the occurrence of water leakage is the highest, and it may be difficult to detect the water leakage phenomenon with a simple threshold setting. That is, even when the discrete moving average process is performed, local noise may not be completely removed, and there is a risk that a TPM rise that captures noise based on stochastic factors is detected as a transition of the dynamic system. .

  Therefore, in the abnormality monitoring apparatus according to the second embodiment, simple moving average processing (for example, for 120 hours) is performed on the obtained TPM time-series data.

  As shown in FIG. 18, in the time series data of the TPM after the simple moving average, the TPM that captures local noise is smoothed and features during the water leakage period (rising of the TPM that captures the transition of the dynamic system) are extracted. I understand that I can do it.

  According to the time-series data analysis method and time-series data abnormality monitoring apparatus according to the second embodiment of the present invention as described above, the time-series data analysis method and time-series data abnormality monitoring according to the first embodiment are described. In addition to the effects of the apparatus, noise (for example, local noise) based on stochastic factors that could not be removed in the discrete moving average process can be removed.

  As a result, it is possible to more accurately detect the TPM that captures the transition of the dynamic system in which the TPM rise is continuously observed. Further, since the influence of the TPM that captures noise based on the stochastic process factor can be further suppressed, the threshold determination accuracy when determining the threshold of the TPM that captures the transition of the dynamic system is improved.

  Note that the abnormality monitoring device according to the first embodiment or the second embodiment configured as described above and the data determination processing units 6 and 6 ′ of the abnormality monitoring device include, for example, a ROM, a RAM, a CPU, and the like. A predetermined program is read into the computer, and the CPU executes the program. Each means may be configured by executing a predetermined program on a computer, or at least a part of these processing contents may be realized by hardware.

  When the processing means in the device and the processing unit is realized by a computer, the processing contents of the functions that the device and the processing unit should have are described by a program. Then, by executing this program on the computer, the processing means in the apparatus and the processing unit is realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. Examples of the computer-readable recording medium include a recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory. Specifically, for example, a magnetic recording device such as a hard disk device, a flexible disk, a magnetic tape, a DVD (Digital Versatile Disc), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc Read Only Memory), a CD An optical disk such as -R (Recordable) / RW (ReWritable), a magneto-optical recording medium such as MO (Magneto Optical disc), or a semiconductor memory such as a flash memory can be used.

  This program is distributed by, for example, distributing portable recording media such as DVDs and CD-ROMs on which the program is recorded. Furthermore, it is good also as a structure which distribute | circulates this program by transferring this program from a server computer to another computer via a network.

  The time-series data analysis method and time-series data abnormality monitoring apparatus of the present invention are not limited to the uses described in the embodiments, and can be used for analysis of various time-series data. In particular, the discrete moving average process is an effective analysis process for time-series data having a basic period, so in addition to power demand and water demand (distribution), the amount of power generated by a photovoltaic power generation device, etc. It can be used to analyze time-series data with a basic period of days such as time-series data, and time-series data that repeats periodic operations such as time-series data such as manufacturing equipment and power system rotating equipment. . Furthermore, the abnormal state changes according to the progress of deterioration (that is, the dynamic system transitions according to the degree of progress of the abnormality), and the abnormal state is detected according to the progress of the deterioration by analyzing time series data. Can do.

  In the description of the embodiment, the simple moving average process is performed in the moving average process and the discrete moving average process. However, the moving average may be calculated using a weighted moving average, an exponential moving average, a triangular moving average, or the like. it can. However, the simple moving average processing is preferable in that the time series data is evaluated equally (that is, the time series data is evaluated while suppressing the influence on the dynamic system).

  In the time series data abnormality monitoring apparatus of the present invention, a threshold can be set for an increase in TPM that captures the transition of the dynamic system, and a warning is output when the threshold is exceeded.

DESCRIPTION OF SYMBOLS 1 ... Abnormality monitoring apparatus 2 ... Monitoring / control object 3 ... Output device 4 ... Data collection part 5 ... Data storage part 6, 6 '... Data determination processing part 7 ... Data acquisition part (acquisition means)
8 ... discrete moving average processing unit (discrete moving average calculating means)
9: Embedding processing unit (embedding means)
DESCRIPTION OF SYMBOLS 10 ... Data vector selection part 11 ... Neighboring vector detection part 12 ... Tangential direction calculating part (tangential direction calculating means)
13: Parallelism evaluation unit (parallelism calculation means)
14 ... Parallelism determination unit (time-series data evaluation means)
15 ... Current time point updating unit 16 ... Accumulated data updating unit 17 ... Moving average processing unit (moving average calculating means)

Claims (6)

A time series data analysis method for analyzing time series data executed by a computer by an orbital parallel measure method,
The computer is
Calculating a basic period of the time-series data;
Calculating a discrete moving average of the time-series data based on the basic period;
Calculating a trajectory parallel measure of time series data subjected to discrete moving average processing;
A step of analyzing the time-series data based on the calculated orbital parallel measure, and a method of analyzing the time-series data. An analysis method of time series data by an analysis device that analyzes time series data by the orbit parallel measure method,
Calculate a basic period of the time series data, calculate a discrete moving average of the time series data based on the basic period,
Time series data subjected to discrete moving average processing is embedded in an n-dimensional state space, the tangential direction of the trajectory in the data vector selected from the embedded time series data, and the time series embedded in the vicinity of the selected data vector Calculate the parallelism with the tangential direction of the orbit in the neighborhood vector that is data,
A time-series data analysis method, comprising: analyzing the time-series data based on the parallelism. The moving average of the time series data of the orbital parallel measure derived based on the parallelism is calculated, and the time series data is analyzed based on the time series data subjected to the moving average process. 3. The time-series data analysis method according to 2. An acquisition means for acquiring time-series data from an evaluation target;
A discrete moving average calculating means for calculating a basic period of the acquired time series data and calculating a discrete moving average of the time series data based on the basic period;
Embedding means for embedding time series data subjected to discrete moving average processing in an n-dimensional state space;
Tangential direction calculating means for calculating the tangential direction of the trajectory of the embedded time series data in a data vector selected from time series data embedded every predetermined time and a neighborhood vector which is time series data in the vicinity of this data vector When,
Parallelism calculating means for calculating parallelism between the tangential direction of the trajectory in the selected data vector and the tangential direction of the trajectory in the neighboring vector;
A time-series data abnormality monitoring device, comprising: time-series data evaluation means for evaluating the time-series data based on the calculated parallelism. A moving average calculating means for calculating a moving average of time-series data of the trajectory parallel measure derived based on the parallelism calculated by the parallelism calculating means;
5. The time-series data abnormality monitoring apparatus according to claim 4, wherein the time-series data evaluation unit evaluates the time-series data based on time-series data subjected to moving average processing.   A program for causing a computer to function as each means of the time series data abnormality monitoring device according to claim 4 or 5.

Images