JPH04323414A - Draw-up device for time series data rule - Google Patents

Abstract

PURPOSE:To draw up a rule for surely foreseeing the occurrence of a clogging phenomenon due to a ground improving agent in a ground improving device. CONSTITUTION:In a draw-up device 1 for a time series data rule, time series numeric value data, at every data sampling time from a ground improving device 20, are converted into time series mark data via a making-mark means 4, and are stored together with data showing a clogging phenomenon in a qualitative mark memory means 5. A mark row pattern, a precursor for the clogging phenomena, is drawn from the qualitative mark memory means 5 to be stored in a pattern memory means 9 with a factor selection means 7. When a precursor false report is given by a rule having the mark row pattern, another mark row pattern of a second priority is added to the mark row pattern with a false report checking means, and mark row patterns are added successively until a false report is ungiven. This can improve a condition part for a precursor rule.

Description

[Detailed description of the invention]

0001

[Industrial Field of Application] The present invention detects the occurrence of a specific event that has a great impact on the operation of a plant, using multiple types of operational data at each predetermined data sampling time from the plant and the detected specific event. The present invention relates to a time-series data rule creation device that creates rules for making predictions based on the above, and particularly relates to a time-series data rule creation device that can create rules for accurately predicting the occurrence of the specific event described above.

0002

2. Description of the Related Art Depending on the type of plant, the occurrence of a certain event may have a significant impact on the plant's operations. For example, in a ground improvement device that strengthens the ground by mixing earth and sand with a soil improvement agent injected from an excavation shaft that excavates the ground, as described below, the soil improvement agent is sprayed at the tip of the excavation shaft during operation. Certain clogging phenomena (hereinafter referred to as clogging events) may occur. If the above-mentioned clogging event occurs during operation, it is necessary to temporarily stop the above-mentioned ground improvement device, pull out the above-mentioned excavation shaft from the middle, and perform maintenance on the clogged part of the excavation shaft. Since such maintenance requires a great deal of effort and time, it is necessary to take measures from the operational standpoint of the soil improvement equipment before the soil improvement agent becomes clogged. On the other hand, the operator of the ground improvement equipment can predict the clogging event based on various time-series operational data from the equipment up to the clogging event and data related to the occurrence of clogging events detected in the past. have some knowledge to do so. Therefore, it is conceivable to formulate the operator's knowledge into rules and use the rules to predict clogging events in the ground improvement equipment.

0003

[Problems to be Solved by the Invention] However, the above-mentioned predictive knowledge of clogging events possessed by operators is often qualitative knowledge based on experience, and the quantity and quality of this knowledge differs from person to person. On the other hand, when creating rules based on the operational data from the above-mentioned soil improvement equipment up to the clogging event, the above-mentioned operational data is output as an enumeration of chronological numerical data, so it is especially difficult to handle numerical data. It is extremely difficult for inexperienced people to formulate the above predictive knowledge into rules based on these numerical data.
A method for creating rules for reliably predicting the above-mentioned clogging event has not been established. Therefore, the first aspect of the present invention
The purpose of this invention is to provide a time-series data rule creation device that can create rules for reliably predicting the occurrence of specific plant events. By the way, even if a clogging event can be reliably predicted using the above rules, if the prediction operation takes a long time, the time to take measures to solve the clogging event will be delayed. Therefore, one method to increase the efficiency of a prediction system and shorten the time required for prediction operations is to create rules that do not have redundant content. Therefore, the second object of the present invention is to create a time series that enables the creation of rules for reliably predicting the occurrence of specific events in a plant, and that also allows the creation of rules whose contents are not redundant. An object of the present invention is to provide a data rule creation device.

0004

[Means for Solving the Problems] In order to achieve the above-mentioned first object, the main means adopted by the first invention is that data is sampled every predetermined time from a plant where a specific event occurs. In a time-series data rule creation device that creates a rule for predicting the occurrence of the specific event based on multiple types of operational data and detected specific events, the operating data is divided into a plurality of sections in advance. Qualitative symbol assigning means for assigning a qualitative symbol expressing the size of the operational data to the operational data included in any of the categories, and the qualitative symbol for each type of prioritized operational data and the detected specific event. a qualitative symbol storage means for storing the above-mentioned data at each data sampling time; and a qualitative symbol storage means for storing a symbol string pattern consisting of qualitative symbols of the highest priority operational data within a predetermined period dating back from the time when a certain specific event occurred. a symbol string extraction means for extracting from the qualitative symbols stored in the qualitative symbol storage means, and a symbol string pattern that is the same as the symbol string pattern consisting of the qualitative symbols of the operation data with the highest priority, from the qualitative symbol strings in the qualitative symbol storage means. , specific event search means for searching the qualitative symbol storage means for the specific event at the data sampling time at the time of occurrence of the specific event corresponding to the searched symbol string pattern; An additional symbol string pattern is created by adding a symbol string pattern of another type of operation data that takes into account the priority order corresponding to the above-mentioned predetermined period to a symbol string pattern that extracts a specific event different from the above. additional pattern creation means that repeats the creation of the additional symbol string pattern until the particular event is extracted by inputting it into the specific event search means in place of the symbol string pattern with the highest priority; and the specific event search means. A pattern storage means for storing a symbol string pattern when the different specific event cannot be extracted and/or an additional symbol string pattern created by the additional pattern creation means, and a symbol string pattern stored in the pattern storage means. and/or constituted as a time-series data rule creation device comprising: a rule creation means for creating a rule having a condition part made up of the above-mentioned additional symbol string pattern and a conclusion part made up of the above-mentioned certain specific event; ing. In addition, the gist of the main means adopted by the second invention in order to achieve the second object is that in addition to the main means according to the first invention,
Deleting at least one qualitative symbol from the symbol string pattern stored in the pattern storage means or the symbol string pattern of the lowest priority of the additional symbol string pattern to create a symbol deletion pattern, and converting the symbol creation pattern into the qualitative symbol string pattern. When the symbol storage means is searched and the specific event is stored at the data sampling time of the occurrence of the specific event corresponding to the searched symbol deletion pattern, the symbol string pattern of the pattern storage device or the additional symbol The present invention is configured as a time-series data rule creation device comprising pattern updating means for rewriting and updating the column pattern into the symbol deletion pattern and repeating the process from deletion to rewriting and updating at least once.

[0005]

[Operation] According to the first aspect of the invention, first, the qualitative symbol assigning means is configured to generate a plurality of qualitative symbols in advance based on a plurality of types of operational data at each predetermined data sampling time from a plant that causes a specific event and a detected specific event. A qualitative symbol expressing the size of the operation data is given to the operation data included in any of the divided divisions of the operation data. Priorities are determined in advance for the various types of operational data, and the qualitative symbols of these operational data and the detected specific events are stored in the qualitative symbol storage means at each data sampling time. The symbol string extraction means extracts a symbol string pattern consisting of qualitative symbols of the highest priority operational data within a predetermined period dating back from the time when a certain specific event occurred from the qualitative symbols stored in the qualitative symbol storage means. do. Further, the specific event retrieval means searches the qualitative symbol strings stored in the qualitative symbol storage means for the same symbol string pattern as the extracted symbol string pattern, and uses the retrieved symbol string pattern in the qualitative symbol storage means. The specific event at the data sampling time at which the corresponding specific event occurs is retrieved from the qualitative symbol storage means. Therefore, when the specific event search means extracts a specific event different from the specific event, the additional pattern creation means adds the symbol string pattern extracted from the specific event different from the specific event.
An additional symbol string pattern is created by adding a symbol string pattern of another type of operation data in consideration of the priority order corresponding to the predetermined period. Then, the additional pattern creation means replaces the additional symbol string pattern with the highest priority symbol string pattern and inputs it into the specific event search means, and repeats creating the additional symbol string pattern until the certain specific event is extracted. . In this way, the symbol string pattern when the specific event retrieval means cannot extract the different specific event and/or the additional symbol string pattern created by the additional pattern creation means are stored in the pattern storage means. . Then, a rule is created by the rule creation means, which includes a condition part consisting of the symbol string pattern and/or the additional symbol string pattern stored in the pattern storage means, and a conclusion part consisting of the certain specific event. Therefore, since the conditions of the rule are strengthened until the specific event search means extracts the specific event, the rule can accurately predict the occurrence of a specific event.

On the other hand, according to the second invention, the first invention
In the invention, for the symbol string pattern and/or additional symbol string pattern stored in the pattern storage means,
A pattern updating means deletes at least one qualitative symbol from the lowest priority symbol string pattern of each of the patterns to create a symbol deletion pattern, and retrieves the symbol deletion pattern from the qualitative symbol storage means. If the specific event is stored at the data sampling time of the specific event occurrence corresponding to the symbol deletion pattern searched at this time, the pattern updating means updates the symbol string pattern of the pattern storage means or the additional symbol. Rewrite and update the column pattern to the above symbol deletion pattern. The process from deletion to rewriting/updating is repeated at least once by the pattern updating means. Therefore, when the above-mentioned symbol string pattern or additional symbol string pattern that has been rewritten and updated by the pattern updating means is applied as a condition part of a rule for predicting a certain specific event, its pattern structure is compared with the pattern before symbol deletion. becomes smaller.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Examples embodying the present invention will be described below with reference to the accompanying drawings to provide an understanding of the present invention. It should be noted that the following examples are examples of embodying the present invention, and are not intended to limit the technical scope of the present invention. Here, FIG. 1 is a schematic configuration diagram showing a ground improvement device to which a time-series data rule creation device according to an embodiment of the present invention is applied, FIG. 2 is a block diagram of the above-mentioned time-series data rule creation device, and FIG. is a graph showing the change over time of the residual of the numerical data from the above-mentioned soil improvement device, and FIG. 4 shows the storage state of the qualitative symbols and clogging event data stored in the qualitative symbol storage means of the above-mentioned time-series data rule creation device. An explanatory diagram, FIG. 5 is an explanatory diagram showing a certain clogging case consisting of a symbol string pattern of pattern length M for all factors involved in the clogging event, and FIG. 6 is a flowchart showing the processing procedure of the above-mentioned time series data rule creation device. . As shown in FIG. 1, the ground improvement device 20 to which the time-series data rule creation device 1 according to the present embodiment is applied includes an excavation injection machine 21 that excavates the ground 28 and injects a powdered ground improvement agent; It mainly consists of an improvement agent silo 22 for temporarily storing the soil improvement agent, and a feeder 23 that supplies a predetermined amount of the soil improvement agent from the improvement agent silo 22 to the excavation injection machine 21 by pneumatic feeding. The excavation injection machine 21 has an excavation shaft 2 which is rotationally driven by a motor 26.
5, insert the excavation shaft 25 into the ground 2 while stirring the earth and sand.
8 is provided with stirring blades 27 that dig into the interior. Further, the improvement agent silo 22 is equipped with a rotating slit vane type supply foil (not shown), and the amount of the soil improvement agent supplied is determined by the rotation speed of the supply foil, the pressure of the supply air 24, etc. The above-mentioned ground improvement agent is applied so that the excavation shaft 25 is in the ground 28.
When it is extracted from the ground, it is injected together with air from the injection port near the stirring blade 27 into the ground 28, and is mixed with the surrounding earth and sand at a predetermined ratio, thereby combining with the earth and sand. As a result, improved columns 29 are formed in the ground 28. During operation, the soil improvement device 20 may become clogged with the ground improvement agent near the injection port, and the operation may have to be stopped midway. If the operation is stopped midway in this way, it takes a lot of effort and a long time to restart the operation. The clogging event as described above does not occur suddenly, but characteristic fluctuations in the pressure of the supply air 24 and the rotational speed of the supply foil occur as a precursor.

[0008] Therefore, the present inventors conducted the following in order to reliably predict the occurrence of the above-mentioned clogging event based on fluctuations in various numerical data (hereinafter referred to as factors of the clogging event) that are precursors to the above-mentioned clogging event. Time series data rule creation device 1
(hereinafter referred to as rule creation device 1). As shown in FIG. 2, the rule creation device 1 includes a ground improvement device 2
Noise removing means 2 that removes noise contained in various numerical data such as the pressure of the supply air 24 input every 50 msec from 0, the rotation speed of the supply foil in the supply machine 23, etc., and the numerical data after noise removal. The residual e(k)(
(4)) and removes the tendency of change in the above numerical data; and the trend removing means 3 which calculates the change tendency of the numerical data (see formula (4)); Qualitative symbols NB to PB (Figure 3) expressing the magnitude of the residual e(k) included in the division of
clogging event data (Y
/N), a qualitative symbol storage means 5 for storing qualitative symbols for each type of numerical data at each data sampling time, and a symbol string consisting of qualitative symbols of operational data within a predetermined period dating back from the time when the clogging event occurred. A pattern extraction means 8 for extracting patterns from the qualitative symbols stored in the qualitative symbol storage means, symbol string patterns related to all factors extracted by the pattern extraction device 8, and each qualitative character constituting these symbol string patterns. A pattern storage means 9 for storing flags corresponding to the symbols, a false alarm check means 10 for checking false alarms of a rule whose condition part is the symbol string pattern stored in the pattern storage means 9, and a false alarm check means 10 for checking the false alarm of a rule whose condition part is the symbol string pattern stored in the pattern storage means 9. A redundancy checking means 11 for checking the redundancy of contents, a factor selection means 7 for selecting the type (factor) of the numerical data of the symbol string pattern to be incorporated into the condition part of the rule, and a plurality of The rule integrating means 12 integrates the contents of the condition part of the rule into one rule. The pattern storage means 9 has a pattern storage area 9p for storing the symbol string pattern extracted by the pattern extraction means and a flag storage area 9f for storing the flags. It is provided for each factor in the storage area 9p and for each qualitative symbol of pattern length M. The noise removal means 2, the trend removal means 3, the symbolization means 4, the factor selection means 7, the pattern extraction means 8, the false alarm checking means 10, the redundancy checking means 11, and the rule integration means 12 are the rule creation device, respectively. This is realized by an operation program stored in advance in a memory included in a calculation unit (not shown).

The rule creation device 1 according to this embodiment is configured as described above. Next, a processing procedure for creating a rule for predicting a clogging event of the ground improvement agent in the ground improvement device 20 using the rule creation device 1 will be described with reference to FIG. 6 and FIGS. 1 to 5. Note that S1, S2, . . . in FIG. 6 indicate operation steps of the rule creation device 1, respectively. First, time-series numerical data x(i) (i is 50 msec
Indicates the data sampling time for each i=1, 2, 3・
) is input to the noise removal means 2 of the rule creation device 1. The noise removing means 2 extracts the numerical data x(i
) based on the moving average numerical data y(j), y(j+1), .
... is obtained at each time i using equation (1) ((1)-1, (1)-2, . . . ). However, x(i); i=1, 2, 3,... (5
0msec) y(i); i=1, 2, 3,... (n=10
) As a result, the initial numerical data x(i
) is removed. Note that instead of the moving average numerical data, an average value for each period of the 10 points may be calculated.

[0010] Subsequently, the moving average numerical data from the noise removal means 2 is inputted to the trend removal means 3 (2) to (
3) The change tendency is removed by the equation. y(j)=a(k)+b(k
)(k-j)Δt
+e(j)
...(2) However, Δt=
tj+1 −tj (N=20)k=1,
2, 3,... j=k-N+1, k-N+2,..., k
Y(j)=A(k)+B(k)(k-j)
Δt...(3)
e(k)=y(k)−Y(k)
...(4) In other words, a(k) and b(k) are respectively A( The residual e(k) obtained using equation (4) from the trend equation shown in equation (3), where k) and B(k), does not include the change trend of the moving average numerical data mentioned above. It turns out. On the other hand, in the rule creation device 1, the selection priority order of each factor xN of the N factor group X shown by equation (5) is set and input into the factor selection means 7 from the outside, for example, manually (S1). X={x1,
x2 ,...,xN}...
(5) In this case, priorities are set in order of factors that are empirically important indicators for the ground improvement device 20 in predicting the occurrence of a clogging event. For example, the first priority factor x1 is the rotation speed of the supply wheel, the second priority factor x2 is the pressure of the supply air 24, and the following are similarly set in priority order and stored in the factor selection means 7. .

[0011] Subsequently, a plurality of threshold values -m3 to +p3 for discriminating the residual e(k) into a plurality of categories are set and input into the encoding means 4 from the outside and stored (S2, FIG. 3). . Therefore, when the residual e(k) is input to the encoding means 4 (S3), the encoding means 4 converts the input residual e(k) and the threshold value into the equation (6) (( 6)-1～
(6)-7), and assign qualitative symbols NB to PB to the residual e(k) to express the magnitude of the value (
S4). e≦-m3 →NB (6)-1-m3
<e≦−m2 →NM (6)−2−
m2 <e≦−m1 →NS (6)−
3-m1 <e≦+p1 →ZE (6
)-4+p1 <e≦+p2 →PS
(6)-5+p2 <e≦+p3 →PM
(6)-6+p3 <e →P
B (6)-7 Note that the above qualitative symbol indicates that the value of the residual e(k) is NB = extremely small, NM = small,
NS=slightly small, ZE=almost no change, PS=slightly large, PM=large, PB=extremely large. That is, the symbolizing means 4 is the qualitative symbol assigning means according to the present invention. These qualitative symbols for each factor at each time j are stored in the qualitative symbol storage means 5 as time-series symbol data, as shown in FIG. Further, clogging event data (Y/N) from the clogging detector is also stored at each time j. In this case, the clogging event data is indicated by Y when a clogging event occurs and there is an input from the clogging detector, and N when there is no input. Usually, the qualitative symbols and clogging event data related to a plurality of clogging events that have occurred in the past are used for each clogging event. On the other hand, a period length M of a predetermined period extending back in time from the time when the clogging event occurred (for example, time j=11 in FIG. 4) in anticipation of the prediction period L is set and input into the pattern extraction means 8 from the outside and stored. (S5). The above period length M
is the period length set to specify the symbol string pattern in the time series direction consisting of qualitative symbols for each factor x1 to xN within this period, and is the pattern length M of the symbol string pattern.
represents. In the device of this embodiment, a rule for predicting the jam event is created based on the symbol string pattern within the pattern length M. FIG. 5 shows a blockage case P consisting of symbol string patterns of all factors within the pattern length M. The above-mentioned clogging cases P are created for each of a plurality of clogging events, and are each created in the pattern storage area 9 of the pattern storage means 9.
p for each case. Then, the factor extraction count K of the counter of the factor selection means 7 is initialized to a value of 1 (S6).

First, the factor selection means 7 selects the factor x1 having the first priority according to the set priority order (S7). Next, the pattern extraction means 8 extracts symbol string patterns p1,1 (x1), p1,2 (
x1 ), .
). P1 = (p1,1
OR p1,2 OR...)...
(7) The above symbol string pattern is a so-called matching pattern used as a condition part of a rule. In this case, the pattern extraction means 8 extracts the symbol string pattern p1, 1 from the symbol string pattern p1,1.
The matching pattern group at this time is extracted from the pattern storage area 9p by setting flags in the flag storage area 9f corresponding to the qualitative symbols (x1), p1, 2(x1), . . . That is, the pattern extracting means 8 and the flag in the flag storage area 9f are the symbol string extracting means according to the present invention. Subsequently, the first matching pattern group P1 is checked for occurrence of false alarms by the false alarm checking means 10 (S10). In this case, a false alarm means, for example, when predicting the clogging event using a rule whose condition part is the symbol string pattern for each case constituting the first matching pattern group P1, the false alarm actually occurs in the ground improvement device 20. This means that the occurrence of the above-mentioned clogging event is predicted even though no clogging event has occurred. Therefore, the false alarm checking means 10 searches the qualitative symbol strings in the qualitative symbol storage means 5 for the symbol string pattern of the factor x1 that is the same as each symbol string pattern of the first matching pattern group P1. If the same symbol string pattern is retrieved, the blockage event data Y (a certain specific event) is retrieved from the qualitative symbol storage means 5. That is, the false alarm check means 10 is the specific event search means according to the present invention. If the clogging event data at the data sampling time is Y (S11, N
O) It is considered that the rules that incorporate the searched symbol string pattern do not generate false alarms for predicting blockage events. Therefore, the matching pattern group P1 is subsequently led to the redundancy elimination step (S30 onwards).

On the other hand, when the clog event data is N (different specific event) (S11, YES), the same symbol string pattern as each symbol string pattern of the first matching pattern group P1 is stored in the qualitative symbol storage means 5. However, depending on this symbol string pattern, the above-mentioned false alarm will be issued. Therefore, the conditions for rules consisting of such symbol string patterns need to be strengthened. Therefore, the false alarm checking means 10 checks all the symbol string patterns pK,i, pK,j, which will generate the above-mentioned false alarm.
... (hereinafter referred to as a false alarm pattern. The subscript i indicates the i-th clogging case, and j indicates the j-th clogging case) from the first matching pattern group P1 (S20).
, based on these false alarm patterns, the Kth (K=1)th false alarm pattern EK is determined as shown in equation (8) (S21)
. EK = {pK,i OR PK,j
OR...}...(8) Then, if there is a factor with the next highest priority (S22, YES), the factor selection means 7
increases the number of factor extractions K to K+1 (S23) and selects the next-ranked factor xK+1, for example x2 (S24)
). Then, a symbol string pattern with a factor xK+1 corresponding to the period length M of the symbol string pattern that causes the false alarm is extracted from the case in the pattern storage area 9p (S25). The symbol string pattern of the above factor xK+1 is added to the symbol string pattern of factor xK (AND condition)
additional patterns are created. For example, the additional pattern for the above-mentioned false alarm patterns pK,i, pK,j, . . . is expressed as equation (9) ((9)-1, . . .). In addition, in equation (9), for example, the symbol string pattern pK+
For subscripts 1,i,1, K+1 has priority K+
It is the first, i is the jam case number, and 1 indicates the identification number when there is more than one symbol string pattern in the case. Therefore, the pattern extraction means 8
The additional symbol string pattern created by and step S1
From the symbol string patterns determined not to generate false alarms in step 1 (these patterns are, for example, pK,s, pK,t), the Kth+1st matching pattern group pK+1 is created.
is determined as in equation (10). (S26). PK+1 = (pK,s O
R pK,t OR...
OR (pK,i AND
pK+1,i,1)
OR (pK,i AND pK+
1, i, 2)
OR...
OR (pK,j AND pK+1,j,1
) OR・
・・）
...(10) PK+1
=(pK+1,1 OR pK+1,2
OR pK+1,
3...)
...(11) Equation (11) is obtained by organizing and integrating the subscripts of each symbol string pattern on the right side of Equation (10) into a new subscript. The above K+1st matching pattern group PK+1 is subjected to the same false alarm check process as above in step S10, and until it is determined that the matching pattern group at that time will not issue a false alarm (S11, NO),
The above steps S10, S11, S20 to S26 are repeated. As a result, a symbol string pattern of the next factor is added to the false alarm pattern in the matching pattern group each time the repeating step is executed, taking priority into consideration. Therefore, when it is determined in step S11 that no false alarm will occur, the matching pattern group at this time includes all the minimum factors necessary for predicting the clogging event. That is, the factor selection means 7, the pattern extraction means 8, and the flag in the flag storage area 9f are the additional pattern creation means according to the present invention. Note that if the next factor is not found during the execution of the above repeating step even though it is still necessary to incorporate the next factor into the matching pattern group (S22, NO), the currently set above threshold value is It is considered that the rule conditions cannot be strengthened to prevent further false alarms under the above pattern length setting conditions. Therefore, the processing step returns to step S2, and the stratification conditions for the residual e(k) in the symbolizing means 4, ie, the threshold value and the pattern length in the pattern extracting means 8, are manually changed and set (S2, S5).

As mentioned above, the flag storage area 9f
The final matching pattern group PF that does not cause a false alarm of clogging event prediction is determined by the flag of
is followed by pattern length redundancy elimination step (S3
0 onwards). The above matching pattern group PF
An example of each symbol string pattern (including the additional symbol string pattern) is shown in equation (12).

[Equation 1] Note that the symbol element Z in equation (12) is the above qualitative symbol NB~PB
shows. Therefore, in step S30, the redundancy check means 11 temporarily deletes the last symbol element Zr,M from the symbol string pattern of the lowest priority factor xr of a certain symbol string pattern pF,i. The above symbol element Zr,
M is temporarily deleted by defeating the corresponding flag in the flag storage area 9f. Furthermore, redundancy check means 1
1 searches the qualitative symbol storage means 5 for a symbol string pattern that is the same as the symbol deletion pattern, and the jam event data Y is stored at the data sampling time at the time of occurrence of the jam event corresponding to the retrieved symbol deletion pattern. Check to see if there is one. That is, a false alarm check process similar to step S10 above is executed. Therefore, if the above-mentioned clogging event data Y is not stored (S31, Y
ES), the redundancy check means 11 checks the last symbol element Z at this point in the factor Xr of the symbol deletion pattern.
A symbol deletion pattern is created by further deleting r, M-1. The operation of deleting the symbol elements from the end of the symbol string pattern of the factor xr is sequentially repeated until the jam event data Y is extracted. Then, when the above-mentioned clogging event data Y is extracted (S31, NO), the matching pattern group PF stored in the pattern storage area 9p
The certain symbol string pattern pF,i is rewritten and updated to the symbol deletion pattern at this time (S32). The processing steps (S30 to S32) from the creation of the symbol deletion pattern to the rewriting update described above are also executed for each of the remaining symbol string patterns of the matching pattern group PF. That is, the flag in the flag storage area 9f, the false alarm check means 10, and the redundancy check means 11
is the pattern updating means referred to in the present invention. This time, on the contrary,
The processing operations of sequentially deleting, checking for false alarms, and rewriting and updating the symbol deletion pattern are performed from the foremost symbol element Zr,1 of the rewritten and updated symbol deletion pattern factor xr toward the last part (S33 to S35). . These processing operations are sequentially executed for each symbol string pattern of the matching pattern group PF in the same manner as steps S30 to S32 above. Note that the deletion start position of the symbol element to be deleted from the symbol string pattern is not limited to the first or last position, but may be, for example, any position.

In this way, the rule creation device 1 incorporates the minimum necessary factors into the symbol string pattern of the matching pattern group PF, while preventing false alarms from occurring from the lowest priority factors of the symbol string pattern. Since the symbol element Z can be deleted within the range, redundancy in the condition part of the rule using this matching pattern group PF can be eliminated. As a result, for example, the calculation speed of rule inference can be improved. On the other hand, the rule integration means 12 determines the level of the symbol element at a certain sampling time (qualitative symbol NP~
PB) differ by only one rank and all other remaining symbol elements are the same, these symbol string patterns are integrated into one (S36). For example, two symbol string patterns p1,1 (
x1 ), p1,2 (x2 ), when the levels of the symbol elements at the top are PB and PM, respectively, with a level difference of 1 rank, and all other remaining symbol elements are the same, rule integration The means 12 integrates the level of the symbol element at the beginning as a common PBM into one new symbol string pattern as shown in equation (15), and rewrites and updates the pattern storage area 9p. This further eliminates redundancy in the matching pattern group PF. p1,1 (x1)={PB,
......} ...(13)
p1,2 (x1)={PM,・
...} ...(14)
p1,1 (x1)={PBM,・
...} ...(15) In this way, the matching pattern group PF finally rewritten and updated by the flag in the flag storage area 9f is used by the above-mentioned calculation unit as a rule for predicting a clogging event in the ground improvement device 20. The blockage event data Y is used as the conclusion part of the rule. That is, the rule of the present invention in which the arithmetic unit creates a rule having a condition part consisting of the symbol string pattern and/or the additional symbol string pattern stored in the pattern storage means and a conclusion part consisting of the certain specific event. It is a means of creation.

As described above, the rule creation device 1 of this embodiment converts the numerical data at each data sampling time from the ground improvement device 20 into qualitative symbols, and then converts the numerical data from the ground improvement device 20 into time-series symbols consisting of these qualitative symbols. It is possible to extract row patterns that are related to the clogging event of the ground improvement device 20. This symbol string pattern is easy to recognize as a pattern even for people who are unfamiliar with handling numerical data. Therefore, based on the extracted symbol string pattern, rules for clogging prediction can be easily created even manually. On the other hand, if a false alarm occurs in the symbol string pattern used in the condition part of the rule, another symbol string pattern is added to the symbol string pattern in consideration of priority until the false alarm no longer occurs. This makes the conditions of the rule stricter, so when the rule is used, the clogging event can be predicted with extremely high accuracy. In addition, the rule creation device 1 deletes the symbol elements of the symbol string pattern related to the lowest priority factor within the range that does not cause false alarms, or integrates extremely similar symbol string patterns into one. Therefore, the contents of the rule condition part are no longer redundant. Although the rule creation device 1 of this embodiment was applied to the clogging of the soil improvement agent in the excavation shaft 25 of the soil improvement device 20, the rule creation device 1 is not limited to this, and is applied to cases that may have a large impact on a certain plant. This is suitable for predicting specific events with high accuracy based on time series data from the plant.

[0017]

[Effects of the Invention] The first invention is constructed as described above. Therefore, it is possible to create rules for predicting the occurrence of specific plant events with extremely high accuracy. Moreover, since the second invention is configured as described above,
It is possible to create a rule for predicting the occurrence of the specific event in the plant with extremely high accuracy, and the content of the condition part is not redundant.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram showing a ground improvement device to which a time-series data rule creation device according to an embodiment of the present invention is applied.

FIG. 2 is a block configuration diagram of the time-series data rule creation device.

FIG. 3 is a graph diagram showing changes over time in the residual of numerical data from the ground improvement device.

FIG. 4 is an explanatory diagram showing the storage state of qualitative symbols and clogging event data stored in the qualitative symbol storage means of the time-series data rule creation device.

FIG. 5 is an explanatory diagram showing a case of a blockage consisting of symbol string patterns of pattern length M for all factors involved in the blockage event.

FIG. 6 is a flowchart showing the processing procedure of the time-series data rule creation device.

[Explanation of symbols]

1... Time series data rule creation device 4... Symbolization means (qualitative symbol assigning means) 5... Qualitative symbol storage means 8... Pattern extraction means (symbol string extraction means) 9... Pattern storage means 9p... Pattern storage area 9f... Flag storage Area 10... False alarm checking means 11... Redundancy checking means 20... Ground improvement device

Claims (2)

[Claims] [Claim 1] Time-series data for creating a rule for predicting the occurrence of the specific event based on a plurality of types of operational data at each predetermined data sampling time from a plant that causes the specific event and the detected specific event. In the rule creation device, a qualitative symbol assigning means for assigning a qualitative symbol expressing the size of the operational data to operational data included in any of the divisions of the operational data divided into a plurality of divisions in advance, and a priority order. Qualitative symbol storage means for storing the qualitative symbols of various operational data defined in the above and the detected specific event for each data sampling time, and the highest priority within a predetermined period retroactive from the time when a certain specific event occurred. a symbol string extracting means for extracting a symbol string pattern consisting of qualitative symbols of the operational data of the ranking from the qualitative symbols stored in the qualitative symbol storage means; A symbol string pattern that is the same as a symbol string pattern consisting of qualitative symbols of data is searched, and the specific event at the data sampling time at the time of occurrence of the specific event corresponding to the searched symbol string pattern is stored in the qualitative symbol storage means. A specific event search means to be searched, and a symbol string pattern of another type of operation data that takes into account the priority order corresponding to the predetermined period, based on the symbol string pattern from which the specific event search means has extracted a specific event different from the certain specific event. is added to create an additional symbol string pattern, and the additional symbol string pattern is input to the specific event search means in place of the symbol string pattern with the highest priority, and the additional symbol string is used until the specific event is extracted. Additional pattern creation means that repeatedly creates a sequence pattern, and a symbol string pattern when the specific event search means cannot extract the different specific event and/or an additional symbol string pattern created by the additional pattern creation means. a rule creation means for creating a rule comprising a pattern storage means for storing, a condition part consisting of the symbol string pattern and/or the additional symbol string pattern stored in the pattern storage means, and a conclusion part consisting of the certain specific event. A time series data rule creation device comprising: [Claim 2] Time-series data for creating a rule for predicting the occurrence of the specific event based on a plurality of types of operational data at each predetermined data sampling time from a plant that causes the specific event and the detected specific event. In the rule creation device, a qualitative symbol assigning means for assigning a qualitative symbol expressing the size of the operational data to operational data included in any of the divisions of the operational data divided into a plurality of divisions in advance, and a priority order. Qualitative symbol storage means for storing the qualitative symbols of various operational data defined in the above and the detected specific event for each data sampling time, and the highest priority within a predetermined period retroactive from the time when a certain specific event occurred. a symbol string extracting means for extracting a symbol string pattern consisting of qualitative symbols of the operational data of the ranking from the qualitative symbols stored in the qualitative symbol storage means; A symbol string pattern that is the same as a symbol string pattern consisting of qualitative symbols of data is searched, and the specific event at the data sampling time at the time of occurrence of the specific event corresponding to the searched symbol string pattern is stored in the qualitative symbol storage means. A specific event search means to be searched, and a symbol string pattern of another type of operation data that takes into account the priority order corresponding to the predetermined period, based on the symbol string pattern from which the specific event search means has extracted a specific event different from the certain specific event. is added to create an additional symbol string pattern, and the additional symbol string pattern is input to the specific event search means in place of the symbol string pattern with the highest priority, and the additional symbol string is used until the specific event is extracted. Additional pattern creation means that repeatedly creates a sequence pattern, and a symbol string pattern when the specific event search means cannot extract the different specific event and/or an additional symbol string pattern created by the additional pattern creation means. creating a symbol deletion pattern by deleting at least one qualitative symbol from the symbol string pattern stored in the storing pattern storage means and the symbol string pattern of the lowest priority of the symbol string pattern or the additional symbol string pattern; The symbol deletion pattern is retrieved from the qualitative symbol storage means, and when the specific event is stored at the data sampling time of the specific event occurrence corresponding to the retrieved symbol deletion pattern, the pattern storage means is pattern updating means for rewriting and updating the symbol string pattern or the additional symbol string pattern into the symbol deletion pattern and repeating the processing from the deletion to the rewriting and updating at least once; and the symbol string pattern and the symbol string pattern stored in the pattern storage means. A time-series data rule creation device comprising: a rule creation means for creating a rule having a condition part made up of the additional symbol string pattern and a conclusion part made up of the certain specific event.