JPH07113995B2 - Inference method and device - Google Patents

Description

The present invention relates to an inference method and apparatus, and more particularly to an inference method and apparatus suitable for plant operation support in consideration of prediction of anomaly spread.

[Conventional technology]

Conventionally, the inference device used to support the operation of a plant, as in the invention described in Japanese Patent Laid-Open No. 62-75720, makes inference based on information representing anomalous propagation, and an actual measurement value is set for each abnormal event. It has not been reflected.

[Problems to be Solved by the Invention]

In the above-mentioned conventional technology, when an abnormality occurs in the plant, the ripples are inferred from the cause of the abnormality, but the fact that the state of the plant changes momentarily reflecting the ripples of the abnormality is not considered, and the accuracy is sufficient. It wasn't. That is, only by inferring from the firing point of the abnormality based on the propagation information of the abnormality, the actual state of each event on the propagation path of the abnormality cannot be properly reflected, and accurate prediction of the abnormality spread cannot be performed.

An object of the present invention is to accurately infer a change in the plant state due to the propagation of the anomaly even when the anomaly occurs and the plant state on the transmission path is fluctuating.

[Means for Solving the Problem] The above problem is to measure the state quantity of a plant event, and infer the ripple effect that the measured state quantity has on the state quantities of other events of the plant, in the plant. The state quantity of the event is continuously measured in time series, and the reason why the change of the state quantity has a ripple effect on the state quantity of the other event that is later in time than the change of the state quantity can be inferred. This is achieved by an inference method that is continuously performed based on data for a predetermined calculation.

In addition, the state quantity of the event on the path where the plant event spreads to other events is continuously measured, and the ripple effect that the inferred plant event has on other events is the state quantity of the event on the spread path. The inference method according to claim 1 that is corrected based on

Further, a numerical value indicating the state of the plant event is deduced based on the inferred ripple effect, and the difference between the numerical value indicating the state of the plant event calculated from the measured plant state quantity and the inferred numerical value is reduced. The inference method according to claim 1 or 2 may further include a procedure for changing the data for the predetermined calculation.

The above-mentioned problem is also provided in a device that infers the effect of the state of a lower event of a plant on an upper event, a plant data storage unit that continuously stores the state quantity of a plant event, and a device connected to the storage unit. State degree calculating means for converting the state quantity into a numerical value indicating the degree of state of the event, and the state of the lower order event based on the numerical value indicating the degree of state of the lower order event provided connected to the state degree calculating means. A state degree influence reasoning means for outputting a ripple effect exerted on an upper event as a numerical value indicating the degree of state, a state degree storing means connected to the state degree calculating means for storing a numerical value indicating the state degree, and the state degree State degree synthesis means for correcting the numerical value output by the state degree influence inference means, based on the numerical value stored in the state degree storage means, and connected to the storage means and the state degree influence inference means. It is also achieved by the.

Further, a knowledge base provided to be connected to the state degree calculating means and the state degree influence inference means and for supplying data for calculation to each of the means, and provided to be connected to the state degree calculating means and the state degree influence inference means State degree error detecting means for calculating the difference between the state degree numerical value output by the state degree calculating means and the state degree numerical value output by the state degree influence inference means; and a state degree error detecting means connected to the error detecting means and the knowledge base. And a learning device that rewrites the data for the operation stored in the knowledge base so as to reduce the difference, and the inference device according to claim 4.

Further, the learning device is connected to the degree-of-state error detecting means and stores the degree of state of the plant event in time series, and the difference detected by the error detecting means connected to the history-state degree storing means is zero. Optimal distribution calculation means for calculating data for calculation, history calculation data storage means connected to the optimum distribution calculation means for storing data for calculation in time series, and history calculation data storage Means and data for arithmetic operation provided by the optimal allocation arithmetic means and output by the optimal allocation arithmetic means, and data for arithmetic operation averaged from data for arithmetic operation output by the history arithmetic data storage means. And an arithmetic data averaging means for outputting to the knowledge base.

[Action]

The state quantity of a plant event is continuously measured, and the ripple effect of the measured state quantity of an event (lower event) on another event (upper event) is continuously inferred. It is inferred that the size of the ripple effect on the higher-ranking event based on the fluctuation of the above always reflects the state quantity of the lower-ranking event.

Also, when the state quantity of a lower event affects the state quantity of a higher event through one or more intervening events, the state quantity of the event during that time is also continuously measured, and based on that value, the lower event The inference amount of the ripple effect that the above has on the upper event is corrected.

Furthermore, a procedure is provided to compare the actual state of the plant event with the inferred state, and to change the data of the inference operation so that the inferred state approaches the actual state. When returned, the data of the operation gradually approaches a value that matches the characteristics of the plant event.

〔Example〕

An embodiment of the inference apparatus to which the present invention is applied will be described below with reference to FIG. The inference apparatus 10 of the present embodiment periodically takes in a signal from a sensor indicating the state quantity of an event in each part of the plant 1 to be monitored via the process input device 2, converts it into an engineering unit, and stores it. A plant database 3, which is a plant data storage unit, a knowledge base 4 that stores know-how related to plant operation of an operator, an inference processing device 5 connected to the knowledge base 4 and the plant database 3, and the inference processing device. And an inference result storage memory 6 connected to 5. Therefore, the plant database 3 reflects the current state of the plant with a delay of the maximum fetch period, and the present apparatus performs inference processing according to the knowledge base while always reflecting the plant current value of the plant database 3. The result of the inference processing is stored in the inference result storage memory 6 and notified to the operator via the CRT display 7 connected to the inference result storage memory 6.

Next, the inference process in this device will be described with reference to FIG.
This device constantly monitors the process state based on the process state signal input to the plant database 3 via the process input device, and if any abnormality is detected in the plant state, the abnormality detection time point is set as the starting point (ignition point). ), Based on the data stored in the knowledge base 4, the spread of the abnormality that has occurred is inferred. Hereinafter, the state of the plant is expressed by an abnormality degree indicating whether the state is abnormal or not.
The inference processing device 5 includes a process abnormality degree calculating device 5A which is a state degree calculating means connected to the plant database 3, a process abnormality degree determining device 5B connected to the process abnormality degree calculating device 5A, and the process abnormality degree determining device. Anomaly degree spread inference device 5C which is a state degree spread reasoning means connected to 5B, an abnormality degree synthesis device 5D which is a state degree synthesis means connected to the abnormality degree spread reasoning device 5C, and the process abnormality degree computing device 5A The abnormality degree synthesis device 5D is provided with a process abnormality degree storage device 5E that is a state degree storage means connected to the abnormality degree synthesis device 5D, and the process abnormality degree calculation device 5A, the abnormality degree spread reasoning device 5C, and the abnormality degree synthesis device 5D, respectively. Connected to the knowledge base 4,
The abnormal degree synthesis device 5D is connected to an abnormal spread path information storage device 6 which is a memory for storing inference results.

The process abnormality degree computing device 5A uses the membership function representing the relationship between the process amount and the abnormality degree, which is stored in the knowledge base 4 as the know-how of the operator and the plant maker.
The process amount is converted into an abnormal degree normalized from 0 to 1. For example, as shown in FIG. 3 in which the horizontal axis represents the process amount Q and the vertical axis represents the abnormality degree E, when the process amount is 250 ° C., the process abnormality degree corresponding to it is stored in the knowledge base 4. 0 using the stored membership function M.
It is required to be 8 (80%). The membership function is determined based on the know-how of the operator etc. in a form that matches the characteristics of the process amount. The process abnormality degree determining device 5B determines whether the process abnormality degree calculated by the process abnormality degree calculating device 5A exceeds a threshold value for determining whether the process is abnormal. FIG. 4 shows an example of knowledge for inferring an abnormal influence. The figure shows retroactively cause for A5 becomes abnormal event is generated from the A ₄ to A _1.
Here, A _{1 to} A ₅ (hereinafter referred to as A _i ) are process amount data
It has anomaly degree α _i converted from B _i . When a certain event, for example, A ₂ is determined to be abnormal by the process abnormality degree determining device 5B, the abnormality degree influence inferring device 5C follows the route of FIG.
Infer how the anomaly of A ₂ affects the anomaly degree α ₃ of A ₃ . In this case, the abnormality degree of A ₃ inferred from the abnormality degree α _{2 of} A _{2 and} the abnormality degree of A ₃ calculated from the process amount B ₃ are synthesized by the abnormality degree synthesizing device 5D, and based on this result, A A predicted anomaly degree Z of ₃ is inferred.

FIG. 5 is a flow chart showing the flow of this processing. When the event at the final stage is inferred, the result is stored in the abnormality degree transmission route information storage device 5E and is displayed as a guide message or the like on the CRT display 7 Is displayed in.

Figures 6A-6C show the propagation of anomalies from event A ₁ to event A ₅ and how they are predicted by inference. First, at time t ₁ , the abnormality of A ₁ is detected, and the degree of abnormality of A _{2 to} A ₅ due to the spread of this abnormality is inferred. In Figure 6A, which shows the condition at time t ₁ , the anomaly at A ₁ has just occurred and is still
Since it does not spread to A _{2 and} later, the process abnormality degree α (Δ in the figure) of A _{2 to} A ₅ shows a value close to normal. On the other hand, according to the inferred abnormality degree R (◯ in the figure), the occurrence of the abnormality of A ₂ and A ₃ is predicted. The composite anomaly degree (x in the figure) is a value obtained by combining the process anomaly degree Δ indicating the current state of each event and the ripple effect due to the anomaly degree inferred in the previous stage,
This is the predicted abnormality degree Z of the stage.

When the Product 6B showing the time t _2, the abnormality is spread to A _2, the process abnormality of the A ₂ △ is higher than ○ inference error probability. The inference of the abnormality of the A _3, process error probability inference degree of abnormality A ₂ of A ₂ is synthesized, (× in the figure) Synthesis abnormality degree Z is used, the accuracy of the inference is enhanced. In the present embodiment, the maximum of the process abnormality degree and the inference abnormality degree is taken as the synthesis of the abnormality degree, but there is a method such as a weighted average of both.

In FIG. 6C showing the time t ₃ , the abnormality of A ₁ is spreading to the events after A ₃ .

Next, FIG. 7 shows an example of a mechanism in which a learning device is added to the extraordinary spread inference to more accurately predict the spread. In this example, in addition to the device shown in FIG.
And an abnormality degree error detecting device 5H which is state degree error detecting means connected to the abnormality degree influence inferring device 5C, and a learning device 5G connected to the abnormality degree error detecting device 5H and the knowledge base 4. There is. In addition to online inference, this is to correct the inference knowledge base based on the inference result and process anomaly level.In the anomaly level error detection device 5H, the process anomaly level is set when the plant abnormal phenomenon is settled. Arithmetic unit
Process anomaly degree α output by 5A and anomaly degree spread reasoning device
The difference (called an error) between the inference abnormality degrees R output by 5C is calculated, and the obtained result is sent to the learning device 5G. Learning device 5G
Changes the abnormal propagation coefficient β stored in the knowledge base 4 so as to minimize the error sent from the abnormality degree error detection device 5H. Here, the inference is performed according to the changed abnormal propagation coefficient β, the learning that the error is evaluated again is continued, and the abnormal propagation coefficient β that matches the characteristics of the process is determined.

The details of the above inference will be described below with reference to FIG. 8 by taking a thermal power plant as an example. FIG. 8 shows the causal relationship as the fault tree (FT), which is the knowledge for predicting the abnormality that occurs in the plant. For example, the event "casing thermal deformation" means that it occurs due to any of the events "large load change rate", "steam condition sudden change", and "heater temperature extraordinary". Further, in each event, a relational expression for deriving the process abnormality degree Y of the event and an upper event from the self event (here, as an expression representing the positional relationship between the events, the direction of the top of the fault tree,
That is, the abnormal ripple propagation coefficient β is defined, which indicates the degree of ripple to the left in the figure as the upper side and the opposite direction as the lower side.

The process abnormality degree Y of an event indicates a state that the event means, and is represented by a continuous amount between 0 and 1. For example, Y = 0 means that the event is abnormal to some extent 0, that is, normal, and if Y = 1, it means that the event is totally abnormal. The process amount abnormality degree of an event is obtained from a process input value according to a function (membership function) defined based on the state represented by the process amount. Reasoning (forward reasoning) is performed in the upper direction by the event group having these abnormalities,
Extent of occurrence of vibration, which is the last observation event (abnormality)
Is predicted.

In order to infer the anomalous degree of the event "casing thermal deformation" based on the anomaly levels of its subordinate event groups, "large load change rate", "sudden change in steam condition", and "heater temperature high", these event groups Is expressed as a production rule (hereinafter referred to as a rule). The degree of occurrence of "casing thermal deformation" from "large load change rate" is "large load change rate"
Assuming that 100% is transmitted from the anomaly degree of, this relationship is expressed as the following rule.

If the "load change rate is large", "casing thermal deformation" occurs.

Similarly, the degree of occurrence of "casing thermal deformation" due to "sudden change of steam condition" and the extent of occurrence of "casing thermal deformation" due to "heater temperature high" are expressed by the following rules.

If "steam condition sudden change", "casing thermal deformation" occurs.

If "heater temperature is very high", "casing thermal deformation" occurs.

Here, when a plurality of lower events spread to upper events, the degree of the spread is represented by a combine function cf. The combine function cf (x ₁ x ₂ ) is represented by cf (x ₁ , x ₂ ) = x ₁ + x ₂ −x ₁ * x ₂ when the inputs x ₁ and x ₂ are integers.
For example, if the "amount of load change" abnormality is Y ₁ = 0.3 b "steam condition sudden change" Y ₂ = 0.4 c "heater temperature extraordinary" abnormality Y ₃ = 0.5, "casing thermal deformation""Abnormality level Y ₄ is ~
It is inferred as follows from the rule of 1 and the abnormalities of a to d.

i) Rule execution Y ₄ ′ = cf (Y ₄ , Y ₁ ) = cf (0,0.3) (1) = 0.3−0 * 0.3 = 0.3 ii) Rule execution Y ₄ ″ = cf (Y ₄ ′, Y) _{2) = cf (0.3,0.4) ...} (2) = 0.3 + 0.4-0.3 * 0.4 = 0.58 iii) rule execution _{Y 4 = cf (Y 4 "} , Y 3) = cf (0.58,0.5) ... (3 ) = 0.58 + 0.5-0.58 * 0.5 = 0.79 As mentioned above, the abnormality of "Large load change rate" is 0.3, the abnormality of "Rapid change of steam condition" is 0.4, and the abnormality of "heater temperature high" is If it is 0.5, it is inferred that the extent to which "casing thermal deformation" becomes abnormal is 79% due to the spread of anomalies of these subordinate events. In addition, the above (1), (2),
In Eq. (3), the cf function contains the abnormalities Y ₄ , Y ₄ ′, Y ₄ ″ of the upper events.
Is included, but this is just in the calculation logic.

Next, the operation of the inference processing apparatus will be described in detail with reference to FIG.

(1) The process amount of each event is monitored at a certain scan speed based on an abnormality degree definition formula that defines the abnormality of each event. The process amount is based on the membership function defined in the abnormality definition formula, and the process abnormality in the range of 0 to 1 α _j
Is converted to.

(2) If the process abnormality degree α _j of each event calculated based on the abnormality degree definition formula exceeds the abnormality determination level (threshold value) TH, it is determined that an abnormal event has occurred, and the event is determined to be abnormal. With the ignition point as the starting point, anomalous influence reasoning on higher-order events is started.

(3) Propagation and propagation of anomalies from lower-level events to higher-level events often involves a time delay, which causes a delay in prediction. However, before the event actually becomes abnormal, the lower-level event group The time delay is compensated by making a prediction using the anomaly inference value. Therefore, the predictive abnormality degree Z _j based on the inference result of the event _j is the inference abnormality degree R _j that infers the abnormality spread from the lower event group and the process abnormality degree α _j defined by the process amount.
And are combined, and Z _j = cpf (α _j , R _j ) cpf: a combined function.

(4) The propagation amount X _{j of the} degree of abnormality from the event j to the upper event k is
It is calculated by multiplying the predicted abnormal degree Z _j of the event j by the abnormal ripple propagation coefficient β _j . That is, X _j = β _j * Z _j . However, when the ripple of the anomaly is settled, X _j = β
You can think of it as _j * α _j .

(5) Lower events that affect the order event k multiple (J ₁ through J _n)
In some cases, the propagation amounts X _{j1 to} X _jn of the abnormalities of the respective events are combined to form the inference abnormality R _k of the event k. R _k is represented by R _k = cf (X _j1 , X _j2 , ..., X _jn ) by using a combine function cf showing the relationship between the lower event group J ₁ , ... J _n and the upper event k. Here, the combine function cf is defined as follows according to the relationship between upper / lower events.

(I) OR: cf (x 1, x 2) = 1- (1-x 1) * (1-x 2) (ii) AND: cf (x 1, x 2) = x 1 * x 2 events k the prediction error probability Z _k of the process abnormality degree alpha _k, which is defined from the process of event k, from the inference error probability R _k,
The composite function cpf, represented by _{Z k = cpf (α k,} R k).

Propagation quantity X _k abnormal to events further higher from the event k, using the abnormal degree spread propagation coefficient beta _k, indicated by X _k = beta _k * Z _k, the order event is L, event k is k if there ₁ to k _n, the inference error probability R _L of events L uses a combined function _{cf, R L = cf (X} k1, X k2, ... X kn) represented by. Furthermore, the predictive anomaly degree Z _L of event L is the composite function
It is shown by Z _L = cpf (α _L , R _L ) using cpf.

(6) By repeating the above (3) to (5), the ripple reasoning in the fault tree is performed.

(7) To determine the abnormal spread route, the rank calculation of each spread route generated from the ignition point is performed, and the one with the higher rank value (RNK) is determined as the upper spread route. The rank calculation is performed by averaging the predictive abnormalities Z on the same propagation route.

n: the number of events on the same propagation route from the ignition point to the final stage event Next, FIG. 10 shows an example of the configuration of the learning device 5G. Connected to the abnormality degree error detection device 5H, a history abnormality degree storage device 5GA which is history state degree storage means is provided, and the history abnormality degree storage device 5GA has an optimum distribution calculation device 5 which is optimum distribution calculation means.
GB is connected. The optimal distribution arithmetic device 5GB is connected with a history propagation coefficient storage device 5GC, which is history calculation data storage means, and a distributed processing device 5GD is connected to the output side of the history propagation coefficient storage device 5GC. Further, a propagation coefficient averaging device 5GE, which is a calculation data averaging means, is provided, which is connected to the output side of the optimum distribution computing device 5GB and the history propagation coefficient storage device 5GC. The output side of the propagation coefficient averaging device 5GE is connected to the abnormal propagation route information storage device 5E, the knowledge base 4, and the history propagation coefficient storage device 5GC. In the inference device shown in FIG. 7 equipped with the learning device 5G having this configuration, the event j determined by the process abnormality degree computing device 5A
The process abnormality degree αj is transmitted to the history abnormality degree storage device 5GA in the learning device 5G via the abnormality degree error detection device 5H and is stored therein. The optimal distribution arithmetic unit 5GB consists of the time-series abnormality degree propagation coefficient β _je for each event stored in the history propagation coefficient storage device 5GC and the process abnormality degree α stored in time series in the history abnormality coefficient storage device 5GA. _It reads _j , determines the optimum intermediate propagation coefficient β _jm based on this, and outputs it to the propagation coefficient averaging device 5GE. The transmission coefficient averaging device 5GE is
When this intermediate propagation coefficient β _jm is input, the corresponding time-series anomaly degree propagation coefficient β _je is read from the history propagation coefficient storage device 5GC, and the input intermediate propagation coefficient β _jm and the time-series anomaly degree propagation coefficient β j are input. _Find the average value β _jm / 2 + β _je / 2 with _je ,
This is output to the abnormal propagation path information storage device 5E, the knowledge base 4, and the history propagation coefficient storage device 5GC as the propagation coefficient β _ja by this learning. The propagation coefficient β _ja input to the history propagation coefficient storage device 5GC is stored as the latest time-series anomaly spread propagation coefficient β _je , used for the next learning process, and sent to the knowledge base 4 for online inference. Used as data. These time-series anomaly degree spread propagation coefficient β _je
Is zero if it matches the characteristics of the process, and therefore the variance of the time series anomalous degree propagation coefficient β _je is due to an unknown quantity, and indicates the certainty of the coefficient β _je. Become. The distributed processing device 5GD obtains the degree of dispersion of the propagation coefficient β _ja input to the history propagation coefficient storage device as a dispersion value, and outputs / displays this dispersion value as an “ambiguous amount” in inference, thereby providing a higher accuracy. High inference results are presented to the operator.

The processing flow of this learning device is shown in FIG. In step 1, in the time series t ₀ , t ₁ , ... T _m at the time when the abnormal phenomenon is settled, the upper event K propagated from the lower events J _{1 to} J _{n is} propagated.
The process error degree error with the inference error degree is calculated. α
_j1 t ₀ , α _j2 t ₀ , ... α _jn t ₀ are lower events J ₁ ~ at time t ₀
In the process abnormality degree of J _n , α _k t ₀ is the process abnormality degree of the upper event K at time t ₀ . β _{j1 to} β _jn are event J ₁ to
The anomaly degree transmission coefficient β _je of J _n , and the process abnormality degree α _j1 t ₀ , ..., α in the time series t _{0 to} t _n .
_{Using j2} t ₀ ,…, α _jn t ₀ ,…, α _j1 t _n ,…, α _jn t _n ,
Reasoning anomaly degree of event K calculated by cf function and actual event K
Of time t ₀ ~t process in the _m degree of abnormality _{α k t 0, ... α k} t m,
The system of equations that makes the error of
_{It is} possible to set _{j1 to} β _jn as unknowns. α _j1 t ₀ , α _j2 t ₀ ,
…, Α _jn t ₀ ,…, α _j1 t _m ,…, α _jn t _m , and α _k t ₀ ,
... α _k t _m is a process abnormality degree and is known from time t _{0 to} t _m .

In step 2, as the solution of the simultaneous equations established in step 1, (m-n + 1) sets of anomalous degree propagation factors are obtained.

In step 3, the change amount E of each group is obtained from the (m-n + 1) group of the degree-of-abnormality propagation factors, and step 4
Anomalousness propagation coefficient (β _j1 ′,
One set of β _j2 ′, ..., β _jn ′) is obtained. This (β _j1 ′, β _j2 ′, ..., β _jn ′) is the aforementioned intermediate propagation coefficient, β _jm .

In step 5, the average value of the intermediate propagation coefficient obtained in step 4 and the anomalous degree propagation coefficient stored in the history propagation coefficient storage device 5GC in time series is calculated, and the obtained values (β _j1 , β _j2 , ..., β _jn ) are sent to the abnormality degree propagation route information storage device, the history propagation coefficient storage device, and the knowledge base 4 as the propagation coefficient β _ja by learning. The propagation coefficient β _ja obtained by the above process is used as the propagation coefficient for online inference.

〔The invention's effect〕

According to the present invention, since the actual state quantity at the time of inference of a plant event is used for inference, a highly accurate inference value that reflects the actual state of the plant is obtained and corresponds to the predicted plant state. There is an effect that makes it possible. In addition, it has a learning function and builds a knowledge base that matches the characteristics of the plant, and has the effect of increasing the reliability of inference.

[Brief description of drawings]

FIG. 1 is a conceptual diagram showing the overall configuration of an inference apparatus to which the present invention is applied, FIG. 2 is a block diagram showing the detailed configuration of the inference processing section of the inference apparatus shown in FIG. 1, and FIG. 3 is membership. Fig. 4 is a graph showing an example of a function, Fig. 4 is a flow chart conceptually showing how event anomalies spread and propagate, Fig. 5 is a flow chart showing inference processing procedures, and Figs. 6A to 6C are event abnormal spreads. -Conceptual diagram showing the state of propagation and prediction, Fig. 7 is a block diagram showing the configuration of the embodiment of the present invention having a learning function, Fig. 8 is a fault tree diagram, and Fig. 9 is an abnormality due to fault tree FIG. 10 is a flow chart showing the spread prediction, and FIG. 10 is a block diagram showing a detailed configuration example of the learning device.
The figure is a flowchart showing the calculation procedure of the learning function. 3 ... Plant data storage means (plant database), 4 ... Knowledge base, 5 ... Inference processing device, 5A ... State degree calculation means (process abnormality degree calculation device), 5C ... State degree influence inference means (abnormality influence influence inference device) , 5D ... State degree synthesis means (abnormality degree synthesis apparatus), 5E ... State degree storage means (process abnormality degree storage apparatus), 5G ... Learning apparatus, 5GA ... History state degree storage means (history abnormality degree storage apparatus), 5GB ... Optimal distribution calculation means (optimal distribution calculation device), 5GC ... History calculation data storage means (history propagation coefficient calculation device), 5GE ... Calculation data averaging means (propagation coefficient averaging device), 5H ... State degree error detection means (abnormal) Error detector).

Claims (6)

[Claims] 1. A method of measuring a state quantity of a plant event and inferring a ripple effect of the measured state quantity on a state quantity of another event of the plant, wherein the state quantity of the plant event is measured in time series. For the predetermined calculation, the inference of the ripple effect that the change of the state quantity exerts on the state quantity of the other event temporally after the change of the state quantity is performed for a predetermined calculation. An inference method characterized by being performed continuously based on data. 2. The state quantity of an event on a path where a plant event propagates to another event is continuously measured, and the inferred ripple effect that the plant event has on another event is the event on the ripple path. The inference method according to claim 1, wherein the inference method is corrected based on the state quantity. 3. A numerical value indicating the state of a plant event is deduced based on the inferred ripple effect, and the difference between the numerical value indicating the state of the plant event calculated from the measured plant state quantity and the inferred numerical value. 3. The inference method according to claim 1, further comprising a step of changing data for a predetermined operation so as to reduce 4. An apparatus for inferring the effect of a lower event state of a plant on an upper event, comprising plant data storage means for continuously storing the state quantity of a plant event, and a plant data storage means connected to the storage means. A state degree calculating means for converting a state quantity into a numerical value indicating the degree of state of an event, and a state of the lower event based on a numerical value indicating the degree of state of the lower order event, which is provided in connection with the state degree calculating means. And a state degree memory means for storing the numerical value indicating the state degree, which is connected to the state degree calculating means, and outputs the influence effect on the state degree as a numerical value indicating the degree of state. And a state degree synthesizing means which is provided in connection with the state degree influence inference means and corrects the numerical value output by the state degree influence inference means based on the numerical value stored in the state degree storage means. Inference and wherein the. 5. A knowledge base, which is provided in connection with the state degree calculation means and the state degree spread inference means, and supplies data for calculation to each of the means, and is connected to the state degree calculation means and the state degree spread inference means. State degree error detecting means for calculating the difference between the state degree numerical value output by the state degree calculating means and the state degree numerical value output by the state degree influence inference means, and connected to the error detecting means and the knowledge base. And a learning device that rewrites data for arithmetic operations stored in the knowledge base so as to reduce the difference, and the inference device according to claim 4. 6. A learning device is connected to the degree-of-state error detecting means and stores the degree of state of a plant event in time series, and the error detecting means is connected to the history-state degree storing means to detect it. Optimal distribution calculation means for calculating data for calculation in which the difference becomes zero, history calculation data storage means connected to the optimum distribution calculation means for storing data for calculation in time series, and the history Of the calculation averaged from the data for calculation output by the optimum distribution calculation means provided connected to the calculation data storage means and the optimum distribution calculation means and the data for calculation output by the history calculation data storage means 6. The inference apparatus according to claim 5, further comprising: an arithmetic data averaging unit that outputs the data for the calculation to a knowledge base.