JPH11110367A - Automatic evaluation structure/parameter adjusting device for total state evaluation of plant - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the burden of a plant operator unskilled in the setting of a fuzzy proposition by selecting and extracting an explanation parameter based on a significant explanation parameter index. SOLUTION: This device is provided with a 1st-order adjusting system (fuzzy proposition/membership function/fuzzy measure determining part) 10, 2nd-order adjusting system (G.A. type evaluating structure/parameter adjusting part) 20 and 3rd-order adjusting system (adjustment content collecting part) 30. In the 1st-order adjusting system 10, a target parameter and its total evaluation system are applied to a time sequential data base (DB) 101 of set plant data, and time sequential signal processing such as the arithmetic processing of a move average value/tendency value (change amount)} 102 is performed by a separately set signal processing parameter 111. Then, data to perform total state evaluation (evaluation object data) are separated by a threshold value corresponding to the target parameter and by considering only these separated evaluation object data, the explanation parameter is determined according to the significant explanation parameter index not to be affected by the relation between the explanation parameters in all the data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for automatically adjusting an evaluation structure and parameters in comprehensive evaluation of a plant state.

[0002]

2. Description of the Related Art Conventionally, in comprehensive evaluation of a plant state, measured value information from a plant is not monitored as it is, but signal processing such as consideration of dead time, smoothing, and calculation of a tendency value is performed. Is evaluating the state of the plant. As a method of evaluating the plant data after the signal processing, a linguistic evaluation of a plant state by a membership function / fuzzy proposition, that is, a state evaluation by a language such as “high temperature” or “flow rate value is increasing” is used. Is being done. The fuzzy proposition is knowledge about the behavior of an explanatory variable explaining the state of a plant to be recognized, and has a function of converting time-series data into a degree of fitness for each knowledge. A fuzzy proposition usually consists of a plurality of explanatory variables and their membership functions. Furthermore, the plant state converted into the degree of conformity to the linguistic evaluation by a plurality of fuzzy propositions is analyzed by a fuzzy measure / integration in consideration of the interaction between the fuzzy propositions, and the interaction between the plurality of evaluation items (synergistic and offsetting). Are integrated as one comprehensive evaluation value of the state of the plant, and then notified to the plant operator.

In the prior art, signal processing parameters such as shift of dead time, smoothing, calculation of a tendency value, and the like are fixed by a fixed setting method by determining an optimum parameter by cross-correlation analysis, or thereafter, the state of the plant is changed. Manual adjustments were made to improve the evaluation accuracy. In addition, the selection of explanatory variables and the determination of fuzzy propositions are performed directly by the plant operator or by an engineer based on information received from the plant operator, and each time, the fuzzy measure is automatically re-identified and the plant is re-identified. In order to evaluate the quality of the comprehensive evaluation of the condition, the adjustment work was performed manually to improve the evaluation accuracy.

[0004]

Conventionally, human beings have set fuzzy propositions, which has been a heavy burden on plant operators who are not accustomed to setting fuzzy propositions. Also, when the behavior of the plant is complicated, humans cannot acquire appropriate knowledge, and setting the fuzzy proposition itself may be difficult. On the other hand, the fuzzy proposition is set by selecting significant explanatory variables based on the correlation between the objective variable and the explanatory variables, and then setting a membership function that captures the behavior for those explanatory variables. However, in this method, when there are many measured values such as in a plant, it is difficult to obtain a clear correlation between the objective variable and each explanatory variable due to the intertwining of those variables, and qualitatively You are even jeopardized to get the right properties. Furthermore, it is not possible to obtain quantitative characteristics. However, it is not difficult to find a qualitative relationship for data having a large fluctuation such as an abnormal state, and the present invention focuses on this.

[0005] Therefore, a first object of the present invention is to provide a clear relationship in correlation analysis because many measured values are entangled with each other for time-series data subjected to signal processing based on set signal processing parameters. Even if it is not possible to find, properly extract explanatory variables that explain the state of the plant, determine a fuzzy proposition / membership function that converts the behavior of that variable into information on the degree of conformity to the state of the plant to be explained,
A fuzzy measure is identified based on the fuzzy proposition conformance information of the time series data and the plant state to be comprehensively evaluated, and the evaluation structure and parameters for the comprehensive state evaluation of the plant are automatically determined to comprehensively evaluate the state. Calculate accuracy. In addition, it is generally difficult for the operator to set the optimal signal processing parameters due to the complexity of the target system in the comprehensive evaluation of the plant status, and the behavior of the signal processing parameters, such as the process characteristics changing over time, is difficult. In operational systems, where adjustments must be made every time the characteristics change and evaluation structures and parameters must be adjusted continuously at regular intervals, the time series data for continuous time zones The automatic adjustment of the state comprehensive evaluation structure and parameters must be performed. Further, when selecting an explanatory variable (fuzzy proposition) from the time-series data after signal processing, it is necessary to select an optimal combination for improving accuracy. Further, it is not realistic to search all combinations of the signal processing parameters and fuzzy propositions in order to select the optimum ones from the viewpoint of the amount of calculation. Therefore, a second object of the present invention is to search and adjust an optimal signal processing parameter and an evaluation structure / parameter by adding a signal processing parameter search means capable of searching within a meaningful time, and to improve the time of the process. The idea is to adapt to changes.

[0006]

In order to solve the above-mentioned problems, the invention according to claim 1 provides an objective variable and an overall evaluation method thereof to a time-series data of a plant, and uses a signal processing parameter separately set. Performs sequential signal processing and evaluates fuzzy propositions, membership functions, and fuzzy measures that give a comprehensive evaluation of the overall state of the plant.
A primary adjustment system that identifies the parameters as well as performs the fitness calculation for each fuzzy proposition, fuzzy integration calculation, comprehensive evaluation of the state using the fuzzy integration value, and comprehensive calculation of the state evaluation accuracy, and a plurality of time-series data, A secondary adjustment system for sequentially adjusting the signal processing parameters and the evaluation structure / parameters; and a tertiary adjustment system for generating an evaluation structure / parameter with high evaluation accuracy from the evaluation structures / parameters sequentially adjusted by the secondary adjustment system. And an evaluation structure having
In the automatic parameter adjustment device, the primary adjustment system includes:
A signal processing means for calculating a moving average value and a change amount according to a set signal processing parameter with respect to a time series data of a plant; a means for separating state comprehensive evaluation target data using a threshold value for an objective variable; and an explanatory variable Means for calculating the difference between the mean of abnormal data and the average of all data and the variance of all data, and calculating the ratio thereof as an explanatory variable significant index indicating the degree of deviation of abnormal data from all data And means for selecting an explanatory variable in order from the variable having the largest explanatory variable significance index. That is, the invention according to claim 1 separates the signal processing means for the time series data of the plant and the data (evaluation target data) for performing the state comprehensive evaluation by the threshold value for the objective variable, and separates only the separated evaluation target data. In consideration of the above, comprising means for determining an explanatory variable according to an explanatory variable significance index not affected by the relationship between explanatory variables in all data, selecting an explanatory variable by the explanatory variable significance index,
The extraction variable determining means to be extracted is characterized.

According to a second aspect of the present invention, for the explanatory variable selected in the first aspect of the present invention, means for determining a membership function from the sign of the explanatory variable significance index, an average value and a maximum value of all data or Means for using the minimum value to cover the abnormal data and determining the membership function parameters so that the value of the membership function increases as the deviation of the abnormal data from all data increases, and a selected explanation. Means for determining a fuzzy proposition from a variable and the membership function. In other words, according to the second aspect of the present invention, for the selected explanatory variable, a membership function parameter for converting measured value data obtained from the plant into a degree of conformity to the state to be evaluated is defined as a significance index and a statistical index for each explanatory variable. The fuzzy proposition / membership function determining means for automatically determining a plurality of fuzzy proposition / membership functions is characterized by the means for determining based on the quantity.

According to a third aspect of the present invention, there is provided a means for converting measured value data of a plant into a degree of conformity to a state to be evaluated using the membership function and the fuzzy proposition determined in the second aspect, and a fuzzy measure integration. Means for converting the fitness of a plurality of fuzzy propositions into one fuzzy integral value;
Means for identifying a fuzzy measure based on the degree of conformity of the fuzzy proposition and the target state evaluation value signal of the plant; means for generating a state evaluation value signal of the plant from the fuzzy integration value and a threshold value for the fuzzy integration value; Means for performing correct / incorrect judgment of the state comprehensive evaluation from the actual value in the measured value data of the plant, and counting the number of correct answers in the correct / incorrect judgment result,
Means for calculating the state total evaluation accuracy. That is, a third aspect of the present invention provides a means for converting measured value data of a plant into a degree of conformity to a state to be evaluated using the membership function and the fuzzy proposition determined by the second aspect, and a plurality of means by fuzzy measure integration. Computes the fuzzy proposition goodness of fit taking into account the interaction between fuzzy propositions, and uses that value as the overall state evaluation value of the plant,
A state in which the plant state is comprehensively evaluated taking into account the explosive change in process behavior (synergistic effect) that cannot be evaluated and judged by simple weighted averaging based on that value (synergistic effect) and the effect of suppressing the evaluation level of the explanatory variables with similar behavior (cancelling effect). It is characterized by comprehensive evaluation means and plant state comprehensive evaluation accuracy calculating means for calculating a state comprehensive evaluation success rate based on plant data in a certain time zone to evaluate an evaluation structure and evaluation parameters.

According to a fourth aspect of the present invention, there is provided means for generating a set of signal processing parameters according to the first aspect by random numbers,
As a result of the signal processing using the generated signal processing parameters, the fuzzy proposition, the membership function, and the fuzzy measure (as described above) for different time series data generated by using the means of claim 1 to claim 3 A set is called an evaluation structure / parameter, and the sum of the signal processing parameters and the signal processing parameters is called an individual), and a means for determining the overall evaluation accuracy of the state, and an evaluation accuracy lower than the set threshold value A means for deleting an individual, a means for obtaining an interaction coefficient between fuzzy propositions from a fuzzy measure for an individual having an evaluation accuracy lower than a set threshold value for a remaining individual, and a minimum interaction coefficient Means for replacing the fuzzy proposition with a fuzzy proposition by the next candidate of the explanatory variable significance index in claim 1 (fuzzy proposition with the fuzzy proposition replaced) And a means for newly generating a set of signal processing parameters by random numbers for the deleted individual (the fuzzy proposition is called a mutated fuzzy proposition). Means for re-identifying the fuzzy measure using the method and recalculating the overall state evaluation accuracy of the plant; and time-series data (adjustment data) used in implementing the means of claim 1 to claim 3. Means for setting time-series data (evaluation data) for several times the time zone, and calculating the state evaluation accuracy of the individual using the evaluation data, and when the rate of improvement in the evaluation accuracy is minimal, or When the re-evaluation is repeated the set number of times, the repetition calculation is terminated, and at this time, the individual having the highest evaluation accuracy is stored as an evaluation structure / parameter. That is, the invention according to claim 4 searches for an optimal combination of a signal processing parameter and a fuzzy proposition to be optimized to improve the success rate of state determination, and a series of the invention according to claims 1 to 3 described above. Means for finding the optimal combination of signal processing parameters and fuzzy propositions with respect to the operation and state evaluation accuracy by a genetic algorithm (GA) type method, and behavior similar to the interaction coefficient between the fuzzy proposition fitness calculated from the fuzzy measure Automatically adjusts GA-type signal processing parameters and fuzzy propositions by increasing the frequency of selection of explanatory variables and fuzzy propositions that are effective for comprehensive evaluation of states, and using optimal signal processing parameters and fuzzy proposition interaction coefficients The means are characterized.

According to a fifth aspect of the present invention, in the individual deleting process according to the fourth aspect, means for correcting an operation so as not to delete the evaluation structure / parameter (an individual having the previous maximum evaluation accuracy) stored at the time of the previous adjustment; A means is provided for correcting the operation so that when the evaluation accuracy is improved by more than the set rate compared to the previous time, the operation is deleted. That is, according to a fifth aspect of the present invention, in the process of sequentially adjusting the time-series data of a plurality of time zones according to the fourth aspect,
Means for preventing the signal processing parameter / fuzzy proposition saved during the previous adjustment from being excluded from the search target at the time of this adjustment. A signal processing parameter / fuzzy proposition forgetting (over-adjustment) preventing means having a means for excluding the signal processing parameter / fuzzy proposition stored during the adjustment.

According to a sixth aspect of the present invention, in the apparatus for automatically adjusting the evaluation structure and parameters in the comprehensive evaluation of the state of the plant according to the fifth aspect, the tertiary adjustment system converts the evaluation structure and parameters using a plurality of time-series data. Means for storing each evaluation structure / parameter when continuously adjusted,
Means for calculating the state comprehensive evaluation accuracy for each time-series data for each evaluation structure / parameter; means for calculating the average value of the state comprehensive evaluation accuracy for each time-series data; Means for storing parameters, means for calculating the state comprehensive evaluation accuracy for each evaluation structure / parameter for each time-series data, means for calculating the average value of the state comprehensive evaluation accuracy for each evaluation structure / parameter, and the average A means for selecting the evaluation structure and parameters obtained when adjusting the time-series data with the minimum value, and adding a fuzzy proposition with the largest interaction coefficient of the fuzzy measure from the selected evaluation structure and parameters As a fuzzy proposition in the evaluation structure / parameter with the maximum average value of the state comprehensive evaluation accuracy. Means for generating a new combined fuzzy proposition by adding the additional fuzzy proposition to the tafuzzy proposition, means for calculating the state comprehensive evaluation accuracy of the master fuzzy proposition and the combined fuzzy proposition for all time series data, Means for making the composite fuzzy proposition a master fuzzy proposition when the state comprehensive evaluation accuracy of the composite fuzzy proposition is greater than the state comprehensive evaluation accuracy of the master fuzzy proposition. That is, the invention according to claim 6 provides the state comprehensive evaluation structure and parameters (fuzzy proposition, membership function, fuzzy measure) according to claims 4 and 5.
Means for storing the evaluation structure / parameter adjustment result based on the time-series data of each time zone at the end of the sequential adjustment of the signal processing parameters, and all the evaluation structures / parameters stored for the individually adjusted individual data. Means for re-evaluation (calculation of state total evaluation accuracy), means for selecting an evaluation structure / parameter having the maximum state total evaluation accuracy of all data as a master evaluation structure / parameter, and a means for selecting time-series data of each time zone. A means for synthesizing the evaluation structure / parameter with the lowest state comprehensive evaluation accuracy to the master evaluation structure / parameter is provided to prevent over-adjustment for each adjustment data at the time of continuous adjustment, and comprehensive state evaluation for all data It is characterized by an evaluation structure / parameter automatic adjustment result re-evaluation means for outputting a highly accurate evaluation structure / parameter.

The present invention has the following functions. In the comprehensive evaluation of the state of the plant, (1) To obtain an explanatory variable significance index that takes into account only abnormal data, even when it is difficult to extract the explanatory variable for the objective variable as a whole by correlation analysis of the time-series data of the plant. Then, when the target variable fluctuates greatly as in the case of an abnormality, for each explanatory variable, the case where the abnormal time data greatly deviates is quantified, and the explanatory variable explaining the plant abnormality is automatically set. Can be extracted.

(2) The polarity of the significance index indicating the degree of deviation of the abnormal data is used for the explanatory variable determined in the above (1), so that the form of the membership function can be changed between the abnormal data and the overall data. Can be determined in such a direction that the degree of matching also increases when the deviation of In addition, by using the average value and the maximum value or the minimum value of the explanatory variables, the parameter of the membership function is set to 0 so that the discrepancy between the abnormal data and all data is eliminated. Can be automatically determined so that the degree of conformity of the variable value that maximizes the difference between the two is set to 1.

(3) The fuzzy measure can be automatically identified by the fuzzy proposition and membership function determined by the above (1) and (2), and the overall state evaluation structure and parameters of the plant can be determined. By using the evaluation structure and parameters of the above, comprehensive evaluation of the state of the plant taking into account the interaction between fuzzy propositions to the explanatory variables is performed, and correctness / incorrectness is determined, thereby automatically determining the accuracy of the comprehensive evaluation of the state of the plant. It can be carried out.

(4) In order to improve the overall evaluation accuracy of the state of the plant, the optimal combination of the signal processing parameters for the time series data can be efficiently searched by performing the GA type search. Specifically, first, a plurality of sets of signal processing parameters are randomly generated, a state comprehensive evaluation structure based on them is determined, and evaluation accuracy is calculated. Next, depending on the magnitude of the evaluation accuracy, the respective signal processing parameters, fuzzy propositions, and fuzzy measures are subjected to improvement work such as deletion / replacement / mutation to adjust the evaluation structure / parameter adjustment / evaluation accuracy for a new individual. Is repeatedly performed. Also, by the interaction coefficient between the fuzzy proposition fitness calculated from the fuzzy measure,
The search efficiency can be further improved by excluding the unnecessary candidates from the search target.

(5) Based on certain time series data, in the overall state evaluation structure of the plant, the evaluation structure at the time of successive adjustment of parameters, and the parameter search process, the evaluation structure and parameters of the maximum evaluation accuracy at the previous adjustment are selected. By controlling the frequency, it is possible to suppress over-adjustment (over-fitting) for changes in time-series data,
A stable evaluation structure and parameters can be obtained.

(6) The most reliable evaluation structure / parameter (master fuzzy proposition) can be selected by analyzing the evaluation structure / parameter sequential adjustment result based on a plurality of time data in a certain period. Furthermore, as a result of the above analysis, fuzzy propositions (additional fuzzy propositions) can be extracted from the time series data that is most difficult to adjust and can be additionally synthesized with the master fuzzy proposition, thereby obtaining evaluation structures and parameters with higher evaluation accuracy. Can be.

As described above, the optimum state comprehensive evaluation structure and parameters for performing the comprehensive state evaluation of the plant can be automatically determined, and it is possible to dynamically respond to the temporal change of the plant characteristics.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. Here, an example of the automatic adjustment of the evaluation structure / parameter for predicting a certain plant state in the comprehensive evaluation of the state of the plant (first embodiment) and a case where a certain manipulated variable (set value) of the process is output on behalf of the operator Of automatic adjustment of evaluation structure / parameters of the second example (second embodiment)
It is assumed that

First, FIG. 1 is a conceptual diagram showing the overall configuration of the present invention which is common to both the first and second embodiments. The evaluation structure / parameter automatic adjustment apparatus of the present invention comprises a primary adjustment system (fuzzy system). Proposition, membership function, fuzzy measure determination unit) 10, secondary adjustment system (GA type evaluation structure / parameter adjustment unit) 20, and tertiary adjustment system (adjustment content summarization unit) 3
0 is provided. In the primary adjustment system 10, a time series DB (database) 101 of the set plant data is
An objective variable and its comprehensive evaluation method are given, time-series signal processing (operation processing of moving average value / trend value (change amount), etc.) 102 is performed using a separately set signal processing parameter 111, and fuzzy logic used for comprehensive evaluation of the state of the plant A proposition and membership function 112 and a fuzzy measure 113 are identified to obtain a state comprehensive evaluation structure and parameters, and a fitness calculation process 114 for each fuzzy proposition and a fuzzy integration calculation process 11
5. Perform state comprehensive evaluation processing 116 based on integral values and state comprehensive evaluation accuracy calculation processing 117. The fuzzy proposition and membership function 112 and the fuzzy measure 113 are stored in the evaluation structure / parameter DB 118, and the evaluation structure / parameter DB 118 and the signal processing parameter 11
1 is stored in the individual DB 110. Here, details of the explanatory variable determination processing 103, the membership function parameter determination processing 104, and the fuzzy measure identification processing 105 will be described later.

The secondary adjustment system 20 uses a plurality of time-series data and uses the signal processing parameters 111 and
The sequential adjustment of the fuzzy proposition / membership function 112 is automatically performed. That is, the adjustment system 20 includes a process 201 for starting automatic adjustment, a process 202 for generating a plurality of sets of the signal processing parameters 111 using random numbers, a process 203 for calculating the evaluation accuracy for each individual, and a process for determining the end of the repetition (stop) of the calculation. 204, an individual selection process based on the calculated evaluation accuracy (deletion of parameters) 205, the selection prohibition process 206 for the individual having the highest evaluation accuracy 206, the primary adjustment system 10
And a multiplication process 208 for generating a random number of a set of signal processing parameters. The sequential adjustment of the fuzzy proposition / membership function 112 in the adjustment system 20 employs a GA type optimal parameter search method.

In the tertiary adjustment system 30, the primary and secondary adjustment systems 1
From the evaluation structures / parameters successively adjusted through 0 and 20, those having the longest evaluation accuracy are extracted as master fuzzy propositions. Furthermore, for time-series data having a low evaluation accuracy in the master fuzzy proposition, additional fuzzy propositions (explanatory variables) are extracted as synergistic maximum propositions from evaluation structures and parameters obtained when the time-series data is adjusted. By synthesizing with the fuzzy proposition, evaluation structures and parameters with high evaluation accuracy for all time-series data are generated after recalculating the evaluation accuracy and confirming the synthesis effect. That is, the adjustment system 30 performs the adjustment summary start processing 301, the evaluation structure / parameter evaluation accuracy recalculation processing 30 of the evaluation structure / parameters that have been automatically adjusted in the past, and performs the acquisition of data in the time series DB 101 and the evaluation structure / parameter DB 118.
2. Master fuzzy proposition extraction processing 305, over-adjustment structure extraction and synergy effect maximum proposition (additional fuzzy proposition) extraction processing 306, master fuzzy proposition and synergy effect maximum proposition synthesis processing 307, evaluation accuracy recalculation and synthesis An effect confirmation process 308 and a stop process 309 are provided.

Next, the whole of the explanatory variable determination processing 103 and the membership function parameter determination processing 104 will be described with reference to FIG. First, sensor information of a plant is fetched, and information of a certain amount (T times) of N variables as shown in Expression 1 is temporarily stored as data 1 in a storage device. Here, xv _t represents the value of the v-th variable at time t. XO _t represents the value of the objective variable at time t.

[0024]

(Equation 1)

Next, a moving average and
The signal processing for calculating the amount of change is performed. High frequency components are removed by signal processing, and the trend value of the trend data is calculated.
Specifically, the data 1 is subjected to moving average processing (generally omitted for simplicity) in the time direction of each variable, and the change amount is calculated (calculation of the change amount of the variable value in a certain time). The result obtained by performing the above is arranged as data A as shown in Expression 2. Here, N is twice the original value. Also, the target variable xO _t is only the variable subjected to the moving average processing, and the target variable for which the amount of change is calculated is treated the same as general xv _t .

[0026]

(Equation 2)

Next, the signal-processed data is transferred to an explanatory variable determining process 103, and significant variables y1, y2,..., Yv,. Then, the determined explanatory variables are transferred to the membership function parameter determination processing 104, and the membership function parameters of the fuzzy proposition that convert the fluctuation of the explanatory variables into the degree of the abnormal sign are determined. Thus, a fuzzy proposition as shown in Expression 3 is determined.

[0028]

(Equation 3)

Here, the types of the membership functions SA and LA are determined in the membership function parameter determination processing 104. FIGS. 3A and 3B show the types and parameters of the membership functions.

Next, the details of the explanatory variable determining process 103 and the membership function parameter determining process 104 will be described with reference to FIGS. 2, in the data A which has been subjected to signal processing, the value of the objective variable xO _t is, the data of the S step before time exceeding the threshold value Th set, before an error occurred data (for convenience, the abnormal data The data A is separated from the data A, and is referred to as data B. Also, the number of cases is U. Next, for each variable xv,
The statistic of the data A at all times and the average value (m _A , m _B ) and the standard deviation σ _A of the data B immediately before the occurrence of the separated error are calculated by Equation 4.

[0031]

(Equation 4)

Next, the explanatory variable significance index I for each variable
_exp is calculated by Expression 5, and the variables are sorted in descending order of the absolute value of the explanatory variable significance index.

[0033]

I _exp = (m _B −m _A ) / σ _A

Here, the target variable itself is excluded from the target of this processing. Then, the top M variables of the rearranged variables are determined as explanatory variables, and are defined as y1, y2,..., Yv,. As a supplement, the explanatory variable significance index I according to Equation 5
_exp represents the ratio of data B, which is data immediately before the occurrence of an abnormality, to the distribution of data at all times, that is, data A, being separated.

FIG. 3 shows the shape of the membership function and the parameters [p1, p2]. That is, the membership function is a membership function shown in the figure, and is a function that outputs a value of [0, 1] according to an input value. Further, the shape of the membership function is determined by fuzzy labels LA and SA, and the bending point of the function is determined by the parameter [p
1, p2]. The above functions are referred to as membership functions.

In FIG. 4, the type of the membership function for each explanatory variable is determined as follows based on the sign of the explanatory variable significance index I _exp calculated in the explanatory variable determination processing 103: If I _exp <0: Member Ship function SA (left-up function) If I _exp > 0 Membership function LA (right-up function) Then, the parameters [p1, p
2] is determined as follows: - If the explanatory variables is not variation I _exp <For _{0 SA [min A, m A} ] ( see FIG. 5) I _exp> For 0 LA [m _A, max _A (See FIG. 6) • When the explanatory variable is a trend value (change amount): If I _exp <0, SA [m _B , 0] (see FIG. 7) If I _exp > 0, LA [0, m _B (See FIG. 8) Here, min _A is the minimum value of data A, max _A is the maximum value of data A, min _B is the minimum value of data B, and max _B is the maximum value of data B.

FIG. 9 is an operation block diagram of the fitness calculation processing 114 for each fuzzy proposition. Using the fuzzy proposition and the membership function determined as described above, for the measured value data of the signal-processed plant, the calculation target variable is selected and the value of the membership function is calculated. (Degree of conformity to the state to be evaluated).

Next, the fuzzy measure identification processing 105 will be described with reference to FIG. First, the fuzzy measure and the fuzzy integral (Choquet integral) are defined. In this embodiment, only a fuzzy measure on a finite set is handled. Therefore, it is set as in Equation 6.

[0039]

X = {1, 2,..., N}

Definition 1: Fuzzy measure A fuzzy measure is defined as in equation (7). Equation 7
Where 2 ^X is the power set of X.

[0041]

(Equation 7)

Definition 2: Fuzzy integral The function f: Choquet integral for the fuzzy measure μ of X → R is defined by equation (8). Where x ₁ , x ₂ , ...,
x _n are those which sorts the elements of X in ascending order the values of the F, is _{f (x 1) ≦ f (} x 2) ≦ ... ≦ f (x n). Equation 9 is a condition.

[0043]

(Equation 8)

[0044]

(Equation 9)

As shown in FIG. 10, the fuzzy measure identification processing 105 receives a fuzzy integral value obtained as a result of performing the fuzzy integration processing of the above-described definition 2 with the fuzzy proposition matching degree as an input, and an objective value in the time series data. Modify the fuzzy measure sequentially to minimize residuals with variables. The convergence determination is performed by comparing the above residual with a set threshold, and when the residual becomes equal to or less than the threshold, and when the number of times of fuzzy measure correction reaches the set number, it is determined by iterative correction and residual evaluation. The convergence judgment / identification ends.

FIG. 11 shows the operation of the fuzzy integral operation processing 115. In this fuzzy integral operation processing, a plurality of fuzzy propositions (eg, proposition 1 and proposition 2) which are inputs are used.
Are sorted in descending order of their values (by the suitability calculation process 114 for each proposition), a sequence of differences of the proposition suitability is obtained, and a fuzzy integral value is obtained by calculating an inner product with the fuzzy measure μ. FIG. 12 is an example of the fuzzy measure μ. This fuzzy measure is a function for the set, and the sum of the fuzzy measure of the fitness of Proposition 1 and the fuzzy measure of the fitness of Proposition 2 is smaller than the fuzzy measure of the fitness of Proposition 1 and the fitness of Proposition 2. This shows a synergistic effect when viewing Proposition 1 and Proposition 2 at the same time.

FIG. 13 shows an example of fuzzy integration.
The fuzzy integral is a linear sum operation (area calculation) in consideration of the interaction by the fuzzy measure. In the fuzzy integral by the fuzzy measure shown in FIG. Differently, it indicates that the fuzzy integral value is determined by the interaction between the explanatory variables. FIG. 14 shows a state comprehensive evaluation 116 and a state comprehensive evaluation accuracy calculation 117 based on fuzzy integral values.
3 is an operation block diagram of FIG. The state comprehensive evaluation 116 based on the fuzzy integrated value is performed by comparing the fuzzy integrated value with a threshold value for the fuzzy integrated value, and using the result as a state comprehensive evaluation signal. Further, using the target variable data (measured value data) of the time series database, the actual plant state is compared with the state comprehensive evaluation signal to determine whether the overall state evaluation is correct or not. The number of correct answers is counted from the judgment result and registered in the database as the state total evaluation accuracy.

FIG. 15 shows a configuration of the GA type signal processing parameter / fuzzy proposition automatic adjustment process, which is obtained by extracting a part of the primary adjustment system 10 and the secondary adjustment system 20 in FIG. Equivalent to. When the automatic adjustment of the signal processing parameter / fuzzy proposition is started (processing 201), a plurality of sets of signal processing parameters (part of an individual) for signal processing are generated by random numbers (processing 202). Next, the primary adjustment system 10 is operated for each generated individual, and the evaluation structure and parameters (fuzzy proposition and membership function 1
12. Obtain the fuzzy measure 113). Next, the state comprehensive evaluation accuracy is calculated to calculate the state comprehensive evaluation accuracy DB11.
7A and obtain the evaluation accuracy for each individual (processing 20
3). Next, the evaluation structure / parameter is deleted according to the value of the evaluation accuracy (process 205). This is called selection processing, and the deleted individual is called a dead individual. Next, a fuzzy proposition replacement improvement process 207 based on a fuzzy measure is performed on the individuals remaining after the selection process, thereby constructing a fuzzy proposition that is expected to improve evaluation accuracy. Next, the primary adjustment system 10 is operated again to recalculate the evaluation accuracy for the changed individual (processes 203 and 204). The series of processing up to this point is repeated a set number of times (this series of processing is called generational change and repeated is called evolution). Furthermore, the individual who died in the selection process 205 is subjected to re-calculation of the primary adjustment system operation / evaluation accuracy as a new individual by generating a combination random number of signal processing parameters. This is called a mutation process 208.

FIG. 16 shows the individual selection process 205 described above.
3 is an operation block diagram of FIG. The state total evaluation accuracy for each individual is sorted in descending order, and individuals below the threshold are deleted according to the evaluation accuracy selection threshold. This result is stored in the individual DB 110
Register again. FIG. 17 shows fuzzy proposition replacement improvement processing 2
It is an operation block diagram of 07. Individuals with low overall state evaluation accuracy are selected from living individuals who have not died.
Next, from the fuzzy measure corresponding to the selected individual, calculate the interaction coefficient (Shapley value) between the fitness of the fuzzy proposition which is the input of the fuzzy integral, and select the fuzzy proposition (explanatory variable) with the smallest interaction coefficient I do. Here, the Shapley value is a fuzzy measure μ (μ is a whole set X = ｛x ₁ , x ₂ ,
, X _n } is a set function on 2 ^X , x ₁ , x ₂ ,.
x _n respectively, the values obtained from the fuzzy proposition is fit), φ (μ) = { φ (μ) (x 1), φ (μ) (x 2), ...
.., Φ (μ) (x _n )}, and is calculated by Expression 10. Note that | · | is the number of elements in the set, on condition of Expression 11.

[0050]

(Equation 10)

[0051]

[Equation 11]

Next, the fuzzy proposition (explanatory variable) having the minimum interaction coefficient is replaced with a fuzzy proposition based on the next candidate of the explanatory variable significance index in the explanatory variable determining means (claim 1). Individual DB1 as a replaced fuzzy proposition (new fuzzy proposition) with the replaced fuzzy proposition
10 (evaluation structure / parameter DB 118).

FIG. 18 shows a calculation process 2 of evaluation accuracy for each individual.
FIG. 3 is an operation block diagram of a process No. 03 and a repetition end determination process 204. First, the calculation of the evaluation accuracy for each individual is performed by the primary adjustment system 10.
Is performed using time series data of a set multiple including the time used for automatic adjustment. Next, in the repetition end determination process 204 of the secondary adjustment system 20, when the number of repetitions reaches the upper limit of the number of adjustments, or when the evaluation accuracy improvement rate is minimized, the automatic adjustment ends. Finally, the individual with the maximum evaluation accuracy as a result of the adjustment is regarded as a sub-optimal individual, and the individual DB 11
0 (evaluation structure / parameter DB 118).

FIG. 19 is an operation block diagram of the fuzzy proposition forgetting prevention processing (selection prohibition processing 206 for individuals with the highest evaluation accuracy). In the figure, an individual (sub-optimal individual) with the highest evaluation accuracy is selected from among the state comprehensive evaluation accuracy based on the previously adjusted time-series data. Next, in the selection process 205 for individuals based on the current adjusted time-series data, a selection prohibition command is added so that the selection process is not performed for the selected individual. However, if the state total evaluation accuracy has improved by more than the set rate from the previous time due to this adjustment,
This selection prohibition processing is not performed.

FIG. 20 shows a past automatic adjustment evaluation structure / parameter evaluation accuracy recalculation processing 30 in the tertiary adjustment system 30.
2 is an operation block diagram of a master fuzzy proposition extraction process 305. FIG. In the figure, the time series DB 10 adjusted in the past
The state comprehensive evaluation accuracy of all combinations of 1 and the adjusted evaluation structure / parameter DB 108 is reevaluated and stored in the evaluation accuracy DB 117A. Next, the average value of the state comprehensive evaluation accuracy of each time-series DB is calculated for each evaluation structure / parameter. Next, a fuzzy proposition is selected as a master fuzzy proposition from the evaluation structure in which the average value of the state comprehensive evaluation accuracy is the maximum, and is temporarily stored together with the membership function.

FIG. 21 is an operation block diagram of the extraction processing of the over-adjusted structure and the extraction processing 306 of the proposition with the maximum synergistic effect in the tertiary adjustment system 30. Evaluation accuracy DB117
From A, the average value of the evaluation accuracy of each evaluation structure / parameter is calculated for each time-series DB 101. Next, an evaluation structure with the minimum evaluation accuracy average value is selected. Next, a fuzzy proposition having the greatest synergy (interaction) is selected as an additional fuzzy proposition from the selected evaluation structure, and is temporarily stored together with the membership function.

FIG. 22 is an operation block diagram of the first embodiment which is an example of the automatic adjustment of the evaluation structure / parameter for predicting a certain plant state. In the figure, an objective variable to be subjected to abnormality prediction is selected from time-series data. that time,
The abnormal state of the objective variable is defined depending on whether the value is above or below the set threshold value. Next, lag time and smoothing / trend value calculation time width candidates are set for all explanatory variables, and these are set as signal processing parameter candidates. Finally, the definition of the objective variable and the abnormal state, and the signal processing parameters are transferred to the first to third order adjustment systems 10 to 30, and the state comprehensive evaluation structure and parameters are automatically adjusted according to the above-described procedure.

FIG. 23 is an operation block diagram of the second embodiment which is an example of the automatic adjustment of the evaluation structure / parameter when outputting a certain operation amount (set value) of the process on behalf of the operator. In the figure, lag time and smoothing / trend value calculation time width candidates are set for all explanatory variables from time-series data, and these are set as signal processing parameter candidates. Next, a process operation amount by an operator or the like is set as a target for state comprehensive evaluation, and this operation amount is added to the time-series data as an objective variable. Finally, the operator's operation amount and signal processing parameters are transferred to the first to third order adjustment systems 10 to 30, and the state total evaluation structure and parameters are automatically adjusted according to the above-described procedure.

[0059]

As described above, according to the present invention, (1) even when there is a non-linear characteristic between the objective variable and each explanatory variable, an explanatory variable explaining the state to be comprehensively evaluated is selected. Can be. (2) When performing a comprehensive evaluation of the state of a plant, it is possible to automatically extract, from all time-series variables, explanatory variables for which knowledge based on the experience of the operator has been acquired in the form of a hearing. (3) Capture the behavior of the explanatory variables that explain the plant state,
It is possible to automatically determine the membership function of the fuzzy proposition to be converted to the fitness of [0, 1], and automatically evaluate the overall state of the plant and calculate the evaluation accuracy by fuzzy integration. It is possible to reduce the load involved in extracting knowledge about the plant state, which was performed by an engineer who has acquired the knowledge. (4) The search for the optimal combination of the signal processing parameters is efficiently performed by performing the GA type selection in consideration of the interaction between the fuzzy propositions. (5) In the successive adjustment of the evaluation structure / parameters at a plurality of consecutive times, over-adjustment at each time can be suppressed by considering the evaluation structure / parameter adjusted last time, and a stable evaluation structure / parameter can be obtained. it can. (6) After continuous adjustment for a certain period of time, by analyzing the adjustment contents, the evaluation accuracy is high over a long period of time, an effective master evaluation structure / parameter is obtained, and the evaluation structure that falls into a local over-adjustment state is obtained. By additionally synthesizing the parameters, it is possible to further improve the overall state evaluation accuracy over a long period of time.

[Brief description of the drawings]

FIG. 1 is a conceptual diagram showing the overall configuration of the present invention.

FIG. 2 is an operation block diagram illustrating an explanatory variable determination process.

FIG. 3 is a diagram showing types and parameters of membership functions.

FIG. 4 is an operation block diagram of a membership function parameter determination process.

FIG. 5 is an explanatory diagram of a membership function.

FIG. 6 is an explanatory diagram of a membership function.

FIG. 7 is an explanatory diagram of a membership function.

FIG. 8 is an explanatory diagram of a membership function.

FIG. 9 is an operation block diagram of a fitness calculation process for each fuzzy proposition.

FIG. 10 is an operation block diagram of a fuzzy measure identification process.

FIG. 11 is an operation block diagram of a fuzzy integration operation process.

FIG. 12 is a diagram illustrating an example of a fuzzy measure.

FIG. 13 is a diagram illustrating an example of fuzzy integration.

FIG. 14 is an operation block diagram of state comprehensive evaluation processing and evaluation accuracy calculation processing by fuzzy integration.

FIG. 15 is a diagram showing a configuration of GA type signal processing parameter / fuzzy proposition automatic adjustment means.

FIG. 16 is an operation block diagram of an individual selection process.

FIG. 17 is an operation block diagram of a fuzzy proposition replacement improvement process.

FIG. 18 is an operation block diagram of a calculation process of evaluation accuracy for each individual and a repetition end determination process.

FIG. 19 is an operation block diagram of a selection prohibition process for an individual with the highest evaluation accuracy.

FIG. 20 is an operation block diagram of past automatic adjustment evaluation structure / parameter evaluation accuracy recalculation processing and master fuzzy proposition extraction processing.

FIG. 21 is an operation block diagram of the extraction processing of the over-adjusted structure and the extraction processing of the proposition with the maximum synergistic effect.

FIG. 22 is an operation block diagram of the first embodiment of the present invention.

FIG. 23 is an operation block diagram of the second embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 Primary adjustment system 20 Secondary adjustment system 30 Tertiary adjustment system 101 Time series DB 102 Signal processing 103 Explanation variable determination processing 104 Membership function parameter determination processing 105 Fuzzy measure identification processing 110 Individual DB 111 Signal processing parameters 112 Fuzzy proposition Membership function 113 Fuzzy measure 114 Goodness-of-fit calculation processing for each fuzzy proposition 115 Fuzzy integration calculation processing 116 Comprehensive state evaluation processing based on integral values 117 Comprehensive state evaluation accuracy calculation processing 117A Comprehensive state evaluation accuracy DB 118 Evaluation structure / parameter DB 201 Automatic adjustment Start processing 202 Processing of generating a plurality of sets of signal processing parameters by random numbers 203 Calculation processing of evaluation accuracy for each individual 204 Repetition end determination (stop) processing 205 Selection processing of individual based on evaluation accuracy (parameter deletion) 206) Selection prohibition processing for individuals with the highest evaluation accuracy 207 Fuzzy proposition replacement improvement processing 208 Propagation processing by generating random numbers of signal processing parameter sets 301 Adjustment start processing 302 Evaluation accuracy of evaluation structures and parameters automatically adjusted in the past Recalculation processing 305 Extraction processing of master fuzzy proposition 306 Extraction of over-adjusted structure and extraction processing of maximum synergistic effect proposition 307 Synthesis processing of master fuzzy proposition and maximum synergy effect proposition 308 Evaluation accuracy recalculation and confirmation processing of synthesis effect 309 Stop processing

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Makoto Uekusa 1 Fuji-cho, Hino-shi, Tokyo Inside Fujifacom Control Co., Ltd.

Claims (6)

[Claims] An objective variable and a comprehensive evaluation method are given to a time series data of a plant, and a time series signal processing is performed by a signal processing parameter set separately to give a comprehensive evaluation of a total state of the plant. Identify fuzzy propositions, membership functions, fuzzy measures as evaluation structures and parameters, calculate the fitness of each fuzzy proposition,
A primary adjustment system for performing a fuzzy integral operation, a state comprehensive evaluation based on a fuzzy integral value, and a state comprehensive evaluation accuracy calculation; and sequentially adjusting the signal processing parameters and the evaluation structure / parameter as an individual using a plurality of time-series data.
An evaluation structure / parameter automatic adjustment device comprising: a secondary adjustment system; and a tertiary adjustment system that generates an evaluation structure / parameter with high evaluation accuracy from evaluation structures / parameters sequentially adjusted by the secondary adjustment system. The primary adjustment system separates the time series data of the plant from the signal processing means that calculates the moving average value and the amount of change using the set signal processing parameters, and the state evaluation data using the threshold value for the objective variable. Means to calculate the difference between the mean of abnormal data and the average of all data, and the variance of all data for each explanatory variable, and calculate the ratio between them as an explanatory variable indicating the degree to which abnormal data deviates from all data. Means for calculating as a significant index, and means for selecting an explanatory variable in order from a variable having a large explanatory variable significant index, a plant state characterized by comprising: Evaluation in case evaluation structure parameter automatic adjusting apparatus. 2. The evaluation structure / parameter automatic adjustment apparatus in the comprehensive state evaluation of a plant according to claim 1, wherein the primary adjustment system determines a membership of the selected explanatory variable from a sign of an explanatory variable significance index. Using the means to determine the function and the average value and the maximum or minimum value of all data, cover the abnormal data, and the value of the membership function increases as the deviation of the abnormal data from all data increases Means for determining a membership function parameter, and means for determining a fuzzy proposition from a selected explanatory variable and the membership function as described above. Automatic adjustment device. 3. The evaluation structure / parameter automatic adjustment apparatus in the comprehensive state evaluation of a plant according to claim 2, wherein the primary adjustment system uses a determined membership function and a fuzzy proposition.
Means for converting the measured value data of the plant into the degree of conformity to the state to be evaluated, means for converting the degree of conformity of a plurality of fuzzy propositions into one fuzzy integral value by fuzzy measure integration, degree of conformity and purpose of the fuzzy proposition Means for identifying a fuzzy measure based on a total state evaluation value signal of the plant to be performed; means for generating a total state evaluation value signal of the plant from the fuzzy integration value and a threshold value for the fuzzy integration value; Means for performing a correct / incorrect judgment of the state comprehensive evaluation from the actual value; and means for counting the number of correct answers in the result of the correct / incorrect judgment and calculating the state comprehensive evaluation accuracy. Evaluation structure / parameter automatic adjustment device. 4. The evaluation structure / parameter automatic adjustment device in the comprehensive state evaluation of a plant according to claim 3, wherein the secondary adjustment system generates a set of signal processing parameters by random numbers, and the generated signal processing. Means for determining the evaluation structure / parameters for each different time-series data generated as a result of performing the signal processing by the parameters by the signal processing means and calculating the state total evaluation accuracy; Means for deleting the individual having a low value, for the remaining individuals, means for obtaining an interaction coefficient between fuzzy propositions from an individual whose evaluation accuracy is lower than a set threshold from its fuzzy measure, Means for replacing the smallest fuzzy proposition with a fuzzy proposition by the next candidate of the explanatory variable significance index; For the individual, re-identify the fuzzy measure using the means for newly generating signal processing parameters by random numbers, the replaced fuzzy proposition, and the fuzzy proposition mutated by new generation of signal processing parameters, and the state of the plant Means for recalculating the overall evaluation accuracy; means for including the time-series data used in claim 1 as adjustment data, and setting time-series data in a time zone several times larger than the adjustment data as evaluation data Calculate the state evaluation accuracy of the individual using the evaluation data, and when the rate of improvement of the evaluation accuracy is minimal, or when re-evaluation is repeated a set number of times, this repetition calculation ends, Means for storing an individual having the highest evaluation accuracy as a sub-optimal individual, and an evaluation structure and parameters in the comprehensive state evaluation of the plant, comprising: Data automatic adjustment device. 5. The evaluation structure / parameter automatic adjustment apparatus in the comprehensive state evaluation of a plant according to claim 4, wherein the secondary adjustment system corrects an operation so as not to delete a sub-optimal individual stored at the time of the previous adjustment. And a means for correcting the operation so that when the evaluation accuracy is improved by more than a set rate compared to the previous time, the operation is corrected so as to delete the evaluation accuracy. -Automatic parameter adjustment device. 6. The evaluation structure / parameter automatic adjustment apparatus according to claim 5, wherein the tertiary adjustment system adjusts the evaluation structure / parameter continuously using a plurality of time series data. Means for storing each evaluation structure / parameter of the above, means for calculating the state total evaluation accuracy for each time-series data for each evaluation structure / parameter, and means for calculating the average value of the state total evaluation accuracy for each time-series data Means for storing the evaluation structure / parameter having the maximum average value; means for calculating the state total evaluation accuracy for each evaluation structure / parameter for each time-series data; and state total evaluation for each evaluation structure / parameter. Means for calculating the average value of accuracy, and the evaluation structure obtained when adjusting the time series data with the minimum average value Means for selecting parameters, means for storing the fuzzy proposition with the largest interaction coefficient of the fuzzy measure as an additional fuzzy proposition from the selected evaluation structure and parameters, and evaluation structure for which the average value of the state comprehensive evaluation accuracy is the largest Means for adding the additional fuzzy proposition to the master fuzzy proposition as a fuzzy proposition in the parameters to generate a new composite fuzzy proposition, and for all time-series data, the state of the master fuzzy proposition and the composite fuzzy proposition Means for calculating the overall evaluation accuracy; and means for using the composite fuzzy proposition as a master fuzzy proposition when the state overall evaluation accuracy of the composite fuzzy proposition is greater than the state overall evaluation accuracy of the master fuzzy proposition. Evaluation system / parameter automatic adjustment device for comprehensive evaluation of plant condition.