JP2024029816A - State estimation device, state estimation method and program - Google Patents

Abstract

The present invention provides a state estimation device, a state estimation method, and a program that can accurately estimate a target variable and analyze the learning status of a database. A state estimation device 1 includes a database creation unit 111 that creates a database DB that stores a plurality of data sets including an explanatory variable x representing the state of an estimation target P and an objective variable y, and a database DB that trains the database DB. The learning unit 112 includes a learning unit 112 and an estimating unit 114 that estimates an objective variable y corresponding to an explanatory variable x, which is a measured value of the estimation target P, based on the learned database DB. The database creation unit 111 selects a variable with high importance among measurable state variables as the explanatory variable x, and the learning unit 112 creates an estimation database using the initial data set before learning and the learning data set. The estimation data set of the estimation database that is created is a data set that includes the data of the initial data set and the amount of data correction by learning. [Selection diagram] Figure 2

Description

The present invention relates to a state estimation device, a state estimation method, and a program.

Conventionally, state estimation methods have been developed that estimate the value of a predetermined objective variable (output variable) based on the value of an explanatory variable (input variable) measured as a parameter representing the state of an object. This state estimation method is called a soft sensor, and it uses physical quantities (measurements) to measure parameters that are difficult to measure, such as various parameters in a plant in a high-temperature, high-pressure environment, product quality, etc. It is used when estimating using (value).

For example, as in Patent Document 1, a state estimation method has been developed that estimates an output value based on an input value of a soft sensor and a plurality of stored data. In the soft sensor disclosed in Patent Document 1, the accuracy of the soft sensor is improved by learning stored data.

International Publication No. 2002/006953

In the soft sensor disclosed in Patent Document 1, historical data, which is a combination (data set) of input values and output values, is distributed into an input space quantized by a mesh of a predetermined size to create a case base. Output values are estimated based on this case base and the measured current input values. In the case-based method disclosed in Patent Document 1, values of a data set as a case are corrected by learning for each mesh, and a case-based data set is created.

However, since soft sensors estimate target variables that are difficult to measure using explanatory variables that are state variables that can be measured during the operation of the estimation target, it is difficult to estimate target variables with high accuracy.

Further, in the data set of Patent Document 1, only data that has been overwritten and corrected through learning is saved, and it is difficult to check the learning status of the input values and output values of the initial data set after correction.

The present invention has been made in view of the above-mentioned circumstances, and provides a state estimation device, a state estimation method, and a program capable of estimating a target variable with high accuracy and analyzing the learning status of a database. With the goal.

In order to achieve the above object, a state estimation device according to a first aspect of the present invention includes:
a database creation unit that creates a database that stores a plurality of data sets including explanatory variables and objective variables that represent the state of the estimation target;
a learning unit that trains the database;
an estimating unit that estimates the objective variable corresponding to the explanatory variable that is the measurement value of the estimation target based on the database learned by the learning unit;
The database creation unit selects, as the explanatory variable, a variable with a high degree of importance among state variables that can be measured during the operation of the estimation target,
The learning unit updates an initial data set before learning with a learning data set to perform learning and create an estimation database;
The estimation data set of the estimation database is a data set that includes data of an initial data set and the amount of data correction by learning.

Further, the correction amount is a correction value of the target variable, which is derived based on the difference between the estimated target variable and the actually measured target variable.
It may also be a thing.

Further, the database creation unit interpolates the target variables included in the plurality of data sets acquired for creating the database based on a change over time of the predetermined explanatory variable.
It may also be a thing.

Further, the database creation unit classifies the plurality of data sets acquired for creating the database into a plurality of data groups based on an environmental index that affects the relationship between the explanatory variable and the objective variable. and creating the database for each data group,
It may also be a thing.

Further, the estimating unit includes:
The degree of similarity between the measured explanatory variable and the explanatory variable in the estimation data set is calculated to extract neighboring data, and based on the number of neighboring data extracted for each database, the similarity is calculated for each database. estimating the objective variable based on the neighborhood data and the weight;
It may also be a thing.

The state estimation method according to the second aspect of the present invention includes:
A variable with a high degree of importance is selected as an explanatory variable among state variables that can be measured during the operation of the estimation target, and a plurality of data sets including the explanatory variables representing the state of the estimation target and a target variable are stored in a database. an initial database creation step of creating an initial database using
a learning step of training the initial database using a training dataset to create an estimation database;
an estimation step of estimating the target variable corresponding to the explanatory variable that is the measurement value of the estimation target based on the estimation database;
The estimation data set stored in the estimation database is a data set that includes data of an initial data set and the amount of correction of the data by learning.

The program according to the third aspect of the present invention is
computer,
A variable with a high degree of importance is selected as an explanatory variable among state variables that can be measured during the operation of the estimation target, and a plurality of data sets including the explanatory variables representing the state of the estimation target and a target variable are stored in a database. a database creation section that creates an initial database;
a learning unit that causes the database to learn using a training data set to create an estimation database that stores an estimation data set that includes data of an initial data set and an amount of correction of data due to learning;
an estimation unit that estimates the target variable corresponding to the explanatory variable that is the measurement value of the estimation target based on the estimation database;
operate as

According to the state estimation device, state estimation method, and program of the present invention, highly important state variables are selected and used as explanatory variables included in the data set stored in the database, so the objective variable can be estimated with high accuracy. It can be performed. In addition, since the estimation data set stored in the database is a data set that includes the data of the initial data set and the amount of correction, it is possible to check the amount of correction from the initial data set, making it easy to analyze the learning situation. be able to.

Conceptual diagram showing an overview of state estimation according to an embodiment of the present invention Functional block diagram of state estimation device according to Embodiment 1 Flowchart showing the flow of state estimation processing according to Embodiment 1 Conceptual diagram showing how to interpolate objective variables in basic data Functional block diagram of state estimation device according to Embodiment 2 Flowchart showing the flow of state estimation processing according to Embodiment 2 Conceptual diagram showing a method for estimating a target variable according to Embodiment 2 Graph showing example importance of state variables Graph showing an example of state estimation when variable selection is performed Graph showing an example of state estimation without variable selection

Hereinafter, a state estimating device and a state estimating method according to an embodiment of the present invention will be described with reference to the drawings.

(Embodiment 1)
(Overview of state estimation system)
FIG. 1 is a conceptual diagram showing an overview of state estimation according to this embodiment. Here, y(t) is the output of the estimation target P (controlled target), x _n (t) is the state variable, and ^y(t) is the estimated value of the output y(t) calculated by the soft sensor.

The state variables x _n (t) may include variables that have a weak influence (low importance) on the output y(t). If the output y(t) is estimated based on the state variable x _n (t) with low importance, it will be susceptible to noise and the estimation accuracy may deteriorate. In this embodiment, among the state variables x _n (t), variables that have a strong influence (high importance) on the output y (t) are selected as the state variables x _rn (t), and the selected A state estimation device that functions as a soft sensor is configured using the state variable x _rn (t).

Further, in this embodiment, a discrete time nonlinear system shown in the following equation is considered as a controlled object that is an estimation object P.
Here, f(·) represents a nonlinear function, φ(t-1) represents the state of the system before time t, and φ is also called an information vector.

φ(t-1) is defined by the following equation.
Here, n _y is the order of the output y, and n _xn is the order of the state variable x _rn .

(state estimation device)
As shown in the block diagram of FIG. 2, the state estimation device 1 according to the present embodiment includes a control section 11, a storage section 12, a display section 13, and an input section 14.

The control unit 11 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc., and controls the operation of the state estimation device 1. Further, the control unit 11 acquires the value of a parameter (explanatory variable x) representing the current state of the estimation target P, and based on the value of the acquired explanatory variable x and the database DB stored in the storage unit 12. Then, the objective variable y is estimated.

The control unit 11 realizes each function of the control unit 11 shown in FIG. 2 by loading various operating programs and data stored in the ROM, storage unit 12, etc. of the control unit 11 into the RAM and operating the CPU. . Thereby, the control unit 11 operates as a database creation unit 111, a learning unit 112, an explanatory variable acquisition unit 113, and an estimation unit 114.

The database creation unit 111 generates 1 included in the data set φ for estimating the target variable y from a plurality of state variables x _n included in the basic data PD acquired as past operation data or simulation data of the estimation target P. Alternatively, two or more state variables x _rn are selected as explanatory variables x. Further, the database creation unit 111 creates a database DB in which the data set φ is accumulated, based on the basic data PD related to the selected explanatory variable x.

The learning unit 112 trains the initial dataset ~φ(j) stored in the database DB using the training dataset φ ^* (t) (teacher data), and estimates the estimation dataset φ(j). generate. Each data set includes an explanatory variable x, an objective variable y, etc., and details of each data set will be described later. The initial dataset ~φ(j) and the learning dataset φ ^* (t) are, for example, datasets that are generated by sorting basic data PD acquired in advance and include mutually different data.

The explanatory variable acquisition unit 113 acquires the value of the explanatory variable x selected by the database creation unit 111 from among the parameters measured as representing the state of the estimation target P. The explanatory variable x is, for example, a physical quantity such as the temperature of a material, the weight, or the pressure inside an apparatus that can be measured using a physical sensor (hard sensor) in the estimation target P, which is a chemical plant.

The estimation unit 114 uses the estimation data set φ(j) learned by the learning unit 112 and stored in the database DB of the storage unit 12, the explanatory variables x of the estimation target P acquired by the explanatory variable acquisition unit 113, etc. Based on , calculate the estimated value ^y of the objective variable y.

The storage unit 12 is a nonvolatile memory such as a hard disk or a flash memory, and stores a program for estimating the objective variable y from the explanatory variable x, the objective variable ^y estimated by the estimation unit 114, and the like. The storage unit 12 also includes a database DB that stores a plurality of estimation data sets φ(j) for estimating the objective variable y. The initial data set ~φ(j) according to the present embodiment includes an explanatory variable x representing the state of the estimation target P and an objective variable y. Furthermore, the estimation data set φ(j) is obtained by adding the correction amount c to the initial data set ~φ(j) including the explanatory variable x and the objective variable y.

The display unit 13 is a display device that displays information such as an explanatory variable x indicating the current state of the estimation target P, an estimated target variable ^y, and is, for example, a liquid crystal monitor.

The input unit 14 is an input device such as a keyboard, a mouse, or a touch panel for inputting various setting values such as learning conditions for the estimation data set φ(j), instructions to start and end estimation of the target variable y, and the like.

Next, a method for estimating the state of the estimation target P using the state estimating device 1 will be specifically described with reference to the flowchart of FIG. In the state estimation process according to this embodiment, an explanatory variable x used for state estimation is selected from the acquired basic data PD, a database DB for state estimation is created, and the database DB is trained. Then, the objective variable y is estimated using the learned database DB.

As shown in FIG. 3, as an initial database creation step, the database creation unit 111 of the state estimation device 1 acquires basic data PD such as operation data of the estimation target P (step S11). The basic data PD may be stored in advance in the storage unit 12, or may be stored in an external storage device such as a server connected to the state estimation device 1 via a network.

Here, the explanatory variables x included in the basic data PD are state variables x _n that can be measured during the operation of the estimation target P, whereas the objective variables y cannot be measured during the operation of the estimation target P. This is a difficult variable. For example, the target variable y is the composition of a compound that is difficult to measure unless the chemical plant that produces the compound is stopped. In this case, the target variable y of the basic data PD acquired as operation data is only the data when the estimation target P stops operating, as shown in FIG. Therefore, in this embodiment, the objective variable y is interpolated based on the predetermined state variable x _n and used as the basic data PD.

Specifically, the database creation unit 111 selects one of the state variables _xn whose temporal change tendency is considered to be close to the target variable y. Then, the database creation unit 111 converts the data between the initial value of the objective variable y (time t=0) and the measured data (time t=t _m ) into the time change of the selected state variable x _n , e.g. Interpolation is performed based on the rate of change at each measurement time from t=0 to _tm . Although FIG. 4 shows interpolation when both the objective variable y and the selected state variable x _n simply increase, interpolation is actually performed based on the tendency of change such as increase or decrease. In addition, the method of selecting the state variables x _n to be referred to when performing interpolation is not particularly limited, and the state whose change tendency is estimated to be more similar to the target variable y based on past operation data, simulation data, etc. It is sufficient to select variables x _n .

The database creation unit 111 selects the state variable x _rn used for state estimation from the state variables x _n included in the acquired basic data PD as the explanatory variable x (step S12). As a variable selection method, a known method used in variable selection and feature value selection can be used. For example, a wrapper method such as a forward method or a backward method, a lasso regression method, a built-in method such as a decision tree, etc. can be used.

The database creation unit 111 extracts data related to the explanatory variable x and the objective variable y from the basic data PD and divides them into two groups, thereby creating an initial data set ~φ(j) (where j is a natural number from 1 to N). (N is 2 or more) and a learning data set φ ^* (t) is generated (step S13). Note that in order to suppress the influence of outliers, it is preferable to remove outliers from the basic data PD in advance.

The initial data set ~φ(j) consists of the explanatory variable ~x(j), which is data that can be measured in the estimation target P, and the explanatory variable ~x(j), which is data that can be measured in the estimation target P, and the explanatory variable ~φ(j), which is data that is difficult to measure during the operation of the controlled target, as shown in equation (3) below. It includes the objective variable ~y(j). The state estimating device 1 creates a database DB (hereinafter also referred to as initial database DB ₀ ) storing the initial data set ~φ(j), and stores it in the storage unit 12 (step S14).

Furthermore, as shown in the above equation, the initial data set ~φ(j) includes a correction amount c(j) that is a parameter for correcting the estimated value of the target variable y by learning described later. However, the correction amount ~c(j) in the initial data set ~φ(j) before learning is set to ~c(j)=0.

Next, as a learning step, the learning unit 112 trains the initial database DB ₀ using the learning data set φ ^* (t) (where t is a natural number from 1 to M) shown in the following formula. .

The learning dataset φ ^* (t) is generated based on the basic data PD, which is pre-measured data, like the initial dataset ~φ(j), and the initial dataset ~φ(j) is This is different data. Furthermore, since the learning data set φ ^* (t) is a data set used for learning the initial database DB ₀ , it does not include the correction amount c as shown in the above equation.

In the learning step, the learning unit 112 selects the learning data set φ ^* (t) (required point) (step S15). The learning unit 112 calculates the degree of similarity S between the selected learning data set φ ^* (t) and the initial data set ~φ(j) stored in the initial database DB ₀ (step S16). Similarity S(φ ^* (t), ~φ(j)) between the training dataset φ * (t) and the initial dataset ~φ (j), that is, the similarity S (φ ^* (t), ~φ (j)) included in the training dataset φ ^* (t) The closeness between the explanatory variable x and the explanatory variable x included in the initial data set ~φ(j) is expressed by the following formula.

Note that h _i is the bandwidth, and ~φ(j, i) represents the i-th element of the explanatory variable x in the j-th data set φ stored in the database DB. Further, n _y and n _x are the orders of the objective variable y and the explanatory variable x, respectively. The method for determining the bandwidth h _i is not particularly limited, and a known method such as a plug-in method can be used.

The similarity S is highest when there is an initial data set ~φ(j) with the same explanatory variable x as the required point φ ^* (t), and in this case, the similarity S is expressed by the following formula. Become.

In addition, if the explanatory variable x of the request point φ ^* (t) and the explanatory variable x of the initial dataset ~φ (j) are not similar, the similarity S(φ ^* (t), ~φ (j) ) approaches 0. Therefore, the learning unit 112 according to the present embodiment selects the initial data set ~φ(j) that satisfies the following conditional expression based on the degree of similarity S as neighboring data (step S17).
Here, T _th is a threshold value, and is set in the range of 0≦T _th ≦1.

Next, using the objective variable y and correction amount c of the initial dataset ~φ(j), which is the neighborhood data selected in step S17, a weighted local linear average method (Linearly Weighted Average: LWA) shown below is performed. The estimated value ^y of the objective variable y is calculated (step S18).
Here, k is the number of neighboring data selected in step S17.

In addition, w _i is a weight for the objective variable y and correction amount c included in the i-th initial data set ~φ(j) selected as neighborhood data, and is given by the following equation.

Through the above procedure, it is possible to calculate the estimated value ^y of the objective variable y for the explanatory variable x in the initial data set ~φ(j) based on the similarity S.

As shown in Equation (8), the learning unit 112 adds the target variable ~y(i) of the neighborhood data and its correction amount ~c(i) to the weight ω calculated using Equation (9). Multiply by _i and add up the values calculated for all neighboring data. Thereby, the learning unit 112 can calculate the estimated target variable ^y(t) corresponding to the degree of similarity S with neighboring data. Note that at the start of learning, the correction amount ~c(i) is 0, and is sequentially updated by the following data corrections.

The learning unit 112 calculates the value based on the error between the estimated value ^y estimated by equation (8) and the objective variable y ^* (t), which is an actual measured value or an interpolated value included in the learning data set φ ^* (t). Then, the correction amount ~c is updated (step S19). Specifically, the learning unit 112 updates the correction amount ~c using the steepest descent method shown in equation (10) below.

Here, ~c(i) is the amount of correction of the neighboring data selected in step S17. Further, η is a learning coefficient, and J(t) is an evaluation criterion defined by the following equations (11) and (12).

As shown in equation (12), the learning unit 112 calculates the correction amount ~c ^new that minimizes the error between the measured or interpolated objective variable y and the estimated objective variable ^y, and stores it in the database DB. The stored neighborhood data correction amount ~c(j) is updated. As a result, the initial data set ~φ(j) is updated sequentially. As shown in equations (10) to (12) above, the correction amount c is a correction value of the objective variable derived based on the difference between the estimated objective variable ^y and the actually measured or interpolated objective variable y. It is.

After updating the correction amount ~c(j), the process returns to step S15, and the learning unit 112 uses the next learning data set φ ^* (t) to learn the initial data set ~φ(j). The learning in steps S15 to S19 is repeated until the evaluation criterion J(t) becomes sufficiently small, that is, until it becomes equal to or less than a preset threshold Th (NO in step S20).

When the evaluation criterion J(t) becomes sufficiently small, that is, less than or equal to the preset threshold Th, the learning unit 112 ends the learning (YES in step S20). As a result, the learning unit 112 stores the initial data set ~φ(j) for which learning has been completed in the database DB as an estimation data set φ(j), which is a data set that allows more appropriate estimation. database DB (hereinafter also referred to as estimation database DBp).

As described above, in the estimation dataset φ(j) according to the present embodiment, the explanatory variable ~x(j) and the objective variable ~y(j) of the initial dataset ~φ(j) are not changed through learning. First, the independently provided correction amount ~c is updated and learned. As a result, the estimation dataset φ(j) after learning retains the explanatory variable ~x(j) and objective variable ~y(j) of the initial dataset ~φ(j), so it is updated by learning. It becomes possible to check parameters easily.

Subsequently, as a state estimation step, the state estimation device 1 estimates the state of the estimation target P using the learned estimation database DBp. When state estimation is started, the estimation unit 114 of the state estimation device 1 acquires the explanatory variable x (required point) of the estimation target P (step S21).

The estimation unit 114 calculates the degree of similarity S between the obtained explanatory variable x and the estimation data set φ(j), and selects neighboring data of the obtained explanatory variable x (step S22). The method of calculating the similarity S and the method of selecting neighboring data are the same as steps S16 and S17 in the learning step. That is, the estimating unit 114 calculates the similarity S between the acquired explanatory variable x and the explanatory variable x of the estimation data set φ(j) in the same manner as in equation (5). Then, the estimation unit 114 selects a predetermined number of estimation data sets φ(j) from those with the highest calculated similarity S as neighboring data.

The estimation unit 114 estimates the target variable y based on the neighborhood data selected in step S22 (step S23). The method for estimating the target variable y is the same as the method for estimating the target variable y in step S18 in the learning step. That is, the estimation unit 114 applies the weighted local linear average method similar to equations (8) and (9) to the estimation data set φ(j), which is the neighboring data selected in step S22, to estimate the Calculate the objective variable ^y.

The state estimation device 1 repeats the state estimation processing in steps S21 to S23 until the state estimation processing is completed (NO in step S24). Then, when the termination conditions such as input of a termination instruction to the input unit 14 and completion of a predetermined estimation processing time are satisfied (YES in step S24), the state estimation device 1 terminates the state estimation processing.

As explained above, according to the state estimation device and state estimation method according to the present embodiment, the data set φ is constructed based on the explanatory variable x selected as a highly important state variable, and the data set φ is Create a database DB used for estimation. Further, the estimation data set φ(j) is configured as data including the initial data set ~φ(j) and the correction amount c(j), and the database DB is trained. Therefore, by using the database DB configured and trained using explanatory variables with high importance, it is possible to estimate the target variable y with high accuracy. Furthermore, since the amount of correction c(j) from the initial data set ~φ(j) can be confirmed, it is possible to easily grasp the correction status. Thereby, it is possible to easily analyze changes in the correction amount c, the validity of the initial data set ~φ(j), the number of neighboring data, and the like.

(Embodiment 2)
In the first embodiment described above, there is one database DB, but two or more databases DB can also be used. In this embodiment, a plurality of databases DB are created based on the environmental index EI that influences the operational state of the controlled object, that is, the relationship between the explanatory variable x and the objective variable y, and these databases are used to determine the objective. The state estimation device 2 that estimates the variable y will be explained.

In this embodiment, the configuration of database DB is different from Embodiment 1, and other configurations are the same as in Embodiment 1, so the same reference numerals are given and explanations are omitted.

As shown in FIG. 5, the state estimation device 2 according to this embodiment includes a plurality of databases DBn. The state estimation device 2 according to this embodiment includes four databases DBn (databases DB1 to DB4). The flow of state estimation according to this embodiment will be described below with reference to the flowchart of FIG. 6.

As shown in FIG. 6, as an initial database creation step, the state estimation device 2 acquires basic data PD including a plurality of data sets for creating the database DBn, such as operational data of the estimation target P (step S31). . The basic data PD according to this embodiment is given as a plurality of data groups with different environmental indicators EI. The environmental index EI is an index that affects the operating state of the estimation target P, that is, the relationship between the explanatory variable x and the objective variable y, and is, for example, an index defined based on the temperature, humidity, etc. around the estimation target P. It is. The basic data PD acquired in step S31 is a data group such as operational data acquired when the environmental index EI is far beyond a predetermined range, and for example, the basic data PD is acquired in each season where the environmental temperature differs greatly. This is operational data.

The database creation unit 111 selects a state variable _xrn with a large degree of importance as an explanatory variable x from the state variables _xn included in the acquired basic data PD (step S32). The variable selection method is similar to step S12 of the first embodiment. Further, the database creation unit 111 classifies the basic data PD related to the selected explanatory variable x into a plurality of data groups based on the environmental index EI. For example, the database creation unit 111 classifies the basic data PD into data groups for each season with different environmental temperatures based on the explanatory variable x (step S33).

The database creation unit 111 generates an initial dataset ~φ(j) and a learning dataset φ ^* (t) for each classified data group using the same method as in Embodiment 1 (step S34), A plurality of initial databases DB ₀ are created (step S35).

Subsequently, as a learning step, the learning unit 112 performs learning using the learning data set φ ^* (t) for each initial database DB ₀ (step S36). The learning step process (step S36) is the same process as steps S15 to S20, which are the learning step processes according to the first embodiment, for each initial database _DB0 . As a result, a plurality of learned databases DBn are created.

Subsequently, as a state estimation step, the state estimation device 2 estimates the target variable y using the plurality of created databases DBn (FIG. 7). Specifically, the explanatory variable acquisition unit 113 acquires the measured value of the explanatory variable x of the estimation target P (step S37). The estimation unit 114 calculates the degree of similarity S between the request point based on the measurement value and the estimation data set φ(j) in each database DBn, and calculates the degree of similarity S between the request point based on the measured value and the estimation data set φ(j), and calculates the degree of similarity S between the request point based on the measurement value and the estimation data set φ(j), Data set φ(j) is selected as neighboring data (step S38).

The estimation unit 114 calculates the estimated value ^y of the target variable y for each database DBn based on the selected neighborhood data (step S39). The method of calculating the estimated objective variable ^y is the same as that in step S23 according to the first embodiment.

Furthermore, the estimating unit 114 counts the number of data N ₁ to N ₄ exceeding the reference similarity Ss in each of the databases DB1 to DB4 (step S40). The estimation unit 114 uses the counted data numbers N ₁ to N ₄ to calculate weights v ₁ to v ₄ according to the following equations.

The estimation unit 114 calculates the estimated value ^y(t) corresponding to the required point based on the weighted average method shown in the following formula (step S41).

The state estimation device 2 repeats the state estimation processing in steps S37 to S41 until the state estimation processing ends (NO in step S42). Then, when termination conditions such as inputting a termination instruction to the input unit 14 and completion of a predetermined estimation processing time are satisfied (YES in step S42), the state estimation device 2 terminates the state estimation processing.

Thereby, the objective variable y can be estimated based on the environmental index EI, with the relationship between the explanatory variable x and the objective variable y largely reflecting the influence of the database DBn having a similar tendency. Therefore, even in a state where the environmental index EI changes significantly, it is possible to estimate the state with higher accuracy.

(Numeric example)
A numerical example of state estimation according to an embodiment of the present invention will be described below. In this example, state estimation in a polycondensation reaction process related to the production of polyester will be described. This process increases the polycondensation reaction by controlling temperature and pressure to produce polyester.

In polycondensation reactions in the manufacturing process of polymer compounds, the viscosity of the compound is generally used as an indicator of product quality. The viscosity of a compound in a polycondensation reaction is usually measured after the reaction is completed, and it is difficult to measure it during the reaction. For example, in the polycondensation reaction of polyester, limiting viscosity (intrinsic viscosity) and reduced viscosity are important indicators of product quality. Therefore, in this example, state estimation is performed using reduced viscosity as the objective variable y. In addition, there are many state variables x _n serving as explanatory variables x, such as temperature, pressure, and raw material flow rate during the reaction, but by selecting variables in the initial database creation step, 183 state variables x _n can be measured during the reaction. From the above, 23 state variables x _rn of high importance were selected as explanatory variables x.

In this example, random forest was used to analyze the importance of state variables x _n and select variables. FIG. 8 shows a portion of the importance levels of the state variables x _n calculated by random forest. As shown in FIG. 8, it can be seen that by applying random forest, state variables x _n with high importance can be selected as explanatory variables x.

Furthermore, in this example, the basic data PD obtained as the actual measurement data was divided into data for the initial data set ~φ(j), data for the learning dataset φ ^* (t), and data for verification. Further, the number k of neighboring data when selecting neighboring data was set to 3, and the learning coefficient η used for learning was set to 0.2.

Examples of estimation results using the database DB created under the above conditions are shown in FIGS. 9 and 10. In the graphs of FIGS. 9 and 10, the time on the horizontal axis is normalized from 0 to 100, and the reduced viscosity on the vertical axis is normalized in the range from 0 to 1. FIG. 9 is a graph showing the results of estimation performed by selecting variables and setting the number of explanatory variables x to 23. As shown in FIG. 9, the viscosity can be estimated to gradually increase so as to approach the true value (actually measured value) of the reduced viscosity indicated by the circle, and it can be seen that appropriate state estimation is performed.

FIG. 10 shows the results of state estimation when 183 state variables x _n are used as explanatory variables x without variable selection. As shown in FIG. 10, compared to the case of FIG. 9, it can be seen that the true value and estimated value of the reduced viscosity deviate from each other, and the estimation accuracy has deteriorated.

In this example, estimation was performed using a plurality of pieces of verification data, and the results were evaluated using the error rate _Ie shown in the following formula.

The average value of the error rate _Ie for all verification data in this example was 2.91%, confirming that effective state estimation could be performed. In addition, the average error rate _Ie when _no variables are selected is 3.24%. It was confirmed that state estimation can be performed with high accuracy.

In the above embodiment, the learning of the database DB is finished before performing the state estimation process, but the present invention is not limited to this. When the objective variable y can be measured during the operation of the estimation target P, for example, when the objective variable y can be measured at a lower frequency than the explanatory variable x, the measured explanatory variable x and objective variable y are used to create the database DB. It is also possible to have students learn online. In this case, the control unit 11 of the state estimation device 1 performs learning of the estimation database DBp using the explanatory variable x and the objective variable y acquired from the sensor of the estimation target P in operation as the learning data set φ ^* . do it. The learning timing may be, for example, the timing at which the objective variable y is acquired. This makes it possible to perform more accurate state estimation corresponding to state changes of the estimation target P.

Further, the state estimation method according to the embodiment described above can be realized using a normal computer system. For example, a computer program for executing the state estimation according to the above embodiment is distributed via a network such as the Internet, and the computer program is installed on a computer, thereby causing the computer device to execute the above state estimation. It can function as a state estimation device.

The present invention is suitable for state estimation when a plurality of state variables can be measured. Further, the present invention is suitable for estimating the state of a system in which the state of the estimation target changes with time.

Reference Signs List 1, 2 state estimation device, 11 control unit, 111 database creation unit, 112 learning unit, 113 explanatory variable acquisition unit, 114 estimation unit, 12 storage unit, 13 display unit, 14 input unit, DB database, P estimation target

Claims (7)

a database creation unit that creates a database that stores a plurality of data sets including explanatory variables and objective variables that represent the state of the estimation target;
a learning unit that trains the database;
an estimating unit that estimates the objective variable corresponding to the explanatory variable that is the measurement value of the estimation target based on the database learned by the learning unit;
The database creation unit selects, as the explanatory variable, a variable with a high degree of importance among state variables that can be measured during the operation of the estimation target,
The learning unit updates an initial data set before learning with a learning data set to perform learning and create an estimation database;
The estimation data set of the estimation database is a data set that includes data of an initial data set and an amount of data correction by learning.
A state estimation device characterized by: The correction amount is a correction value of the target variable, which is derived based on the difference between the estimated target variable and the actually measured target variable.
The state estimating device according to claim 1, characterized in that: The database creation unit interpolates the target variables included in the plurality of data sets acquired to create the database based on a change over time of the predetermined explanatory variable.
The state estimating device according to claim 1, characterized in that: The database creation unit classifies the plurality of data sets acquired to create the database into a plurality of data groups based on an environmental index that affects the relationship between the explanatory variable and the objective variable, creating the database for each data group;
The state estimating device according to any one of claims 1 to 3. The estimation unit is
The degree of similarity between the measured explanatory variable and the explanatory variable in the estimation data set is calculated to extract neighboring data, and based on the number of neighboring data extracted for each database, the similarity is calculated for each database. estimating the objective variable based on the neighborhood data and the weight;
The state estimating device according to claim 4, characterized in that: A variable with a high degree of importance is selected as an explanatory variable among state variables that can be measured during the operation of the estimation target, and a plurality of data sets including the explanatory variables representing the state of the estimation target and a target variable are stored in a database. an initial database creation step of creating an initial database using
a learning step of training the initial database using a training dataset to create an estimation database;
an estimation step of estimating the target variable corresponding to the explanatory variable that is the measurement value of the estimation target based on the estimation database;
The estimation data set stored in the estimation database is a data set that includes data of an initial data set and an amount of data correction by learning.
A state estimation method characterized by: computer,
A variable with a high degree of importance is selected as an explanatory variable among state variables that can be measured during the operation of the estimation target, and a plurality of data sets including the explanatory variables representing the state of the estimation target and a target variable are stored in a database. a database creation section that creates an initial database;
a learning unit that causes the database to learn using a training data set to create an estimation database that stores an estimation data set that includes data of an initial data set and an amount of correction of data due to learning;
an estimation unit that estimates the target variable corresponding to the explanatory variable that is the measurement value of the estimation target based on the estimation database;
A program that operates as