JP2021174352A - Plant control support device, program and plant control support method - Google Patents

Abstract

To provide a plant control support device that enables an accurate selection of a manipulated variable which contributes to improved operational performance.SOLUTION: While a plant control support device 100 comprises an operation variable determination unit 160 that selects a manipulated variable based on a degree of correlation with an operation performance index and an operation frequency, the plant control support device selects the manipulated variable that contributes to an improvement of operational performance according to the following procedure, and calculates a manipulated amount (value of the manipulated variables). (1) Operation data is acquired from a control device of a plant. (2) Invalid data acquired during a plant outage period, a test period, a failure period, etc. is deleted from the operation data. (3) An operation variable is selected from the operation data, which has a large degree of correlation with an operational performance index and a large fluctuation amount of the manipulated amount as an operation variable that contributes to an improvement of the operational performance. (4) A statistical model that simulates the characteristics of the plant is constructed by inputting the manipulated amount and outputting the operational performance index of the plant. (5) Using machine learning technology, the manipulated amount with the best operational performance index is derived from the statistical model.SELECTED DRAWING: Figure 1

Description

The present invention provides a plant control support device, a program, and a plant control support that enable selection of operation variables to be set / adjusted in order to improve the operability of controlled objects including plants such as power plants, waste incinerators, and factory equipment. Regarding the method.

With the technological innovation of ICT (Information and Communication Technology) and IoT (Internet of Things), attention is focused on the utilization of big data. For controlled objects including plants, a technique for collecting and analyzing a large amount of data (sensor data, measurement data) is attracting attention for the purpose of improving the performance, quality, productivity, etc. of the controlled object. On the other hand, in the power generation business field, there is increasing interest in dealing with environmental problems such as renewable energy and global warming countermeasures. In order to improve the social value, environmental value, and economic value of the power generation business, thermal power plants (thermal power plants) are required to operate in consideration of operational performance such as operating rate and environmental performance.

As an effort to improve the operational performance of a thermal power plant, Patent Document 1 discloses a control device that minimizes the concentration of nitrogen oxides, carbon monoxide, etc. contained in exhaust gas, which is an index of environmental performance. ..
In order to improve the operational performance of a plant, it is necessary to appropriately determine the instrumental variables (instrumental parameters) according to the structure and operational status of the plant to be applied. Patent Document 2 discloses an operation support device that extracts operation parameters that contribute to improving the operation performance of the plant.

Japanese Unexamined Patent Publication No. 2012-141862 Japanese Unexamined Patent Publication No. 2018-081350

The driving support device described in Patent Document 2 extracts operating parameters (operating variables) that contribute to improving the operational performance of the controlled object based on the operating data collected from the controlled object and the design information of the controlled object. However, this extraction method has the possibility of extracting instrumental variables by referring to operation data that should not be referred to originally, and the possibility of extracting instrumental variables that are not the target of operation in normal operation. There is room left.
The present invention has been made in view of such a background, and an object of the present invention is to provide a plant control support device, a program, and a plant control support method that enable accurate selection of instrumental variables that contribute to improvement of operational performance. And.

In order to solve the above-mentioned problems, the plant control support device according to the present invention is a plant control support device that selects an instrumental variable that contributes to the improvement of the operation performance index of the plant, and has a degree of correlation with the operation performance index. It also includes an instrumental variable determination unit that selects the instrumental variable based on the operation frequency.

According to the present invention, it is possible to provide a plant control support device, a program, and a plant control support method that enable accurate selection of instrumental variables that contribute to improvement of operational performance. Issues, configurations and effects other than those described above will be clarified by the description of the following embodiments.

It is an overall block diagram of the plant control support apparatus which concerns on this embodiment. It is a data structure diagram of the test / maintenance / stop information DB stored in the operation data DB which concerns on this embodiment. It is a data block diagram of the failure occurrence information DB stored in the operation data DB which concerns on this embodiment. It is a data structure diagram of the operation data item DB stored in the operation data DB which concerns on this embodiment. It is a data structure diagram of the operation variable keyword DB stored in the operation data DB which concerns on this embodiment. It is a data block diagram of the failure sensor DB stored in the operation data DB which concerns on this embodiment. It is a data composition diagram of the evaluation score DB stored in the processing result DB which concerns on this embodiment. It is a flowchart of the plant control support process which concerns on this embodiment. It is a flowchart of the extraction process of effective analysis period and effective data item which concerns on this embodiment. It is a flowchart of the evaluation score calculation process of the data item which concerns on this embodiment. It is a flowchart of the selection process of the operation variable which concerns on this embodiment. It is a figure which shows the example of the control logic circuit diagram which concerns on this embodiment.

Next, the plant control support device in the embodiment (embodiment) for carrying out the present invention will be described. The plant control support device selects the manipulated variables that contribute to the improvement of operational performance according to the following procedure, and calculates the manipulated variable (value of the manipulated variables). (1) Acquire operation data from the control device of the plant. The operation data includes, for example, set values (designated values, operation amounts) of operation variables such as damper opening, and state quantities indicating the operation status of the plant. (2) Exclude invalid data acquired during the plant outage period, test period, failure period, etc. from the operation data. (3) From the operation data, select an instrumental variable that has a large correlation with an operation performance index such as the amount of nitrogen oxides and a large fluctuation amount of the operation amount as an operation variable that contributes to the improvement of the operation performance. (4) A statistical model that simulates the characteristics of the plant is constructed by inputting the operation amount and outputting the operation performance index of the plant. (5) Using machine learning technology, the manipulated variable with the best operational performance index is derived from the statistical model.

After removing invalid data, the plant control support device selects the instrumental variables that have a high correlation with the operation performance index, and selects the instrumental variables that contribute to the improvement of the operation performance with high accuracy. In addition, the plant control support device selects instrumental variables with a large fluctuation amount of the manipulated variable, and except for the manipulated variables that are not targeted for operation in normal operation (small fluctuation amount), the manipulated variables that contribute to the improvement of operational performance. To select. Furthermore, the plant control support device derives the optimum manipulated variable from a statistical model that simulates the characteristics of the plant. The plant operator will be able to formulate an operation plan for the plant based on the instrumental variables and manipulated variables output by the plant control support device, and will be able to improve the efficiency and operational performance of the planning. ..

≪Configuration of plant control support device≫
FIG. 1 is an overall configuration diagram of the plant control support device 100 according to the present embodiment. The plant control support device 100 is connected to the external device 210 and the plant 220. The external device 210 is a user interface device of the plant control support device 100 that includes a display, a keyboard, a mouse, and the like and is used by the user of the plant control support device 100.

The plant 220 includes equipment 222 that constitutes the plant and a control device 221 that controls the operation of the plant 220. The control device 221 receives the measurement signal indicating the state quantity of the plant output from the device 222, and outputs the measurement signal as the status quantity to the plant control support device 100. Further, the control device 221 receives the operation amount of the operation variable from the plant control support device 100 and outputs it as an operation signal (control signal) to the device 222 to control the operation of the plant 220. The control device 221 may receive the manipulated variable of the manipulated variable from a device different from the plant control support device 100 and output the manipulated signal to the device 222.

The plant control support device 100 is composed of a control unit composed of a CPU (Central Processing Unit), a storage unit composed of a ROM (Read Only Memory), a RAM (Random Access Memory), and an SSD (Solid State Drive), and a communication card. It is a computer including an input / output unit (see the input / output interface 120 of FIG. 1 described later) and the like. The input / output unit transmits / receives communication data to / from the external device 210 or the plant 220.

The input / output interface 120 is composed of a control unit and an input / output unit, and stores data received from the external device 210 or the plant 220 in the operation data DB 130 or the operation data DB 140. Further, the input / output interface 120 transmits the operation amount output by the operation amount determination unit 170, which will be described later, to the plant 220. The external device 210 can access the information stored in the operation data DB 130, the operation data DB 140, and the processing result DB 150, which will be described later, via the input / output interface 120.

≪Plant control support device: Configuration of storage part≫
The storage unit includes an operation data DB 130 (operation data database), an operation data DB 140, and a processing result DB 150. Further, the storage unit shows the procedure of the plant control support process (see FIG. 8 described later), and stores a program for making the computer function as the plant control support device 100. The operation data DB 130 stores design information such as design documents and business data related to the structure of the plant 220. Business data includes, for example, a test period, a failed sensor, an operation parameter (instrumental variable), an analysis target period, an evaluation point of an evaluation item, a parameter referred to in a plant control support process, and the like.

FIG. 2 is a data configuration diagram of the test / maintenance / stop information DB 310 stored in the operation data DB 130 according to the present embodiment. The test / maintenance / shutdown information DB 310 stores the test period, maintenance period, and shutdown period of the plant 220.

FIG. 3 is a data configuration diagram of the failure occurrence information DB 320 stored in the operation data DB 130 according to the present embodiment. The failure occurrence information DB 320 stores the time and degree (failure classification) of the failure that occurred in the plant 220.

FIG. 4 is a data configuration diagram of the operation data item DB 330 stored in the operation data DB 130 according to the present embodiment. The operation data item DB 330 stores the operation data item stored in the operation data DB 140 and its tag (described as "Tag No." in FIG. 4). Operational data items include instrumental variables and state quantities. In the following, the operation data item is also simply referred to as a data item.

FIG. 5 is a data configuration diagram of the operation variable keyword DB 340 stored in the operation data DB 130 according to the present embodiment. The operation variable keyword DB 340 stores keywords (keywords included in the operation data item column) of data items that are operation variables among the data items (see FIG. 4).

In addition to the instrumental variables, the data items include data items that are observed values (plant state quantities) of the equipment 222 of the plant 220. Among the state quantities, there are also observed values corresponding to instrumental variables. Specifically, among the state quantities, there is a state quantity measured by a sensor as a result of operating the device 222 (see FIG. 1) by designating the values of the operation variables (operation amount, set value, designated value). As an example of such a state quantity, there is a damper opening degree measured by a sensor after operating a damper by designating an operation variable called a damper opening degree. In the following explanation, if there is an observed value (state quantity) corresponding to the manipulated variable, the observed value is regarded as the value of the manipulated variable. This is because the observed value is considered to indicate the value after the operation (for example, the opening degree of the damper) more accurately than the value of the manipulated variable.

FIG. 6 is a data configuration diagram of the failure sensor DB 350 stored in the operation data DB 130 according to the present embodiment. The failure sensor DB 350 stores the failed sensor (device) and the failure period.

Returning to FIG. 1, the operation data DB 140 stores operation data which is information on the operation status of the plant. The operation data includes, for example, data (plant state amount) and operation amount (operation variable set value, designated value) acquired by a sensor installed in the device 222 as operation record data. The operation data is time series data that is the value of the operation data item.
The processing result DB 150 stores the processing results of the manipulated variable determination unit 160 and the manipulated variable determination unit 170, which will be described later. In the processing result DB 150, for example, an evaluation score for a data item is stored.

FIG. 7 is a data structure diagram of the evaluation score DB 360 stored in the processing result DB 150 according to the present embodiment. In the evaluation score DB 360, the identification number 361 of the data item 362 which is an instrumental variable, the degree of correlation 363 with the operational performance index (absolute value of the correlation coefficient), the coefficient of variation 364 (the degree of variation of the value of the instrumental variable), and the evaluation. The score 365 is stored.

≪Plant control support device: Control unit configuration: Instrumental variable determination unit≫
Returning to FIG. 1, the control unit includes an instrumental variable determination unit 160 and an operation amount determination unit 170. The instrumental variable determination unit 160 selects an instrumental variable having a large degree of correlation (absolute value of the correlation coefficient) with the operational performance index and a large amount of fluctuation. The instrumental variable determination unit 160 includes an effective analysis period / valid data item extraction unit 161, an evaluation item generation unit 162, an evaluation point acquisition unit 163, a weighting determination unit 164, a data item evaluation unit 165, and an instrumental variable selection unit 166.

The valid analysis period / valid data item extraction unit 161 sets the data item (valid data item) and the period (valid analysis period) to be analyzed for determining the operation variable in the operation data stored in the operation data DB 140. Is extracted. Specifically, the valid analysis period / valid data item extraction unit 161 excludes the test period, maintenance period, and outage period (see FIG. 2) of the plant 220 from the analysis target period stored in the operation data DB 130. This is because the data acquired by the sensor during the test period, maintenance period, and outage period is not the data during operation of the plant 220 to be analyzed, but invalid data (noise).

In addition, the valid analysis period / valid data item extraction unit 161 extracts data items that are instrumental variables. Specifically, the valid analysis period / valid data item extraction unit 161 contains data including keywords stored in the operation variable keyword DB340 (see FIG. 5) among the data items included in the operation data item DB330 (see FIG. 4). Extract items.

Further, the valid analysis period / valid data item extraction unit 161 excludes the failure period included in the failure sensor DB 350 (see FIG. 6) from the analysis target period according to the conditions. This is because the data acquired by the sensor during the failure period is not correct data of the plant 220 to be analyzed, but incorrect (inaccurate) data (noise). The period obtained by excluding the test period, maintenance period, shutdown period, and sensor failure period of the plant 220 from the analysis target period of the operation data is also referred to as an effective analysis period.

In addition, the valid analysis period / valid data item extraction unit 161 determines the state quantity measured by the sensor after the operation by the operation variable value (operation amount, set value / specified value for the plant 220) in the data item according to the condition. Exclude data items that cannot be acquired. For example, when the sensor for measuring the damper opening is out of order, the damper opening data item is excluded as an invalid data item as an instrumental variable.

The evaluation item generation unit 162 calculates the degree of correlation (causal relationship, absolute value of the correlation coefficient) between the value of the manipulated variable in the effective analysis period and the operational performance index, and the coefficient of variation. The operational performance index is an index showing the environmental performance such as the concentration / emission of nitrogen oxides (NOx) and carbon monoxide, and the operating rate of the plant. The smaller the amount of nitrogen oxides and carbon monoxide, the higher the operational performance. The evaluation item generation unit 162 calculates the absolute value of the correlation coefficient with the operation performance index for each data item which is an instrumental variable, and stores it in the column of the correlation degree 363 of the evaluation score DB 360 (see FIG. 7).

Further, the evaluation item generation unit 162 calculates the coefficient of variation for each data item which is an instrumental variable and stores it in the column of the coefficient of variation 364 of the evaluation score DB 360. Specifically, the evaluation item generation unit 162 normalizes the value of the manipulated variable, which is time series data, by, for example, a method of z-score normalization or min-max normalization, and uses the following equation (1) to normalize the coefficient of variation. Is calculated.
Coefficient of variation = standard deviation / mean (1)

Further, the evaluation item generation unit 162 generates a characteristic curve of the value of the manipulated variable in the effective analysis period and the operation performance index. Next, the evaluation item generation unit 162 approximates the characteristic curve with a characteristic curve such as a linear function or a Poisson distribution function, and displays it on the display of the external device 210.

The evaluation point acquisition unit 163 acquires the evaluation points of the correlation degree and the coefficient of variation, which are evaluation items, from the operation data DB 130. The evaluation score is, for example, a score of 5 to 1, which is a score that is a basis for weighting when evaluating the degree of correlation and the coefficient of variation.
The weighting determination unit 164 determines the weighting from the evaluation points of the evaluation items (correlation degree and coefficient of variation). For example, assume that the evaluation point of the correlation degree is 3 and the evaluation point of the coefficient of variation is 4. Then, the weighting of the correlation degree is calculated to be 3 / (3 + 4) = 0.42857, and the weighting of the coefficient of variation is calculated to be 4 / (3 + 4) = 0.57143.

The data item evaluation unit 165 calculates an evaluation score using the following equation (2) based on the correlation degree and the coefficient of variation calculated by the evaluation item generation unit 162 and the weighting calculated by the weighting determination unit 164. It is stored in the evaluation score 365 column of the evaluation score DB 360 (see FIG. 7).
Evaluation score = Correlation degree x Correlation degree weighting + Coefficient of variation x Coefficient of variation weighting (2)

The instrumental variable selection unit 166 selects a predetermined number of data items whose evaluation score is equal to or higher than a predetermined threshold value and whose evaluation score is higher as the instrumental variables that contribute to the improvement of operational performance.
The details of the processing of the instrumental variable determination unit 160 described above will be described with reference to FIGS. 9 to 11 described later. Subsequently, the operation amount determination unit 170 will be described.

≪Plant control support device: Control unit configuration: Operation amount determination unit≫
The manipulated variable determination unit 170 derives the manipulated variable (designated value, set value) of the manipulated variable to be output to the plant 200. The operator of the plant 220 may use the manipulated variable derived by the manipulated variable determination unit 170 as it is as a set value for controlling the device 222 in the control device 221. Further, the operator may formulate an operation plan of the plant with reference to the operation variables and the operation amounts derived from the plant control support device 100, and determine the set value of the control device 221. The operation amount determination unit 170 includes a statistical model 171, an operation learning model 172, a model construction unit 173, an operation amount learning unit 174, and an operation amount calculation unit 175. The statistical model 171 and the operation learning model 172 are stored in the storage unit.

The statistical model 171 is a statistical model that inputs the manipulated variable of the manipulated variable and outputs the operational performance index. For example, the statistical model 171 may be a neural network that inputs the manipulated variable of the manipulated variable and outputs the operational performance index. Specifically, it may be a neural network trained using learning data in which the operation amount is input from the operation data and the operation performance index is output (correct answer data). Further, it may be a statistical model based on the Radial Basis Function Network described in Patent Document 1. Alternatively, for example, a boiler model of a thermal power plant may be constructed, and the parameters of the model may be adjusted using the operation data to obtain a statistical model 171.

The operation learning model 172 is a model for deriving the manipulated variable of the manipulated variable having the best operational performance index, and is, for example, a reinforcement learning model in which the statistical model 171 is the control target (environment). The operation learning model 172 may also be a calculation model of an optimization algorithm such as a genetic algorithm or a nonlinear programming method.
The model building unit 173 builds the statistical model 171. For example, if the statistical model 171 is a neural network, the model construction unit 173 prepares learning data in which the operation amount is input from the operation data and the operation performance index is output (correct answer data), and the learning data is used. Train statistical model 171.

The operation amount learning unit 174 trains the operation learning model 172 with the statistical model 171 as a control target. For example, when the operation learning model 172 is a reinforcement learning model, the operation amount learning unit 174 may train the operation learning model 172 using a normalized Gaussian function network, which is a method of the Actor-Critic learning method. good.
The operation amount calculation unit 175 calculates the operation amount for which the operation performance index is optimal using the operation learning model 172, and stores it in the processing result DB 150. The operation amount calculation unit 175 may output the operation amount to the plant 220 via the input / output interface 120.

≪Plant control support processing≫
FIG. 8 is a flowchart of the plant control support process according to the present embodiment. With reference to FIG. 8, a process in which the plant control support device 100 selects an instrumental variable that contributes to the improvement of operational performance and a process in which the optimum manipulated amount of the instrumental variable is derived will be described. The plant control support process may be executed repeatedly, or may be executed at a predetermined timing such as when the plant control support device 100 is started for the first time or when a predetermined period has passed since the previous process. , It may be executed according to the instruction of the user.

In step S11, the instrumental variable determination unit 160 determines whether or not the instrumental variable is determined, and if it is determined, proceeds to step S12 (step S11 → YES), and if it is not determined, it proceeds to step S16 (step S11 → NO). move on. The operation variable determination unit 160 may inquire the user via the external device 210 to determine whether or not the determination is correct, or may acquire and determine the correctness parameter in the operation data DB 130, or the previous time. It may be determined that the operation variable is determined when a predetermined time has elapsed from the determination.

In step S12, the instrumental variable determination unit 160 executes the extraction process of the valid analysis period / valid data item. The details of the effective analysis period and the extraction process of the effective data items will be described with reference to FIG. 9 described later.
In step S13, the evaluation item generation unit 162 of the operation variable determination unit 160 calculates the evaluation items of the data items. Specifically, the evaluation item generation unit 162 calculates the degree of correlation between the value of the manipulated variable and the operational performance index during the effective analysis period, and the coefficient of variation of the manipulated variable. The evaluation item generation unit 162 stores the calculated correlation degree and coefficient of variation in the columns of the correlation degree 363 and the coefficient of variation 364 of the evaluation score DB 360 (see FIG. 7), respectively. Further, the evaluation item generation unit 162 generates a characteristic curve of the value of the operation variable approximated by the linear function or the Poisson distribution function and the operation performance index, and displays it on the display of the external device 210.

In step S14, the instrumental variable determination unit 160 executes the evaluation score calculation process of the data item. The details of the evaluation score calculation process of the data item will be described with reference to FIG. 10 described later.
In step S15, the instrumental variable determination unit 160 executes the instrumental variable selection process. The details of the operation variable selection process will be described with reference to FIG. 11 described later.
In step S16, the operation amount determination unit 170 determines whether or not to learn the operation method, and if it learns, it proceeds to step S17 (step S16 → YES), and if it does not learn, it goes to step S19 (step S16 → NO). move on. The operation amount determination unit 170 may inquire the user via the external device 210 to determine whether or not the learning is correct, or may acquire and determine whether or not the learning is correct or not in the operation data DB 130, or the previous time. It may be determined that learning is performed when a predetermined time has elapsed from the operation method learning.

In step S17, the model building unit 173 builds the statistical model 171.
In step S18, the operation amount learning unit 174 trains the operation method. Specifically, the operation amount learning unit 174 trains the operation learning model 172 with the statistical model 171 as the control target.
In step S19, the operation amount calculation unit 175 calculates the operation amount for which the operation performance index is optimal using the operation learning model 172, and stores it in the processing result DB 150.
In step S20, the operation amount determination unit 170 determines whether or not to end the plant control support process, and if it ends (step S20 → YES), ends the plant control support process, and if it does not end (step S20 → YES). NO) Return to step S11. The operation amount determination unit 170 may inquire the user via the external device 210 to determine whether or not the termination is appropriate, or may acquire and determine the appropriateness / rejection parameter in the operation data DB 130.

≪Valid analysis period / Extraction process of valid data items≫
FIG. 9 is a flowchart of the effective analysis period / valid data item extraction process according to the present embodiment. The valid analysis period / valid data item extraction process (see step S12 in FIG. 8) executed by the valid analysis period / valid data item extraction unit 161 will be described with reference to FIG. The effective analysis period at the start of processing may be an analysis target period set by the user and stored in the operation data DB 130, or may be a period of operation data stored in the operation data DB 130.

In step S31, the valid analysis period / valid data item extraction unit 161 determines whether or not the judgment of the valid analysis period is necessary, proceeds to step S32 if the judgment is necessary (step S31 → YES), and if the judgment is unnecessary (step S31 → YES). NO) Proceed to step S34. The valid analysis period / valid data item extraction unit 161 determines the necessity of determining the valid analysis period by acquiring the necessity flag of the valid analysis period determination set by the user and stored in the operation data DB 130. Alternatively, the user may be inquired and determined via the external device 210.

In step S32, the effective analysis period / effective data item extraction unit 161 excludes the test period, the maintenance period, and the outage period of the plant 220 from the effective analysis period, and sets a new effective analysis period. The valid analysis period / valid data item extraction unit 161 acquires the test period, the maintenance period, and the stop period from the test / maintenance / stop information DB 310 (see FIG. 2).

In step S33, the effective analysis period / effective data item extraction unit 161 further excludes the failure period of the plant 220 from the effective analysis period, and sets a new effective analysis period. The valid analysis period / valid data item extraction unit 161 acquires the failure period from the failure occurrence information DB 320 (see FIG. 3).

In step S34, the valid analysis period / valid data item extraction unit 161 determines whether or not the data item is specified, and if specified (step S34 → YES), proceeds to step S35, and if not specified (step S34 → NO). The process proceeds to step S36. The valid analysis period / valid data item extraction unit 161 may determine the validity of the data item designation by acquiring the validity flag of the data item designation set by the user and stored in the operation data DB 130. , The user may be inquired and decided via the external device 210.

In step S35, the valid analysis period / valid data item extraction unit 161 extracts data items including keywords. Specifically, the valid analysis period / valid data item extraction unit 161 includes keywords stored in the operation variable keyword DB 340 (see FIG. 5) among the data items stored in the operation data item DB 330 (see FIG. 4). Extract data items. By this process, the valid analysis period / valid data item extraction unit 161 extracts the data item which is an instrumental variable.

In step S36, the valid analysis period / valid data item extraction unit 161 is the data item (in the case of step S34 → YES) that is the manipulated variable extracted in step S35, or the data item corresponding to all the manipulated variables (step S34 → NO). In the case of), the process of repeating steps S37 to S40 is started. Hereinafter, the data item to be repeatedly processed will be referred to as a data item to be processed.
In step S37, the valid analysis period / valid data item extraction unit 161 determines the presence / absence of failure sensor information. Specifically, the valid analysis period / valid data item extraction unit 161 proceeds to step S38 if the failure information of the sensor information corresponding to the data item to be processed is in the failure sensor DB 350 (see FIG. 6) (step S37 → YES). If not (step S37 → NO), the process proceeds to step S41.

In step S38, the valid analysis period / valid data item extraction unit 161 determines whether or not the failure period is within a predetermined ratio of the analysis period. Specifically, in the valid analysis period / valid data item extraction unit 161, if the period in which the failure period of the sensor related to the data item to be processed and the valid analysis period overlap is within a predetermined ratio of the valid analysis period (step S38 → YES) Proceed to step S40, and if the predetermined ratio is exceeded (step S38 → NO), the process proceeds to step S39.

In step S39, the valid analysis period / valid data item extraction unit 161 determines that the data item of the failure sensor is invalid and excludes it. Specifically, the valid analysis period / valid data item extraction unit 161 determines that the data item to be processed is an invalid data item and excludes it from the valid data items to be analyzed.
In step S40, the effective analysis period / effective data item extraction unit 161 excludes the sensor failure period from the effective analysis period, and sets a new effective analysis period.
In step S41, the valid analysis period / valid data item extraction unit 161 returns to step S37 if there is an unprocessed data item to be processed, and ends the extraction process of the valid analysis period / valid data item if there is not.

≪Data item evaluation score calculation process≫
FIG. 10 is a flowchart of the evaluation score calculation process of the data item according to the present embodiment. The evaluation score calculation process (see step S14 of FIG. 8) of the data item executed by the instrumental variable determination unit 160 will be described with reference to FIG.
In step S51, the data item evaluation unit 165 determines whether or not the evaluation points of the evaluation items (correlation degree and coefficient of variation) need to be set, and if the setting is necessary (step S51 → YES), the process proceeds to step S52, and if the setting is unnecessary (step S51 → YES). Step S51 → NO) Proceed to step S53. The data item evaluation unit 165 may determine the necessity of setting the evaluation points of the evaluation items by acquiring the necessity flag of the evaluation points of the evaluation items set by the user and stored in the operation data DB 130. Alternatively, the user may be inquired and determined via the external device 210.

In step S52, the evaluation point acquisition unit 163 determines the evaluation points of the evaluation items. Specifically, the evaluation point acquisition unit 163 may acquire and determine the evaluation points of the evaluation items set by the user and stored in the operation data DB 130, or the external device 210. You may make a decision by contacting the user via.
In step S53, the weighting determination unit 164 determines the weighting of the evaluation item from the evaluation points.

In step S54, the data item evaluation unit 165 calculates the evaluation score based on the correlation degree and the coefficient of variation calculated in step S13 (see FIG. 8) and the weighting calculated in step S53, and evaluates the evaluation score DB 360 (see FIG. 8). It is stored in the column of the evaluation score 365 (see FIG. 7).

<< Instrumental variable selection process >>
FIG. 11 is a flowchart of the operation variable selection process according to the present embodiment. The operation variable selection process (see step S15 in FIG. 8) executed by the instrumental variable selection unit 166 of the instrumental variable determination unit 160 will be described with reference to FIG. The data items corresponding to the manipulated variables included in the evaluation score DB 360 shown in FIG. 7 are referred to as selection candidate data items. In FIG. 11, the selection candidate data item is simply referred to as a data item.
In step S61, the instrumental variable selection unit 166 proceeds to step S62 if there is a selection candidate data item whose evaluation score (see the evaluation score column of the evaluation score DB 360 shown in FIG. 7) is equal to or less than a predetermined threshold value (step S61 → YES). If not, the process proceeds to step S63 (step S61 → NO).
In step S62, the instrumental variable selection unit 166 excludes data items whose evaluation score is equal to or less than a predetermined threshold value from the selection candidate data items, and newly sets them as selection candidate data items.

In step S63, the operation variable selection unit 166 newly extracts a predetermined number of data items having the highest evaluation score among the selection candidate data items as selection candidate data items.
In step S64, the instrumental variable selection unit 166 determines whether or not the drawing of the selection candidate data item needs to be confirmed, proceeds to step S65 if confirmation is required (step S64 → YES), and operates if confirmation is not required (step S64 → NO). Finish the variable selection process. The operation variable selection unit 166 determines the necessity of drawing confirmation of the selection candidate data item by acquiring the necessity flag of the drawing confirmation of the selection candidate data item set by the user and stored in the operation data DB 130. Alternatively, the user may be inquired and determined via the external device 210.

In step S65, the instrumental variable selection unit 166 displays the instrumental variables using the drawings. Specifically, the instrumental variable selection unit 166 displays a control logic circuit diagram (see FIG. 12 described later) on the display of the external device 210, highlights the instrumental variables corresponding to the selection candidate data items on the diagram, and highlights them. Prompt the user for confirmation.

FIG. 12 is a diagram showing an example of the control logic circuit diagram 410 according to the present embodiment. The instrumental variable 411 "air flow rate set value" in the control logic circuit diagram 410 is highlighted. By referring to the control logic circuit diagram 410, the user can grasp and confirm the relationship between the operation variables that are candidates for selection and other operation variables, measured values, and circuits.

≪Characteristics of plant control support device≫
The plant control support device 100 selects an instrumental variable (data item) having a large correlation with the operation performance index and a coefficient of variation. When calculating the degree of correlation and the coefficient of variation, the plant control support device 100 does not calculate by referring to all the operation data stored in the operation data DB 140, but the stop period, maintenance period, and sensor of the plant 220. By excluding the failure period of the above, the calculation is limited to the period during which the plant is operating normally and the data is valid.

By selecting an instrumental variable that has a high degree of correlation with the operation performance index, the operation performance of the plant 220 can be improved by setting a small number of instrumental variables. In addition, by selecting an instrumental variable with a large coefficient of variation, it is possible to select an instrumental variable while avoiding unnecessary manipulated variables that are not normally manipulated.
By setting the evaluation points, the user can set the degree of correlation and the weighting of the manipulated variables in selecting the manipulated variables. In addition, the user can set candidates for the selected operation variable by specifying a keyword. The characteristic curve between the value of the manipulated variable and the operational performance index is displayed, and the user can confirm the relationship between the manipulated variable and the operational performance index.
Instrumental variables whose evaluation score is lower than a predetermined value are not selected, and instrumental variables whose correlation with the operational performance index and operation frequency are low are not selected. In addition, by calculating the correlation degree and coefficient of variation from the operation data during the period when the data is valid, it is possible to select the manipulated variables that contribute to the improvement of the operational performance index with high accuracy.

Further, the plant control support device 100 calculates the operation amount at which the operation performance index is the best by using the operation learning model 172 that controls the statistical model 171 showing the relationship between the operation variable and the operation performance index. By controlling the plant 220 based on the calculated operation amount, the operation performance index can be improved.

≪Modification example: Outliers≫
In step S13 (see FIG. 8) in the above-described embodiment, the evaluation item generation unit 162 calculates the correlation degree and the coefficient of variation from the operation data of the effective analysis period excluding the test period including invalid data. The evaluation item generation unit 162 may further exclude outliers from the operation data to calculate the correlation degree and the coefficient of variation. For example, outliers may be excluded using the Smirnov-Grabs test or interquartile range.

≪Transformation example: Invalid data≫
In steps S37 to S40 of the effective analysis period / valid data item extraction process (see FIG. 9) in the above-described embodiment, the sensor failure is performed from the effective analysis period according to the ratio of the sensor failure period in the effective analysis period. The period is excluded, or the data item itself is excluded. On the other hand, the sensor failure period may be excluded from the effective analysis period, or the data item itself may be excluded without considering the ratio.

≪Modification example: Coefficient of variation≫
In the above-described embodiment, the coefficient of variation indicating the operation frequency of the manipulated variable is calculated using the equation (1). The operation frequency of the operation variable may be the number of times the operation variable (set value) is changed within a period of a predetermined length.
Further, if the number of times the manipulated variable is changed per predetermined length period is less than or equal to the predetermined value, the valid analysis period / valid data item extraction unit 161 may remove the manipulated variable from the valid data item as an invalid data item. good.

≪Other variants≫
The present invention is not limited to the above-described embodiment, and can be modified without departing from the spirit of the present invention. For example, the operation data DB 130 and the operation data DB 140 are stored in the plant control support device 100, but may be stored in a device different from the plant control support device 100.

Although some embodiments of the present invention have been described above, these embodiments are merely examples and do not limit the technical scope of the present invention. The present invention can take various other embodiments, and various modifications such as omission and substitution can be made without departing from the gist of the present invention. These embodiments and modifications thereof are included in the scope and gist of the invention described in the present specification and the like, and are also included in the scope of the invention described in the claims and the equivalent scope thereof.

100 Plant control support device 130 Operation data DB
140 Operation data DB (operation data)
160 Instrumental variable determination unit 170 Operation amount determination unit 171 Statistical model 172 Operation learning model (machine learning model)
210 External device 220 Plant 360 Evaluation score DB
361 Identification number 362 Data item (instrumental variable)
363 Correlation degree (absolute value of correlation coefficient with operational performance index)
364 Coefficient of variation (operation frequency)
365 rating score

Claims (14)

A plant control support device that selects instrumental variables that contribute to the improvement of plant operation performance indicators.
A plant control support device including an instrumental variable determination unit that selects the instrumental variables based on the degree of correlation with the operation performance index and the operation frequency. The instrumental variable determination unit is characterized in that the correlation degree and the operation frequency are calculated based on the operation data of the effective analysis period, which is the period during which the plant is operating normally, in the operation data of the plant. The plant control support device according to claim 1. The plant control support device according to claim 1, wherein the operation frequency is a coefficient of variation of the operation variable. The plant control support device according to claim 1, wherein the operation frequency is the number of changes of the operation variable in a predetermined length period. The plant control support device according to claim 1, wherein the degree of correlation is an absolute value of a correlation coefficient between the manipulated variable and the operational performance index. The plant control according to claim 1, wherein the instrumental variable determination unit selects the instrumental variable based on an evaluation score calculated by giving a predetermined weight to the correlation degree and the operation frequency. Support device. The plant control support device according to claim 6, wherein the instrumental variable determination unit excludes instrumental variables whose evaluation score is lower than a predetermined value. The plant according to claim 2, wherein the instrumental variable determination unit calculates the correlation degree and the operation frequency based on the operation data excluding the operation data acquired by the failed device from the operation data. Control support device. The operation variable determination unit approximates the characteristic curve between the value of the operation variable and the operation performance index with one of a predetermined characteristic curve including a linear function and a Poisson distribution function, and displays it on an external device. The plant control support device according to claim 1. The operation variable determination unit according to claim 1, wherein the operation variable determination unit selects an operation variable that contributes to the improvement of the operation performance index from the operation variables including a predetermined keyword in the name among the operation variables. Plant control support device. A machine learning model that controls a statistical model that simulates the relationship between the instrumental variables selected by the instrumental variable determination unit and the instrumental performance index is trained, and the trained machine learning model is used to optimize the instrumental performance index. The plant control support device according to claim 1, further comprising an operation amount determination unit for determining a set value of an instrumental variable. The plant control support device according to claim 11, wherein the machine learning model is a reinforcement learning model. A program for operating a computer as a plant control support device according to any one of claims 1 to 12. It is a plant control support method of a plant control support device that selects instrumental variables that contribute to the improvement of the plant operation performance index.
A plant control support method characterized by executing a step of selecting the instrumental variables based on the degree of correlation with the operation performance index and the operation frequency.

Images