TW202247889A - Device, remote monitoring system, method for controlling device, and method for controlling remote monitoring system - Google Patents

Abstract

This device is for executing control of a plant on the basis of a prediction result obtained using a learning model. When a prediction result obtained using the learning model satisfies a prescribed condition, the device selects, from operational data as additional training data, data that is largely divergent from training data that was used to construct the learning model. Thereafter, new training data including the training data and the additional training data is used to reconstruct the learning model.

Description

Device, remote monitoring system, device control method, and remote monitoring system control method

This case is related to the device, the remote monitoring system, the control method of the device and the control method of the remote monitoring system. In this case, priority is claimed based on Japanese Patent Application No. 2021-061268 filed with the Japan Patent Office on March 31, 2021, and the contents thereof are incorporated herein.

In the case of a wet exhaust gas desulfurization device, which is an example of a plant, the exhaust gas generated by a combustion device such as a boiler is introduced into the absorption tower of the desulfurization device, and the exhaust gas is circulated in the absorption tower. Absorb liquid gas liquid contact. In the process of gas-liquid contact, the SO ₂ in the exhaust gas will be absorbed in the absorbing liquid through the reaction of the absorbent (such as calcium carbonate) in the absorption liquid with the sulfur dioxide (SO ₂ ) in the exhaust gas, and the SO ₂ (exhaust gas for desulfurization). On the other hand, the absorption liquid after absorbing SO ₂ falls down and is stored in the storage tank below the absorption tower. Absorbent is supplied to the storage tank, and the absorption liquid that restores the absorption performance with the supplied absorbent is supplied to the upper part of the absorption tower by a circulation pump for gas-liquid contact with exhaust gas (absorption of SO ₂ ).

In such a wet-type exhaust gas desulfurization device, it takes a long time until changes in the circulation flow rate of the absorbing liquid or the concentration of the absorbent are reflected in the SO ₂ concentration in the exhaust gas. Therefore, as far as the control of the wet exhaust gas desulfurization device is concerned, the relationship between the control parameters such as the circulation flow rate of the absorbing liquid or the concentration of the absorbent and the concentration of _SO2 in the exhaust gas is constructed by machine learning. As a learning model, The SO ₂ concentration in the exhaust gas can be kept below the reference value by determining the circulation flow rate of the absorbing liquid or the control target value of the absorbent concentration based on the SO ₂ concentration in the exhaust gas predicted by the learning model. For example, Patent Document 1 discloses that in the control of such a wet _- type exhaust gas desulfurization device, the first learning model representing the relationship between the amount of circulation of the absorbing liquid and the concentration of SO in the exhaust gas is used to obtain the relevant The control target value of the circulation amount of the absorption liquid, and the _second learning model representing the relationship between the absorbent concentration of the absorption liquid and the SO2 concentration in the exhaust gas is used to obtain the control target value of the absorbent concentration of the absorption liquid, thereby A method of controlling the circulation flow rate of the absorbent or the concentration of the absorbent. [Prior Art Document] [Patent Document]

[Patent Document 1] Japanese Patent Laid-Open No. 2020-11163

(Problem to be solved by the invention)

In the aforementioned factory equipment control, a learning model for predicting a control target value is prepared in advance. The construction of such a learning model is carried out by machine learning using learning data selected from operating data of factory equipment. However, since operating data contains many deviations, it may not be possible to obtain sufficient data depending on the selection method of learning data. case of forecast error. Also, even if the prediction error of the learning model is very small at the beginning, there may be cases where the prediction error becomes larger later on due to changes in the operating conditions of the factory equipment and the construction of the learning model.

When the prediction error of the learning model is insufficient in this way, it is necessary to reconstruct the learning model. The reconstruction of the learning model is carried out, for example, by using learning data to which new data has been added, but since the operation data contains many deviations, the prediction error may also increase depending on the selection of the added data. Furthermore, since the operation data used as the basis of the learning data is huge, it is required to efficiently select the data for reducing the prediction error by reconstructing the learning model.

At least one embodiment of this case was developed in view of the above-mentioned circumstances to provide a device, a remote monitoring system, The control method of the device and the control method of the remote monitoring system are aimed at. (means to solve the problem)

In order to solve the above-mentioned problems, at least one embodiment of the present invention is a device for performing processing related to plant equipment control based on prediction results using a learning model, and is characterized by: The additional learning data selection unit is used to select, from the aforementioned operation data, data with a high degree of deviation from the learning data used in the construction of the aforementioned learning model as additional learning when the prediction result using the aforementioned learning model meets a predetermined condition. information; and The learning model construction unit is used to reconstruct the aforementioned learning model using new learning materials including the aforementioned learning materials and the aforementioned additional learning materials.

In order to solve the above-mentioned problems, at least one embodiment of the remote monitoring system of this application is a remote monitoring system composed of a terminal device capable of communicating with a device for performing processing related to plant equipment control based on prediction results using a learning model. The system is characterized in that the aforementioned device is equipped with: The additional learning data selection unit is used to select from the operation data the data from the learning data used in the construction of the learning model when the prediction result using the learning model satisfies a predetermined condition based on the request from the terminal device. Data with a large degree of deviation are used as additional learning materials; and The learning model constructing unit is used to reconstruct the aforementioned learning model by using the learning materials including the aforementioned learning materials and the aforementioned additional learning materials.

In order to solve the above-mentioned problems, at least one embodiment of the present invention is a method of controlling a device, which is a method of controlling a device for performing processing related to control of plant equipment based on a prediction result using a learning model, and is characterized in that it includes: The step of selecting additional learning materials, which is to select, from the aforementioned operation data, materials with a high degree of deviation from the learning data used in the construction of the aforementioned learning model as additional learning materials when the prediction result using the aforementioned learning model meets a predetermined condition; and The step of constructing the learning model is to construct the learning model again by using the learning material including the learning material and the additional learning material.

In order to solve the above-mentioned problems, at least one embodiment of the present application provides a method for controlling a remote monitoring system comprising a terminal device capable of communicating with a device for performing processing related to plant equipment control based on a prediction result using a learning model. The control method of the remote monitoring system is characterized by: The additional learning data selection step is to select the degree of deviation from the learning data used in the construction of the learning model from the operation data when the prediction result using the learning model satisfies a predetermined condition based on the request from the terminal device. large data as additional learning materials; and The step of constructing the learning model is to construct the learning model again by using the learning material including the learning material and the additional learning material. [Effect of the invention]

According to at least one embodiment of the present invention, it is possible to provide a device, a remote monitoring system, a device control method, and remote monitoring that achieve good control accuracy by efficiently selecting learning materials used in the reconstruction of the learning model. system control method.

Hereinafter, several embodiments of the present invention will be described with reference to the drawings. However, the scope of the present invention is not limited to the following embodiments. The dimensions, materials, shapes, and relative arrangements of components described in the following embodiments do not limit the scope of the present invention thereto, but are merely illustrative examples.

First, the configuration of a wet-type exhaust gas desulfurization device 10 , which is an example of a plant facility to be controlled by the plant facility control device according to at least one embodiment of the present invention, will be described with reference to FIG. 1 . Fig. 1 is a configuration diagram of a wet-type exhaust gas desulfurization device 10 according to an embodiment.

In addition, in the following description, the wet exhaust gas desulfurization device 10 is described as an example of factory equipment, but the control object is not limited to the wet exhaust gas desulfurization device 10, and the control parameters can be widely included. And predict the control target value to control the plant equipment.

The wet exhaust gas desulfurization device 10 is plant equipment for desulfurizing the exhaust gas generated in the combustion device 1 . The combustion device 1 is, for example, a boiler for generating steam, and is configured as a part of power generation plant equipment capable of generating electricity by supplying the steam generated by the combustion device 1 to the generator 5 . The wet-type exhaust gas desulfurization device 10 is provided with: an absorption tower 11 communicated with the combustion device 1 through a pipe 2; a plurality of circulation pumps 12a, 12b provided in the pipe 3 for circulation of the absorption liquid circulating in the absorption tower 11 , 12c (these are collectively referred to as "circulation pump 12"appropriately); for supplying the absorbent contained in the absorption liquid, that is, calcium carbonate (CaCO ₃ ) slurry (slurry) (absorbent slurry) to The absorbent slurry supply part 13 in the absorption tower 11; and the gypsum recovery part 14 for recovering the gypsum in the absorption liquid. The absorption tower 11 is provided with an outflow pipe 16 for allowing the exhaust gas desulfurized in the operation described later to flow out from the absorption tower 11 as an outflow gas, and a gas analyzer ₁₇ for measuring the concentration of SO in the outflow gas is provided in the outflow pipe 16 .

The absorbent slurry supply unit 13 is equipped with: absorbent slurry manufacturing plant 21 for manufacturing absorbent slurry; Absorbent slurry supply piping 22 connecting the absorbent slurry manufacturing equipment 21 with the absorption tower 11; The absorbent slurry supply amount control valve 23 for controlling the flow rate of the absorbent slurry flowing through the absorbent slurry supply piping 22 . The gypsum recovery unit 14 is equipped with: Gypsum separator 25; A pipe 26 for extracting gypsum slurry that connects the gypsum separator 25 to the absorption tower 11; and The pump 27 for pumping out gypsum slurry is installed in the piping 26 for pumping out gypsum slurry.

The wet-type exhaust gas desulfurization device 10 is provided with a control device 15 that is a device that controls plant equipment in at least one embodiment of the present invention. The control device 15 is equipped with an operation data receiving unit 30 electrically connected to the operation data acquisition unit 20. The operation data acquisition unit 20 includes various operation data (such as Various detectors for temperature or pressure of various parts, flow rate of various fluids, etc.). The operation data acquisition unit 20 includes a gas analyzer 17 .

The control device 15 is equipped with: being electrically connected to the first learning model building unit 38 of the operation data receiving unit 30; electrically connected to the first relationship table creation unit 31 of the first learning model construction unit 38; is electrically connected to the circulation flow rate determination unit 32 of the first relationship table creation unit 31; and The circulating pump adjusting part 33 is electrically connected to the circulating flow determining part 32 . The circulation pump regulator 33 is electrically connected to each of the circulation pumps 12a, 12b, 12c.

The control device 15 is further equipped with: is electrically connected to the second learning model construction unit 39 of the operation data receiving unit 30; is electrically connected to the second relationship table creation part 35 of the second learning model construction part 39; The absorbent slurry supply amount determination unit 36 electrically connected to the second relationship table creation unit 35; and The absorbent slurry supply control unit 37 is electrically connected to the absorbent slurry supply amount determination unit 36 . The absorbent slurry supply control unit 37 is electrically connected to the absorbent slurry supply control valve 23 .

The control device 15 further has: The prediction error calculation unit 40 electrically connected to the first learning model construction unit 38 and the second learning model construction unit 39; and It is electrically connected to the additional learning material selection unit 42 of the prediction error calculation unit 40 .

FIG. 2 shows the configuration of a remote monitoring system 44 for remotely monitoring the control state of the wet-type exhaust gas desulfurization device 10 (see FIG. 1 ). The remote monitoring system 44 is equipped with: Distributed control system (DCS) 46 of each machine constituting the combustion device 1 (refer to FIG. 1 ) and the wet exhaust gas desulfurization device 10 (refer to FIG. 1 ); An edge server 48 electrically connected to the DCS 46 and carrying the control device 15; A remote monitoring device 50 such as a desktop personal computer or a tablet computer that is electrically connected to the edge server 48 via a cloud or a virtual private network (Virtual Private Network; VPN). Usually, the control state of the wet exhaust gas desulfurization device 10 can be remotely monitored by the remote monitoring device 50 existing in a place away from the edge server 48 .

Next, the operation of the wet exhaust gas desulfurization device 10 to desulfurize the exhaust gas generated in the combustion device 1 will be described. As shown in FIG. 1 , the exhaust gas generated in the combustion device 1 flows through the pipe 2 to flow into the absorption tower 11 and rises in the absorption tower 11 . By operating at least one of the circulation pumps 12 , the absorption liquid flows through the circulation piping 3 to flow into the absorption tower 11 , and the absorption liquid flows down in the absorption tower 11 . The absorption liquid flowing down in the absorption tower 11 is accumulated in the absorption tower 11 , flows out from the absorption tower 11 by the circulation pump 12 , and circulates through the piping 3 for circulation. In this way, the absorption liquid will circulate in the absorption tower 11 .

In the absorption tower 11, the ascending exhaust gas and the absorbing liquid flowing down are brought into gas-liquid contact. SO ₂ contained in exhaust gas reacts with CaCO ₃ in the absorption liquid as shown in the following reaction formula, and gypsum (CaSO ₄ ･2H ₂ O) is precipitated in the absorption liquid. SO ₂ +CaCO ₃ +2H ₂ O+1/2O ₂ →CaSO ₄ ･2H ₂ O+CO ₂

_In this way, a part of SO2 in the exhaust gas will be removed as gypsum in the absorption liquid, that is, the exhaust gas will be desulfurized, so the SO2 concentration in the effluent gas flowing out from the absorption tower 11 through the outflow pipe ₁₆ is relatively low. The concentration of SO ₂ in the exhaust gas flowing into the absorption tower 11 through the pipe 2 is even lower. The effluent gas flowing out from the absorption tower 11 is released into the atmosphere through the outflow pipe ₁₆ , but the SO2 concentration is measured by the gas analyzer 17 during the process, and the measurement result is transmitted to the control device 15 for operation data reception. Section 30.

The SO ₂ concentration in the effluent gas tends to decrease as the circulation flow rate of the absorbing liquid circulating in the absorption tower 11 increases, unless there is a large change in the CaCO ₃ concentration in the absorbing liquid. By controlling the number of circulation pumps ₁₂ operated by the control device 15 according to the control method described later to control the circulation flow rate, the SO2 concentration in the outflow gas can be controlled, for example, the outflow can be controlled to be below a preset set value. _SO2 concentration in the gas.

The gypsum precipitated in the absorption liquid in the absorption tower 11 is extracted from the absorption tower 11 as gypsum slurry by the pump 27 for extracting the gypsum slurry, and the gypsum slurry flows into the gypsum separator 25 through the pipe 26 for extracting the gypsum slurry. In the gypsum separator 25, gypsum and water are separated, the gypsum is recovered, and the water is sent to a drainage facility (not shown).

Since CaCO ₃ in the absorption liquid reacts with SO ₂ to become gypsum, the concentration of CaCO ₃ in the absorption liquid will decrease as the desulfurization of exhaust gas proceeds. The control device 15 controls the opening degree of the absorbent slurry supply control valve 23 according to the control method described later, and supplies the absorbent slurry produced in the absorbent slurry manufacturing facility 21 to the absorption tower 11 through the absorbent slurry supply piping 22 Inside. Thereby, the CaCO ₃ concentration in the absorbing liquid falls within a preset range, and large fluctuations in the CaCO ₃ concentration during desulfurization of exhaust gas are suppressed.

Next, basic control of the wet-type exhaust gas desulfurization device 10 related to the control device 15 will be described. FIG. 3 is a flowchart showing basic control of the wet-type exhaust gas desulfurization device 10 performed by the control device 15 of FIG. 1 .

As far as the basic control is concerned, after collecting various operating data of the combustion device 1 and the wet exhaust gas desulfurization device 10 in step S1, in step S2, for the various operating data and the effluent gas from the absorption tower 11 The relationship between the future SO ₂ concentration is constructed by machine learning as the first learning model. Next, in step S3, a first relational table described later is created using the constructed first learning model. _In the subsequent step S4, the circulation flow rate of the absorbing liquid at which the concentration of SO2 in the effluent gas becomes below the preset set value is determined according to the first relational table, and in step S5, the circulation is adjusted according to the determined circulation flow rate. The operating conditions of the pump 12. Thereby, the SO ₂ concentration in the effluent gas will be controlled so that it becomes below a preset set value.

Further, after step S1, in addition to steps S2 to S5, in step S12, a second learning model is constructed by machine learning for the relationship between various operating data and the future CaCO ₃ concentration in the absorbent. Next, in step S13, a second relational table described later is created using the constructed second learning model. In the subsequent step S14, the supply amount _of the absorbent slurry whose CaCO concentration is within the preset range is determined according to the second relational table, and in step S15, by controlling the absorbent slurry supply unit 13, that is The opening of the valve 23 is controlled for the amount of the agent slurry supply, and the absorbent slurry is supplied into the absorption tower 11 at the determined supply amount. Thereby, the CaCO ₃ concentration in the absorbing liquid falls within a preset range, and large fluctuations in the CaCO ₃ concentration during desulfurization of exhaust gas are suppressed.

Next, each step of the control method of the wet exhaust gas desulfurization device 10 related to the control device 15 will be described in detail. As for step S1, as shown in FIG. 1 , after the operation data obtaining unit 20 obtains various operation data of the combustion device 1 and the wet exhaust gas desulfurization device 10, the obtained various operation data will be transmitted to the control device 15. And the operation data receiving unit 30 receives it, so that the control device 15 collects various operation data. As mentioned above, the operation data acquisition unit 20 includes the gas analyzer 17, so various operation data include the concentration of SO ₂ in the effluent gas.

In step S2, the first learning model construction unit 38 constructs a first model by machine learning for the relationship between various operating data collected by the control device 15 and the future SO ₂ concentration in the outflow gas. In step S3, the first relational table creation unit 31 creates the circulation flow rate of the absorbent at the first time and the flow rate of the outflow gas at the second time later than the first time using the constructed first learning model. 1st relational table of cross _- correlation of predicted values of SO2 concentration. Since the first relational table is created using the first learning model constructed by machine learning, the first relational table can be quickly created.

In the first relational table, the circulation flow rate of the absorption liquid and the predicted value of the _SO2 concentration in the effluent gas are different in time. If the circulation flow rate of the absorption liquid is set to the current value, the prediction value of the _SO2 concentration in the effluent gas The value is, for example, a predicted value of the SO ₂ concentration several minutes from now. Therefore, the various operation data include at least the concentration of SO2 in the effluent gas at an arbitrary time and the circulation flow rate of the absorbing liquid at a time when the time interval of subtracting the first time from the _second time is longer than the arbitrary time. It is directly predicted from the actual operation data including the concentration of SO2 in the effluent gas at an arbitrary time and the circulation flow rate of the absorption liquid at a time interval that is longer than an arbitrary time by subtracting the first time from the _second time. future SO ₂ concentrations, thus improving the predictive performance of future SO ₂ concentrations. In addition, the shorter the interval between the first time and the second time, the better the predictive performance of the future SO ₂ concentration. Therefore, the interval between the first time and the _second time is ideally the sum of the time required for the SO2 concentration in the effluent gas to change due to the change in the circulation flow rate of the absorbing liquid and the time required for the gas analyzer ₁₇ to measure the SO2 concentration.

In FIG. 4, it is shown that the interval between the first time and the _second time is defined as the time required for the SO2 concentration in the effluent gas to change due to the change in the circulation flow rate of the absorbing liquid and the time required for the gas analyzer ₁₇ to measure the SO2 concentration. Transition (a) of predicted value of SO ₂ concentration by time and time, transition (b) of measured value of SO ₂ concentration by gas analyzer 17, and transition (c) of true value of SO ₂ concentration. In each graph, the farther to the right is the past value, and the far left is the latest value. The latest value of the measured value of the SO ₂ concentration by the gas analyzer 17 is the value at the first time, and the latest value of the predicted value of the SO ₂ concentration is the value at the second time. The interval (i ₎ between the latest value of the measured value of the SO2 concentration by the gas analyzer ₁₇ and the latest value of the true value of the SO2 concentration is equivalent to the time required for the gas analyzer ₁₇ to measure the SO2 concentration, that is, the measurement delay, SO ₂ The interval (ii) between the latest value of the true value of the concentration and the latest value of the estimated value of the SO ₂ concentration corresponds to the time required until the SO ₂ concentration in the effluent gas changes due to a change in the circulation flow rate of the absorbing liquid.

An example of the first relationship table is shown in FIG. 5 . In this embodiment, the first relational table is a graph in which the predicted value of the SO2 concentration in the effluent gas is taken _on the horizontal axis, and the circulation flow rate of the absorbing liquid is taken on the vertical axis, but this does not have to be the case, and may be The form of a matrix or a mathematical expression, etc. In step S4, the circulation flow rate determination unit 32 determines the circulation flow rate Q (control target value) of the absorbent for which the SO ₂ concentration in the effluent gas becomes the preset set value SV in the future based on the first relational table.

In step S5, as shown in FIG. 1 , the circulating pump regulator 33 determines the operating number of the circulating pumps 12a to 12c so as to be equal to or greater than the determined circulating flow rate Q (control target value), and makes the determined The number of circulation pumps operating is operated. For example, when the supply amounts of the three circulation pumps 12a to 12c are the same during operation, the circulation flow rate can be adjusted in three stages. As long as the number of circulation pumps is increased, finer adjustment of the circulation flow can be achieved. Also, for example, when the supply amounts of the three circulating pumps 12a to 12c are different from each other during operation, the circulating flow rates of six stages can be adjusted at most by combining the operating circulating pumps. Further, for example, if each of the three circulation pumps 12a to 12c can adjust the supply amount, finer adjustment of the circulation flow rate is possible.

In addition, the adjustment of the circulating flow rate is not limited to those performed by controlling the number of circulating pumps. The supply rate of the circulation pump may be adjusted using one circulating pump whose supply rate can be adjusted so that the circulation flow rate determined by the circulation flow rate determination unit 32 is used.

By adjusting the circulation flow rate of the absorption liquid circulating in the absorption tower ₁₁ in this way, the SO2 concentration in the future effluent gas can be controlled to be below the preset set value, but for this, the _CaCO concentration in the absorption liquid needs to be constant big change. Therefore, in this embodiment, as described above, in addition to steps S2 to S5, the concentration of CaCO ₃ in the absorption liquid is controlled to be within a preset range through steps S12 to S15. Next, each of steps S12 to S15 will be described in detail.

As far as step S12 is concerned, the second learning model construction unit 39 constructs a second learning model by machine learning for the relationship between various operating data collected by the control device ₁₅ and the future CaCO concentration in the absorption liquid in the absorption tower 11. learning model. In step S13, using the constructed second learning model, the second relational table creation unit 35 creates the second relationship between the supply amount of the absorbent slurry to the absorption tower 11 at the third time and the time later than the third time. 2nd relational table of cross-correlation of predicted values of CaCO ₃ concentration at 4 times. Since the second relationship table is created using the second learning model constructed by machine learning, the second relationship table can be quickly created.

In the second relational table, the supply amount of the absorbent slurry to the absorption tower 11 is different from the predicted value of the CaCO ₃ concentration at different times. If the supply amount of the absorbent slurry is the current value, the predicted value of the CaCO ₃ concentration It is, for example, an estimated value of the CaCO ₃ concentration several minutes from now. Therefore, the various operation data include at least the CaCO ₃ concentration at an arbitrary time and the supply amount of absorbent slurry at a time after the time interval of subtracting the third time from the fourth time is longer than the aforementioned arbitrary time. The future is directly predicted from the actual operation data including the CaCO ₃ concentration at any time and the supply amount of absorbent slurry at a time that has elapsed from the time interval of subtracting the third time from the fourth time than the aforementioned arbitrary time. CaCO ₃ concentration, thus improving the predictive performance of future CaCO ₃ concentrations.

In this embodiment, the CaCO ₃ concentration at an arbitrary time is a value calculated using a simulation model based on mass balance calculation. Since a sensor for detecting the concentration of CaCO ₃ is generally expensive, if such a sensor is installed, the cost of the wet-type exhaust gas desulfurization device 10 will increase. However, if the CaCO ₃ concentration is calculated using a simulation model based on mass balance calculation, an expensive sensor is not required, and an increase in the cost of the wet-type exhaust gas desulfurization device 10 can be suppressed.

In addition, the shorter the interval between the third time and the fourth time, the better the prediction performance of the future CaCO ₃ concentration. Therefore, the interval between the third time and the fourth time is ideally set to be the time required until the CaCO ₃ concentration changes due to the change in the supply amount of the absorbent slurry. The transition of the predicted value and the transition of the true value of the supply amount of absorbent slurry have the same relationship as the transition of the predicted value (a) and the transition (c) of the true value of the SO ₂ concentration in FIG. 4 , respectively. In this embodiment, the CaCO ₃ concentration is calculated using a simulation model based on mass balance calculations, but when the CaCO ₃ concentration is measured by a sensor, the transition of the predicted value of the supply amount of the absorbent slurry and the sensor The transition of the measured value and the transition of the true value respectively form the same relationship as the various transitions (a) to (c) of the SO ₂ concentration in FIG. 4 .

Generally, the number of steps necessary to change the concentration of SO in the effluent gas from the absorption tower ₁₁ is more than the number of steps necessary to change the concentration of CaCO, so the control _of the concentration of SO is more difficult _than the control _of the concentration of CaCO. The delay is large. Therefore, by setting the time from the third time to the fourth time to be shorter than the time from the first time to the second time, the influence _of the control delay can be appropriately considered, so the future CaCO3 concentration can be further increased predictive performance.

An example of the second relationship table is shown in FIG. 6 . In this embodiment, the second relationship represents a graph in which the predicted value of CaCO ₃ concentration is taken on the horizontal axis and the supply amount of absorbent slurry is taken on the vertical axis, but this is not necessarily the case, and it may be a matrix or a mathematical formula etc. form. In step S14, the absorbent slurry supply amount determination unit 36 determines the absorbent slurry supply amount F (control target value) at which the future CaCO ₃ concentration falls within the preset range R based on the second relational table. .

In step S15, as shown in FIG. 1 , the absorbent slurry supply control unit 37 controls the opening of the absorbent slurry supply amount control valve 23 so that the absorbent slurry is supplied into the absorption tower 11 via the absorbent slurry supply piping 22 . The supply amount of the absorbent slurry is close to the determined absorbent slurry supply amount F (control target value). By adjusting the supply amount of the absorbent slurry to the absorption tower 11 in this way, it is possible to control the future CaCO ₃ concentration within a preset range.

Based on the operation data of the combustion device 1 and the operation data of the wet exhaust gas desulfurization device 10 including the circulation flow rate of the absorption liquid in this way, the circulation flow rate of the absorption liquid at the first time and the time later than the first time are prepared. The first relationship table between the SO ₂ concentration in the effluent gas flowing out of the absorption tower 11 at the second time, the future SO ₂ concentration can be directly predicted from the actual operation data, so the prediction of the future SO ₂ concentration can be improved The first relational table of performance, according to the first relational table, determine the circulation flow rate of the absorbing liquid at the first time when the concentration of SO ₂ in the effluent gas at the second time becomes below the preset set value, and at the first time Since the operation conditions of the circulation pumps 12a to 12c are adjusted according to the determined circulation flow rate, the operation conditions of the circulation pumps 12a to 12c can be appropriately adjusted.

As far as this embodiment is concerned, the CaCO ₃ concentration in the absorption liquid will be within the preset setting range through steps S12-S15, but for example, if the CaCO ₃ concentration in the absorption liquid is actually measured by a sensor, according to If the actual measurement value is used to adjust the supply amount of the absorbent slurry to the absorption tower 11 at any time, the steps of steps S12 to S15 are not required. In this case, the control device 15 may not include the second learning model constructing unit 39 , the second relationship table creating unit 35 , the absorbent slurry supply amount determining unit 36 , and the absorbent slurry supply control unit 37 .

Next, a plant equipment control method of an embodiment in which the basic control shown in FIG. 3 is added to the control device 15 will be described. Fig. 7 is a flowchart showing a plant equipment control method according to an embodiment.

In this factory equipment control, in addition to steps S2 to S5 shown in FIG. 3 , a predicted value of the first learning model is calculated in step S100 . Next, in step S101, the analysis result by the gas analyzer 17 is acquired. In the subsequent step S102, by comparing the predicted value calculated in step S100 with the analysis result obtained in step S101, the predicted result of the first learning model, that is, the predicted error is calculated. In the subsequent step S103, when the prediction error calculated in step S102 satisfies a predetermined condition, for example, it is determined that the prediction error is greater than a critical value. When the prediction error is larger than the critical value (step S103: YES), additional learning materials are selected in the subsequent step S104, and the first learning model is reconstructed in step S105 using the additional learning materials selected in step S104. Then, in step S106, a prediction error is calculated for the first learning model reconstructed in step S105, and it is determined in step S107 whether or not the prediction error is equal to or less than a critical value. When the prediction error calculated in step S106 is still greater than the critical value (step S107: NO), the process is returned to step S104, and selection of additional learning materials and reconstruction of the learning model are repeated. Such repetitive processing is performed until the predicted value of the reconstructed first learning model becomes below the critical value. In addition, when the prediction error in step S103 is equal to or less than the critical value (step S103: NO), the process ends, but a series of processes shown in FIG. 7 may be repeatedly performed at predetermined timings.

Next, each step related to Fig. 7 will be described in detail. In step S100 , the predicted value of the SO ₂ concentration in the outflow gas is calculated using the first learning model constructed by the first learning model construction unit 38 . The calculation of the predicted value based on the first learning model in step S100 is the same as in the case of calculating the predicted value of the _SO2 concentration in the effluent gas in order to create the first relational table in the aforementioned step S3. The circulation flow rate of the absorbing liquid at the first time is calculated, and the predicted value of the concentration of SO ₂ in the outflow gas at the second time in the future than the first time is calculated.

In step S101, the actual measurement value of the concentration of SO ₂ in the outflow gas is acquired based on the analysis result of the gas analyzer 17 . This actually measured value is the actual SO ₂ concentration in the outflow gas at the second time corresponding to the predicted value of the SO ₂ concentration in the outflow gas calculated in step S100.

In step S102, the prediction error calculating unit 40 calculates a prediction error as the difference between the predicted value of the _SO2 concentration in the outflow gas calculated in step S100 and the actual measurement value of the _SO2 concentration in the outflow gas obtained in step S101. Difference. This prediction error is an error corresponding to the prediction accuracy of the first learning model constructed by the first learning model construction unit 38 and includes various factors. For example, the operating data received by the operating data receiving unit 30 has many deviations, so the first learning model constructed by machine learning using the operating data as learning data has learning errors caused by the deviations. In addition, since the operating conditions of the plant equipment have changed since the time of model construction, there may be cases where the prediction error becomes large afterward.

In step S103, it is determined whether such a prediction error is larger than a preset threshold value ε. The success or failure determination of step S103 may also be performed as being true when the prediction error continues to be larger than the critical value ε for a predetermined time or more. The magnitude of the prediction error varies depending on the operating state of the wet exhaust gas desulfurization device 10. If it is assumed that a short-term determination is made in step S103, the reconstruction of the first learning model will be Frequent implementation may increase the burden of model management. Therefore, in step S103, if the state in which the prediction error is larger than the critical value ε continues for a predetermined period of time or longer, the establishment determination is made in step S103, whereby the reconstruction of the first learning model can be appropriately carried out, resulting in Efficient model management.

In step S104, when the determination in step S103 is established, the additional learning material selection unit 42 selects additional learning materials included in the learning materials used for reconstructing the first learning model. The learning materials used for reconstruction include new additional learning materials (that is, the learning materials used for reconstruction = Initial learning materials + additional learning materials). The operation data receiving unit 30 continuously receives the operation data, and selects appropriate additional learning data from the operation data received after the previous first learning model was constructed.

In addition, the selection of the additional learning data in step S104 may be carried out for the operation data acquired during the normal operation of the factory equipment. For example, operation data acquired during abnormal operation such as when an abnormality occurs in a factory facility, when the operation is started, and when the operation is stopped is excluded from the selection object of the additional learning data. In addition, when the operation data includes data obtained during such unsteady operation, it may be excluded by performing pre-processing on the operation data.

Here, referring to FIG. 8 , the method of selecting additional learning materials by the additional learning material selection unit 42 will be specifically described. FIG. 8 is a flowchart showing a method of selecting additional learning materials in step S104 of FIG. 7 .

In step S200 , first, by analyzing the operation data received by the operation data receiving unit 30 , at least one explanatory variable of the first learning model is selected from a plurality of parameters included in the operation data. Such explanatory variables can be selected, for example, for each of the plurality of operating data contained in the operating data, by means of multiple regression or the like, to the target variable of the first learning model, that is, the SO ₂ concentration in the effluent gas. Each contribution degree is calculated and performed based on the contribution degree. For example, Z parameters may be selected as explanatory variables in order of greater contribution. By selecting a part of the plurality of parameters included in the operation data as explanatory variables of the first learning model in this way, it is possible to suppress the decrease in learning accuracy compared to the case where all the parameters included in the operation data are set as learning objects. , while effectively reducing the amount of computation during learning.

In step S201, the explanatory variable selected in step S200 among the learning materials (operation data) constructed last time in the first learning model is selected as the initial learning data. At this time, the average value over W time may be used as the initial learning data for V selected from the learning data (operation data) used in the previous construction of the first learning model. In this case, by using the average value over a predetermined period of time as the learning data for a specific parameter included in the operation data, it is possible to effectively reduce the amount of computation during learning while suppressing a decrease in learning accuracy.

In step S202, an additional learning data candidate is selected from the operation data received by the operation data receiving unit 30 for the explanatory variable selected in step S200. The additional learning data candidates are selected from the new operation data received by the operation data receiving unit 30 during the period from the previous construction of the first learning model to the present, and include parameters corresponding to the initial learning data selected in step S201. . For example, when the average value over W time is used as the initial learning material as described above, the average value over W time may be used as an additional learning material candidate.

In step S203, the degree of divergence is calculated for the initial learning material selected in step S201 and the additional learning material candidates selected in step S202. The calculation of the degree of deviation can use, for example, K-nearest neighbor algorithm (k-nearest neighbor algorithm), Mahalanobis distance, etc., to evaluate the degree of deviation by various methods. Then, in step S204, additional learning materials to be added to the learning materials are selected based on the degree of deviation calculated in step S203.

Here, FIGS. 9A and 9B are diagrams showing the process of selecting additional learning materials in step S204 of FIG. 8 .

In the form of FIG. 9A, a plurality of additional learning material candidates Dc1 are displayed for a certain initial learning material Ds in a space defined by arbitrary variables 1 and 2 included in the explanatory variables of the first learning model. , Dc2 , Dc3 , ･･･, distances indicating degrees of deviation between the initial learning material Ds and each additional learning material candidate Dc1 , Dc2 , Dc3 , ･･･ are calculated. In this example, the additional learning material selection unit 42 selects the additional learning material candidate Dc5 having the largest distance among the plurality of additional learning material candidates as the additional learning material.

In addition, in the form of FIG. 9B, in the space defined by arbitrary variable 1 and variable 2 contained in the explanatory variables of the first learning model, for the plurality of additional learning material candidates selected in step S202, Dc1 , Dc2 , Dc3 , ..., display the plural initial learning materials Ds1 , Ds2 , ... selected in step S201 . Then, for each of the additional learning material candidates Dc1, Dc2, Dc3, ..., the distance to the nearest initial learning material is calculated. The additional learning material selection unit 42 selects the one with the largest distance among the plurality of additional learning material candidates as the additional learning material. In FIG. 9A and FIG. 9B, the number of variables used in the additional determination of learning materials is set to 2, but this is not limiting the scope of the present invention, and may be set to 1 or 3 or more during implementation.

In this way, the additional learning material selection unit 42 calculates the degree of deviation between the initial learning material and the additional learning material candidates, and selects the learning material candidates to be added to the learning materials for reconstructing the first learning model based on the degree of deviation. The number of newly added additional learning data can be arbitrary. For example, by selecting additional learning data with a degree of deviation equal to or greater than a predetermined value from the operating data, additional learning data of a number (A) determined from the one with the largest deviation can be selected. material.

In addition, in this embodiment, it is assumed that the first learning model has already been constructed by the first learning model construction unit 38, and the learning data used in the previous construction of the first learning model are used as the initial learning data. However, when there is no construction history of the first learning model (for example, when the first learning model is constructed for the first time), one or more parameters arbitrarily selected from the operation data may be processed as initial learning data. In this case, it is also possible to construct a learning model with a small prediction error when constructing the first learning model for the first time.

Returning to FIG. 7, in step S105, the additional learning data selected in step S104 is added to the initial learning data to create new learning data, and the first learning model is constructed again. This makes it possible to regenerate the first learning model by using new learning materials that add additional learning materials selected from operating data acquired later to the initial learning data used in the previous construction of the first learning model. build.

Then, in step S106, the prediction error is calculated using the first learning model reconstructed in step S105. Calculation of the prediction error in step S106 is the same as in step S102 described above.

In step S107, similarly to step S103, it is determined whether or not the prediction error calculated in step S106 is equal to or less than the threshold value ε. That is, it is determined whether or not the prediction error of the first learning model has been sufficiently improved by the reconstruction. As a result, when the prediction error of the first learning model is improved to be equal to or less than the critical value ε, it is deemed that the prediction accuracy of the first learning model has been improved, and the process ends. On the other hand, when the prediction error of the first learning model is still larger than the critical value ε (step S107: NO), the process returns to step S104. That is, when the improvement of the prediction error of the first learning model is not sufficient even by re-structuring, additional learning data is selected again in step S104, and the learning data is revised again, and the implementation of the first learning model is repeated. build. Such reconstruction of the first learning model is repeated until the prediction error formation critical value ε is equal to or less than step S107.

Here, the change in the predicted value of the first learning model with the number of implementations of the reconstruction will be specifically described. Fig. 10 shows the learning data used in the reconstruction of the first learning model (the SO2 concentration in the outflow gas of the objective variable and the learning data of the explanatory variable X used in the learning model ₎ for each number of implementations of the reconstruction. 11 is a graph showing the transition of the predicted value of the first learning model reconstructed using the learning materials shown in FIG. 10 .

FIG. 10 shows a case in which new additional learning materials are selected in step S104 as the number of times of reconstruction increases, whereby the number of materials included in the learning materials increases. As shown in FIG. 11 , the prediction error of the first learning model reconstructed using such learning data decreases as the number of times reconstruction is performed increases. This indicates that the prediction accuracy of the first learning model is improved by appropriately selecting additional learning data for each reconstruction.

In addition, when the number of implementations of reconstruction increases, the prediction error of the first learning model converges to a predetermined value (near 0.7 in the example of FIG. 11 ). Therefore, in step S107, in addition to or instead of making the predicted value of the first learning model equal to or less than the critical value, the iterative processing after step S104 may be terminated depending on whether the prediction error has sufficiently converged. determination.

The additional learning materials selected in this way are added to the learning materials for constructing the first learning model to create new learning materials, and the first learning model is reconstructed using the learning materials. In this case, by appropriately selecting the additional learning data to be added to the learning data according to the degree of deviation from the initial learning data included in the previous learning data, the prediction error of the first learning model can be effectively reduced. Thereby, even when the prediction error of the first learning model decreases due to some factors, good prediction accuracy can still be obtained by reconstructing the first learning model.

The control device 15 constructs the first learning model with improved prediction accuracy in the first learning model constructing unit 38 again, and can accurately set the control target value of the circulation flow rate based on the predicted value of the first learning model. As a result, the circulation pump adjustment unit 33 can adjust the number of circulation pumps 12 according to the control target value to appropriately control the circulation flow rate.

The reduction of the prediction error by such reconstruction of the learning model is performed similarly for the second learning model processed by the second learning model construction unit 39 . That is, when the prediction error of the second learning model calculated by the prediction error calculation unit 40 is equal to or less than the critical value, the additional learning data selection unit 42 performs additional learning for reconstructing the second learning model. The selection of the additional learning material of the material is performed to reconstruct the second learning model using new learning material including the additional learning material. In this case, by appropriately selecting the additional learning data to be added to the learning data according to the degree of deviation from the initial learning data contained in the previous learning data, the prediction error of the second learning model can be effectively reduced. Thereby, even if the prediction error of the second learning model decreases due to some factors, good prediction accuracy can still be obtained by reconstructing the second learning model.

The control device 15 reconstructs the second learning model with improved prediction accuracy in the second learning model construction unit 39, so that the control target for the absorbent slurry supply amount can be set with high accuracy based on the predicted value of the second learning model. value. As a result, the absorbent slurry supply control unit 37 can control the absorbent slurry supply amount control valve 23 based on the control target value to appropriately control the absorbent slurry supply amount.

In addition, in the above embodiment, CaCO ₃ is used as the SO ₂ absorbent, but it is not limited to CaCO ₃ . For example, magnesium hydroxide (Mg(OH) ₂ ) can also be used as an absorbent for SO ₂ .

In addition, a configuration in which the information processing device 52 shown in FIG. 12 is electrically communicably connected to the edge server 42 on a cloud environment or via a VPN may be used as a device for executing each process of the control device 15 . In this case, the information processing device 52 is equipped with: an operation data receiving unit 30, a first relationship table creation unit 31, a circulation flow rate determination unit 32, a second relationship table creation unit 35, an absorbent slurry supply amount determination unit 36, a first learning The model construction part 38, the second learning model construction part 39, the prediction error calculation part 40, and the additional learning data selection part 42 can also communicate with the circulation flow rate determination part 32 and the absorbent mud supply determination part 36. The control target value is given to the circulation pump adjustment unit 33 and the absorbent slurry supply control unit 37 of the control device 15 to control the circulation pump or the supply amount of the absorbent. In addition, the operation data receiving unit 30 may receive various operation data via the operation data relay unit 43 of the control device 15 , or may receive various operation data from the operation data acquisition unit 20 as described above.

In particular, when calculating on the cloud environment, from the viewpoint of security, the control target value of the circulation pump or the absorbent may not be directly controlled, but only displayed. For example, the operating index map generated on the cloud environment is sent to the customer's own device (terminal device 54) via a dedicated application, and the update of the local operating index map is performed by the customer's hand Condition. On the other hand, the information processing device 52 may also include a circulating pump regulator 33 and an absorbent slurry supply control unit 37 to remotely control the supply of the circulating pump or absorbent. Furthermore, the information processing device 52 may be configured to execute each process in the information processing device 52 in response to a request from the terminal device 54 .

In addition, the constituent elements of the above-described embodiments may be appropriately replaced with known constituent elements without departing from the gist of the present invention, and the above-described embodiments may be appropriately combined.

The contents described in each of the above-mentioned embodiments are grasped as follows, for example.

(1) An apparatus according to an embodiment is an apparatus for performing processing related to control of plant equipment based on prediction results using a learning model, and is characterized in that it includes: The additional learning data selection unit is used to select, from the aforementioned operation data, data with a high degree of deviation from the learning data used in the construction of the aforementioned learning model as additional learning when the prediction result using the aforementioned learning model meets a predetermined condition. information; and The learning model construction unit is used to reconstruct the aforementioned learning model using new learning materials including the aforementioned learning materials and the aforementioned additional learning materials.

According to the form of (1) above, new learning materials are created by selecting additional learning materials to be added to the learning materials for constructing the learning model, and the learning model is reconstructed using the learning materials. At this time, by selecting the additional learning data to be added to the new learning data so as to include the one with the largest degree of deviation from the learning data, it is possible to appropriately reconstruct the learning model. Then, by executing the processing related to the control of the plant equipment based on the prediction result of the learning model reconstructed appropriately in this way, good control accuracy can be obtained.

(2) In other forms, in the form of (1) above, when the prediction result using the reconstructed new learning model satisfies a predetermined condition, the above-mentioned additional learning data selection department is selected as The aforementioned additional learning materials that have not been selected as the aforementioned additional learning materials are further selected as the aforementioned additional learning materials that include the aforementioned materials with a large degree of deviation, The learning material constructing unit implements the reconstruction of the learning model using the new learning material including the additional learning material further selected by the additional learning material selecting unit.

According to the form of (2) above, when the prediction result using the reconstructed new learning model satisfies a predetermined condition, new additional learning materials are added to the learning materials to create new learning materials, and the new learning materials are used. New learning materials are used to implement the reconstruction of the learning model again. By repeatedly implementing such selection of additional learning data and reconstruction of the learning model, the prediction error of the learning model can be sufficiently reduced.

(3) In other forms, in the form of (1) or (2) above, the additional learning data selection unit selects the average value of the parameters included in the operation data for a predetermined period as the additional learning material.

According to the aspect of (3) above, by using the average value over a predetermined period as an additional learning parameter, it is possible to effectively reduce the amount of computation for rebuilding the learning model while ensuring learning accuracy.

(4) In another aspect, in any one of the above (1) to (3), the learning model construction unit executes the learning model when the prediction result continues to meet the predetermined condition for a predetermined time or longer. reconstruction.

According to the aspect of (4) above, whether the prediction result satisfies the predetermined condition is determined based on whether the prediction result satisfies the predetermined condition continuously over a predetermined period of time. Prediction results may vary depending on the operating status of plant equipment. If short-term judgments are assumed, learning model reconstruction will be carried out frequently, and the burden of model management may increase. Therefore, by performing the determination of the continuity over a predetermined period of time as in this embodiment, it is possible to appropriately reconstruct the learning model and achieve efficient model management.

(5) In another aspect, in any one of the above (1) to (4), the learning materials are materials before construction of the learning model or materials used in the previous construction.

According to the aspect of (5) above, the learning model is implemented using new learning materials created by adding additional learning materials to the materials before the learning model was constructed or the learning materials used in the previous construction of the learning model. Rebuild.

(6) In another aspect, in any one of the above (1) to (5), the additional learning data selection unit selects the above-mentioned Additional learning materials.

According to the aspect of (6) above, the selection of the additional learning data is carried out for the operation data acquired during the normal operation of the factory equipment. For example, the operation data acquired during abnormal operation such as when an abnormality occurs in the factory equipment, when the operation is started, and when the operation is stopped, is excluded from the selection of additional learning data, so that the prediction result of the learning model can be obtained appropriately. In addition, when the operation data includes data obtained during such unsteady operation, it may be excluded by performing pre-processing on the operation data.

(7) In any of the above-mentioned (1) to (6) forms, when the prediction result using the aforementioned learning model satisfies a predetermined condition, it means that it is based on the prediction obtained by using the aforementioned learning model. When the prediction error of the value meets the critical value.

According to the aspect of (7) above, whether the prediction result satisfies the predetermined condition is determined based on whether the prediction error based on the prediction value obtained by using the learning model satisfies the critical value. Thereby, even if the prediction error of the learning model decreases due to certain factors, the learning model can be reconstructed using new learning data including effectively selected additional learning data, and good prediction accuracy can be obtained. Then, by performing processing related to the control of plant equipment based on the predicted value of the learning model whose prediction accuracy has been improved in this way, good control accuracy can be obtained.

(8) In another aspect, in the aspect of (7) above, the additional learning data selection unit selects the parameters included in the additional learning data from the operation data based on the degree of contribution to the predicted value. .

According to the aspect of (8) above, by including some parameters selected from the operation data in the additional learning data, it is possible to effectively reduce the amount of computation for rebuilding the learning model while ensuring the learning accuracy.

(9) Regarding other forms, in the form of (7) or (8) above, the plant equipment is degassed by contacting the exhaust gas generated in the combustion device with the absorption liquid circulated in the absorption tower. Sulfur wet flue gas desulfurization unit, The predicted value is the sulfur dioxide concentration of the exhaust gas at the outlet of the absorption tower.

According to the form of (9) above, by performing reconstruction on the learning model for predicting the concentration of sulfur dioxide in the exhaust gas at the outlet of the absorption tower of the wet-type flue gas desulfurization device, when the prediction error is larger than the predetermined value, it can be achieved. The prediction accuracy according to the learned model is properly ensured.

(10) In another aspect, in the aspect of (9) above, the control target value of the circulation flow rate of the absorbent is determined based on the predicted value calculated by the learning model.

According to the aspect of (10) above, by reconstructing, a prediction value is calculated using a learning model with a reduced prediction error, and the control target value of the circulation amount of the absorbent is determined based on the prediction value, whereby a good performance can be obtained. control precision.

(11) In another form, in the form of (9) or (10) above, the control target value of the absorbent supply amount to the absorption tower is determined based on the prediction value calculated by the learning model. .

According to the form of (11) above, by reconstructing, a predicted value is calculated using a learning model with a reduced prediction error, and the control target value of the absorbent supply amount is determined based on the predicted value, whereby good control can be achieved. precision.

(12) A form of remote monitoring system comprising a terminal device capable of communicating with a device for performing processing related to plant equipment control based on prediction results using a learning model, characterized by the aforementioned The device has: The additional learning data selection unit is used to select from the operation data the data from the learning data used in the construction of the learning model when the prediction result using the learning model satisfies a predetermined condition based on the request from the terminal device. Data with a large degree of deviation are used as additional learning materials; and The learning model constructing unit is used to reconstruct the aforementioned learning model by using the learning materials including the aforementioned learning materials and the aforementioned additional learning materials.

According to the aspect of (12) above, new learning materials are created by selecting additional learning materials to be added to the learning materials for constructing the learning model, and the learning model is reconstructed using the learning materials. At this time, by selecting the additional learning data to be added to the new learning data so as to include the one with the largest degree of deviation from the learning data, it is possible to appropriately reconstruct the learning model. Then, by executing the processing related to the control of the plant equipment based on the prediction result of the learning model reconstructed appropriately in this way, good control accuracy can be obtained.

(13) An apparatus control method of an aspect, which is a method for controlling an apparatus for performing processing related to plant equipment control based on a prediction result using a learning model, and is characterized by comprising: The step of selecting additional learning materials, which is to select, from the aforementioned operation data, materials with a high degree of deviation from the learning data used in the construction of the aforementioned learning model as additional learning materials when the prediction result using the aforementioned learning model meets a predetermined condition; and The step of constructing the learning model is to construct the learning model again by using the learning material including the learning material and the additional learning material.

According to the aspect of (13) above, new learning materials are created by selecting additional learning materials to be added to the learning materials for constructing the learning model, and the learning model is reconstructed using the learning materials. At this time, by selecting the additional learning data to be added to the new learning data so as to include the one with the largest degree of deviation from the learning data, it is possible to appropriately reconstruct the learning model. Then, by executing the processing related to the control of the plant equipment based on the prediction result of the learning model reconstructed appropriately in this way, good control accuracy can be obtained.

(14) A control method of a remote monitoring system of an aspect, which is the control of a remote monitoring system composed of a terminal device capable of communicating with a device for performing processing related to plant equipment control based on prediction results using a learning model A method characterized by: The additional learning data selection step is to select the degree of deviation from the learning data used in the construction of the learning model from the operation data when the prediction result using the learning model satisfies a predetermined condition based on the request from the terminal device. large data as additional learning materials; and The step of constructing the learning model is to construct the learning model again by using the learning material including the learning material and the additional learning material.

According to the aspect of (14) above, new learning materials are created by selecting additional learning materials to be added to the learning materials for constructing the learning model, and the learning model is reconstructed using the learning materials. At this time, by selecting the additional learning data to be added to the new learning data so as to include the one with the largest degree of deviation from the learning data, it is possible to appropriately reconstruct the learning model. Then, by executing the processing related to the control of the plant equipment based on the prediction result of the learning model reconstructed appropriately in this way, good control accuracy can be obtained.

1: Combustion device 2: Piping 3: Piping for circulation 5: Generator 10: Wet exhaust gas desulfurization device 11: Absorption tower 12:Circulation pump 13: Absorbent mud supply department 14: Gypsum recycling department 15: Control device 16:Outflow piping 17: Gas analyzer 20: Operation data acquisition department 21: Absorbent mud manufacturing equipment 22: Piping for absorbent slurry supply 23: Absorbent mud supply control valve 25: Gypsum separator 26: Piping for pumping out gypsum slurry 27: Pump for pumping out gypsum slurry 30:Operation Data Receiving Department 31: The first relationship table preparation department 32: Circulation flow determination unit 33: Circulation pump adjustment department 35: The second relationship table preparation department 36: Absorbent mud supply determination department 37: Absorbent mud supply control department 38: The first learning model construction department 39: The second learning model construction department 40: Prediction Error Calculation Department 42:Additional learning materials selection department 43: Operation Data Relay Department 44: Remote monitoring system 48:Edge server 50: Remote monitoring device 52: Information processing device 54: Terminal device 55: Information processing system

[ Fig. 1] Fig. 1 is a configuration diagram of an exhaust gas desulfurization device according to an embodiment. [ Fig. 2 ] is a schematic configuration diagram of a remote monitoring system according to an embodiment. [ Fig. 3 ] is a flowchart showing basic control of a wet-type exhaust gas desulfurization device according to an embodiment. [FIG. ₄ ] It is a graph which shows each transition of the predicted value of the SO2 concentration in the effluent gas, the measured value of the _SO2 concentration by the gas analyzer, and the true value of the predicted value of the _SO2 concentration. [ Fig. 5 ] is a diagram schematically showing an example of a first relational table created in basic control of a wet-type exhaust gas desulfurization device according to an embodiment. [ Fig. 6 ] is a diagram schematically showing an example of a second relational table created in the basic control of the wet-type exhaust gas desulfurization device according to the embodiment. [ Fig. 7 ] is a flowchart showing a factory equipment control method according to an embodiment. [FIG. 8] It is a flowchart which shows the selection method of the additional learning material in step S104 of FIG. [FIG. 9A] It is a figure which shows the process of selecting an additional learning material in step S204 of FIG. [ Fig. 9B ] is a diagram showing the procedure of selecting additional learning materials in step S204 of Fig. 8 . [FIG. 10] It is a figure which shows the distribution of the learning material used for the reconstruction of the 1st learning model by the implementation frequency of reconstruction. [FIG. 11] It is a figure which shows the transition of the predicted value of the 1st learning model reconstructed using each learning data shown in FIG. 10. [ Fig. 12 ] is a configuration diagram of an information processing system according to an embodiment. [FIG. 13] It is a figure which shows the internal structure of the information processing apparatus of FIG. 12 together with a control apparatus.

1: Combustion device

2: Piping

3: Piping for circulation

5: Generator

10: Wet exhaust gas desulfurization device

11: Absorption tower

12, 12a, 12b, 12c: circulation pump

13: Absorbent mud supply department

14: Gypsum recycling department

15: Control device

16:Outflow piping

17: Gas analyzer

20: Operation data acquisition department

21: Absorbent mud manufacturing equipment

22: Piping for absorbent slurry supply

23: Absorbent mud supply control valve

25: Gypsum separator

26: Piping for pumping out gypsum slurry

27: Pump for pumping out gypsum slurry

30:Operation Data Receiving Department

31: The first relationship table preparation department

32: Circulation flow determination unit

33: Circulation pump adjustment department

35: The second relationship table preparation department

36: Absorbent mud supply determination department

37: Absorbent mud supply control department

38: The first learning model construction department

39: The second learning model construction department

40: Prediction Error Calculation Department

42:Additional learning materials selection department

Claims (14)

An apparatus for performing processing related to control of plant equipment based on prediction results using a learning model, characterized by: The additional learning data selection unit is used to select, from the aforementioned operation data, data with a high degree of deviation from the learning data used in the construction of the aforementioned learning model as additional learning when the prediction result using the aforementioned learning model meets a predetermined condition. information; and The learning model construction unit is used to reconstruct the aforementioned learning model using new learning materials including the aforementioned learning materials and the aforementioned additional learning materials. The device described in claim 1, wherein when the prediction result using the reconstructed new learning model meets a predetermined condition, the additional learning data selection unit selects from the aforementioned operation that is not selected as the additional learning data. The data are further selected as the aforementioned additional learning materials comprising the aforementioned data with a large degree of deviation, The learning material constructing unit implements the reconstruction of the learning model using the new learning material including the additional learning material further selected by the additional learning material selecting unit. The device according to claim 1 or 2, wherein the additional learning data selection unit selects an average value of parameters included in the operation data over a predetermined period as the additional learning data. The device according to claim 1 or 2, wherein the learning model constructing unit reconstructs the learning model when the prediction result continues to meet the predetermined condition for a predetermined time or longer. The device according to claim 1 or 2, wherein the learning data are data before construction of the learning model or data used in the previous construction. The device according to claim 1 or 2, wherein the additional learning material selection unit selects the additional learning material from the operation data obtained during the normal operation of the plant. The device according to claim 1 or 2, wherein when the prediction result using the learning model satisfies a predetermined condition, it means that the prediction error based on the prediction value obtained by using the learning model meets a critical value. The device according to claim 7, wherein the additional learning data selection unit selects parameters included in the additional learning data from the operation data based on the degree of contribution to the predicted value. The device as described in Claim 7, wherein the above-mentioned plant equipment is a wet exhaust gas desulfurization device that desulfurizes by contacting the exhaust gas generated in the combustion device and the absorption liquid circulated in the absorption tower, The predicted value is the sulfur dioxide concentration of the exhaust gas at the outlet of the absorption tower. The device according to claim 9, wherein the control target value of the circulation flow rate of the absorbent is determined based on the predicted value calculated by the learning model. The apparatus according to claim 9, wherein the control target value of the absorbent supply amount to the absorption tower is determined based on the predicted value calculated by the learning model. A remote monitoring system comprising a terminal device capable of communicating with a device for performing a process related to plant equipment control based on a prediction result using a learning model, characterized in that the aforementioned device has: The additional learning data selection unit is used to select from the operation data the data from the learning data used in the construction of the learning model when the prediction result using the learning model satisfies a predetermined condition based on the request from the terminal device. Data with a large degree of deviation are used as additional learning materials; and The learning model constructing unit is used to reconstruct the aforementioned learning model by using the learning materials including the aforementioned learning materials and the aforementioned additional learning materials. A device control method for performing processing related to plant equipment control based on a prediction result using a learning model, characterized by having: The step of selecting additional learning materials, which is to select, from the aforementioned operation data, materials with a high degree of deviation from the learning data used in the construction of the aforementioned learning model as additional learning materials when the prediction result using the aforementioned learning model meets a predetermined condition; and The step of constructing the learning model is to construct the learning model again by using the learning material including the learning material and the additional learning material. A control method of a remote monitoring system, which is a control method of a remote monitoring system composed of a terminal device capable of communicating with a device for performing processing related to plant equipment control based on a prediction result using a learning model, characterized by have: The additional learning data selection step is to select the degree of deviation from the learning data used in the construction of the learning model from the operation data when the prediction result using the learning model satisfies a predetermined condition based on the request from the terminal device. large data as additional learning materials; and The step of constructing the learning model is to construct the learning model again by using the learning material including the learning material and the additional learning material.

Images