JP2023173706A - State quantity prediction device, state quantity prediction method, state quantity prediction system, and control method of the state quantity prediction system - Google Patents

Abstract

To predict a state quantity accurately while keeping a cost down using a prediction model obtained by adjusting a learned prediction model corresponding to another plant.SOLUTION: A state quantity prediction device predicts a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model. The state quantity prediction device stores a first prediction model corresponding to a first plant, creates a second prediction model corresponding to a second plant by adjusting the first prediction model using measurement data of the second plant, and predicts a state quantity of the second plant using the second prediction model. The creation of the second prediction model is performed by regularizing and learning a difference between machine learning parameters of the first prediction model and the second prediction model.SELECTED DRAWING: Figure 1

Description

The present disclosure relates to a state quantity prediction device, a state quantity prediction method, a state quantity prediction system, and a control method of a state quantity prediction system.

BACKGROUND ART In equipment such as plants, state quantities may be predicted by calculations using numerical models for purposes such as monitoring, control, or abnormality determination. This type of numerical model includes, for example, a physical model based on a static equilibrium equation or state equation derived from physical knowledge, or a statistical model such as a machine learning model using neural networks or multiple regression analysis. There is.

When using a physical model as a numerical model, the physical model assumes that the true prediction formula is linear and performs approximations such as linearization, or models that assume operating conditions such as static stability. However, with such a physical model, there is a risk that the prediction error will become large if the assumed preconditions are not satisfied. Furthermore, in statistical models such as machine learning models, prediction errors may become large in the extrapolation region of learning data used to construct the model. Therefore, Patent Document 1 proposes a prediction method in which the error in the state quantity predicted by the physical model is corrected by the machine learning model by learning the deviation between the state quantity derived by the physical model and the actually measured quantity by machine learning. There is. This method is a prediction method that combines a physical model and a machine learning model, but the machine learning model does not take into account the dynamic characteristics of the state quantity to be predicted. Therefore, there is a possibility that sufficient prediction accuracy may not be obtained for objects whose state quantities have dynamic characteristics.

One possible solution to such issues regarding dynamic characteristics is to introduce a Recurrent Neural Network (RNN) as a machine learning method. On the other hand, neural ODE (Neural Ordinary Differential Equation), which is a continuous expression neural network that introduces a differential structure, has the advantage of having higher memory efficiency than RNN and being able to handle temporally continuous models. Non-Patent Document 1 discloses a prediction method that can improve prediction accuracy by combining neural ODE and known differential equations.

Patent No. 2882232

Manuel A.Roehrl, Modeling SystemDynamics with Physics Informed Neural Networks Based on Lagrangian Mechanics,IFAC, 2020

When predicting state quantities using a prediction model that combines a machine learning model and a physical model, as in Patent Document 1 and Non-Patent Document 1 above, the prediction model is based on learning data acquired in the plant to be predicted. By learning using , the machine learning parameters included in the machine learning model and the physical parameters included in the physical model are determined. As learning data used for learning such a prediction model, measurement data obtained by a measuring device such as a sensor installed in a plant to be predicted is used.

Here, when trying to learn a prediction model for predicting state quantities for a plant that does not have sensors installed to obtain measurement data that can be used as learning data, it is necessary to use sensors to obtain new learning data. etc. is disadvantageous in terms of cost. In this case, if a trained prediction model has been obtained using learning data in the past at another similar plant, the prediction model corresponding to the other plant is used to predict the prediction model corresponding to the plant where sensors, etc. are not installed. It is possible to use it as a model. However, since each plant has unique characteristics, the prediction accuracy will decrease if the prediction model of another plant is used as is, and the prediction model will need to be fine-tuned.

At least one embodiment of the present disclosure has been made in view of the above-mentioned circumstances, and uses a prediction model obtained by adjusting a learned prediction model corresponding to another plant to reduce cost and improve accuracy. It is an object of the present invention to provide a state quantity prediction device, a state quantity prediction method, a state quantity prediction system, and a control method for a state quantity prediction system that can predict state quantities well.

In order to solve the above problems, state quantity prediction devices according to some embodiments of the present disclosure,
A state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model,
A first prediction for storing a first prediction model that corresponds to a first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. a model storage section,
a measurement data acquisition unit for acquiring measurement data of the second plant;
a second predictive model creation unit for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
a state quantity prediction unit for predicting the state quantity of the second plant using the second prediction model;
Equipped with
The second predictive model creation unit creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

In order to solve the above problems, a state quantity prediction system according to some embodiments of the present disclosure,
A state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, and an information processing device capable of communicating with the state quantity prediction device, the system comprising:
The state quantity prediction device includes:
A first prediction for storing a first prediction model that corresponds to a first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. a model storage section,
a measurement data acquisition unit for acquiring measurement data of the second plant after being requested by the information processing device;
a second predictive model creation unit for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
a state quantity prediction unit for predicting the state quantity of the second plant using the second prediction model;
Equipped with
The second predictive model creation unit creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

In order to solve the above problems, state quantity prediction methods according to some embodiments of the present disclosure include:
A state quantity prediction method for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, the method comprising:
storing a first predictive model that corresponds to a first plant and is the predictive model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned;
a step of acquiring measurement data of the second plant;
creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
predicting the state quantity of the second plant using the second prediction model;
Equipped with
In the step of creating the second predictive model, the second predictive model is created by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

In order to solve the above problems, a control method for a state quantity prediction system according to some embodiments of the present disclosure includes:
A control method for a state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter and an information processing device capable of communicating with each other using a prediction model including a machine learning model and a physical model. hand,
The state quantity prediction device includes:
a measurement data acquisition step for acquiring measurement data of the second plant after being requested by the information processing device;
Using the measurement data, create a first prediction model that corresponds to the first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. A second predictive model for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model stored in a first predictive model storage unit for storage. creation step,
a state quantity prediction step for predicting the state quantity of the second plant using the second prediction model;
Run
The second predictive model creation step creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

In order to solve the above problems, state quantity prediction devices according to some embodiments of the present disclosure,
A state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model,
Predictive model creation for creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant Department and
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit creates the predictive model by regularizing and learning the machine learning parameters of the predictive model.

In order to solve the above problems, a state quantity prediction system according to some embodiments of the present disclosure,
A state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, and an information processing device capable of communicating with the state quantity prediction device, the system comprising:
The state quantity prediction device includes:
Predictive model creation for creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant Department and
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit creates the predictive model by regularizing and learning the machine learning parameters of the predictive model.

In order to solve the above problems, state quantity prediction methods according to some embodiments of the present disclosure include:
A state quantity prediction method for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, the method comprising:
creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant;
predicting the state quantity of the plant using the prediction model;
Equipped with
In the step of creating the predictive model, the predictive model is created by regularizing and learning the machine learning parameters of the predictive model.

In order to solve the above problems, a control method for a state quantity prediction system according to some embodiments of the present disclosure includes:
A control method for a state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter and an information processing device capable of communicating with each other using a prediction model including a machine learning model and a physical model. hand,
The state quantity prediction device includes:
After the request from the information processing device, the machine learning parameters included in the machine learning model and the physical parameters included in the physical model are learned by using learning data acquired in the plant. a predictive model creation unit for creating a predictive model;
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit creates the predictive model by regularizing and learning the machine learning parameters of the predictive model.

According to at least one embodiment of the present disclosure, a state in which a state quantity can be predicted with high accuracy while suppressing costs using a prediction model obtained by adjusting a learned prediction model corresponding to another plant A quantity prediction device, a state quantity prediction method, a state quantity prediction system, and a control method for a state quantity prediction system can be provided.

FIG. 1 is a block diagram showing the configuration of a state quantity prediction device according to an embodiment. 2 is a schematic diagram showing an example of a neural network that is the machine learning model of FIG. 1. FIG. FIG. 3 is a block diagram showing the configuration of a state quantity prediction device according to another embodiment. It is a figure which shows the transition of the predicted value calculated based on the 2nd prediction model adjusted by the actual value of a state quantity. FIG. 1 is a diagram schematically showing the configuration of a desulfurization device. FIG. 1 is a block diagram showing the configuration of a state quantity prediction system according to an embodiment.

Hereinafter, some embodiments of the present invention will be described with reference to the accompanying drawings. However, the configurations described as embodiments or shown in the drawings are not intended to limit the scope of the present invention thereto, and are merely illustrative examples.

A state quantity prediction device 1 according to at least one embodiment of the present disclosure is a device for predicting a state quantity x of a plant corresponding to an input parameter u using a prediction model M. The hardware configuration for realizing the state quantity prediction device 1 is not limited, but may include, for example, a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), a computer-readable storage medium, etc. The information processing device is configured as an information processing device. A series of processes for realizing various functions is stored in a storage medium, etc. in the form of a program, for example, and the CPU reads this program into a RAM, etc., and executes information processing and arithmetic processing. By doing so, various functions are realized. Note that the program may be pre-installed in a ROM or other storage medium, provided as being stored in a computer-readable storage medium, or distributed via wired or wireless communication means. etc. may also be applied. Computer-readable storage media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, and the like.

The prediction model M used by the state quantity prediction device 1 includes a machine learning model M _N and a physical model M _P. In the following embodiments, a physical model M _P for predicting the state quantity x of a plant in a static state is combined with a machine learning model M _N for predicting the dynamic state of the plant. A prediction model M is constructed. The functions of the machine learning model M _N and the physical model M _P included in such a prediction model M are merely examples, and are not limited.

FIG. 1 is a block diagram showing the configuration of a state quantity prediction device 1A according to an embodiment. The state quantity prediction device 1A has a prediction model M in which a machine learning model M _N and a physical model M _P are combined, and specifically, for calculating the first predicted value X1 using the machine learning model M _N. a first predicted value calculation unit 4 for calculating a second predicted value X2 using the physical model _MP , and a second predicted value calculation unit 6 for calculating the second predicted value The state quantity predicted value calculation unit 8 is provided for calculating the predicted value X of the state quantity x.

The physical model M _P is a physical model corresponding to a plant in a static state, and outputs a second predicted value X2 which is a static component of the state quantity x of the plant corresponding to an input parameter u in a static state. It is configured as follows. Such a physical model M _P is generally expressed by the following equation as a relational expression between the static input parameter u and the second predicted value X2.
X2=f(u, θ _P ) (1)
Note that θ _P is at least one physical parameter included in the physical model _MP , and in this embodiment, k physical parameters θ _P1 , θ _P2 , ..., θ _Pk (hereinafter referred to as "physical parameter θ _P" as appropriate) ” (k is a natural number of 1 or more). f is an arbitrary function whose variables are the input parameter u and the physical parameter _θP .

The machine learning model _MN is a machine learning model corresponding to a plant in a dynamic state, and the first predicted value X1 is a dynamic component of the state quantity x of the plant corresponding to the input parameter u during the dynamic state. configured to output. Such a machine learning model M _N is configured, for example, as a neural network that shows the relationship between a dynamic input parameter u and a first predicted value X1 that is a dynamic component of the state quantity x of the plant.

FIG. 2 is a schematic diagram showing an example of a neural network that is the machine learning model M _N of FIG. 1. This neural network includes an input layer 12 having a plurality of nodes into which a plurality of input parameters u1, u2, . It includes an intermediate layer 13 (hidden layer) that includes a plurality of nodes between an input layer 12 and an output layer 14. _Each node of the intermediate layer 13 has machine learning parameters θ _N1 , θ _{N2 ,} . Set.

The first predicted value calculation unit 4 calculates the first predicted value X1 by giving the input parameter u to the machine learning model _MN . The second predicted value calculation unit 6 calculates the second predicted value X2 by giving the input parameter u to the physical model _MP . The state quantity predicted value calculation unit 8 calculates the first predicted value X1 calculated by the first predicted value calculation unit 4 corresponding to the dynamic component of the state quantity x, and the second predicted value X1 corresponding to the static component of the state quantity x. Based on the second predicted value X2 calculated by the predicted value calculation unit 6, a predicted value X of the state quantity x is calculated. In this way, the state quantity prediction device 1A can obtain the predicted value X of the state quantity x including a dynamic component and a static component.

Here, the machine learning parameter θ _N and the physical parameter θ _P included in the prediction model M are determined by learning using learning data (hereinafter, the machine learning parameter θ _N and the physical parameter θ _P are collectively referred to as (in this case, it is appropriately referred to as "learning parameter θ"). Specifically, during learning, initial values θ _N0 and θ _P0 are given to the machine learning parameter θ _N and physical parameter θ _P , respectively, and, for example, the evaluation function F1 defined by the following equation is minimized. The machine learning parameter θ _N and the physical parameter θ _P are searched so that
F1=Σ(X-X') ² (1)
Note that X' is a true value (for example, a measured value) corresponding to the predicted value X' included in the learning data.

Here, the above-mentioned state quantity prediction device 1A uses the prediction model M to predict the state quantity x, but in order to obtain the prediction model M with good prediction accuracy, the learning data used for learning the prediction model M must be The premise is that you have acquired a sufficient amount. Therefore, for example, in a plant for which sufficient training data cannot be acquired, the prediction accuracy of the prediction model M obtained through learning may decrease. In order to solve this problem, it may be possible to add sensors or the like to the plant to acquire sufficient learning data, but this is disadvantageous in terms of cost. In order to solve such problems, in the next embodiment, the learned first prediction model M1 corresponding to the first plant P1 for which sufficient learning data can be acquired is adjusted. By obtaining the second prediction model M2 corresponding to the second plant P2, which is different from the above, it becomes possible to predict the state quantity x in the second plant P2.

FIG. 3 is a block diagram showing the configuration of a state quantity prediction device 1B according to another embodiment. The state quantity prediction device 1B is a device for predicting the state quantity x of the second plant P2, and includes a first prediction model storage unit 15, a measurement data acquisition unit 16, a second prediction model creation unit 17, and a state quantity prediction unit 18.

The first prediction model storage unit 15 is configured to store a first prediction model M1 corresponding to a first plant P1 different from the second plant P2 that is a prediction target of the state quantity prediction device 1B. The first predictive model M1 is the aforementioned predictive model M corresponding to the first plant P1, and is a predictive model that includes a machine learning model M _N and a physical model M _P , as illustrated in FIG. In the first plant P1, it is possible to acquire sufficient operation data as learning data for learning the first prediction model M1, and the first prediction model M1 uses such sufficient operation data as learning data. Through learning, it is constructed to have sufficient prediction accuracy. The first predictive model M1 constructed in this way is readably stored in the first predictive model storage unit 15.
Note that the first plant P1 and the second plant P2 are similar to each other. Whether or not the first plant P1 and the second plant P2 are similar can be determined by, for example, the degree of similarity between the equipment configuration of the plants, the specifications of each equipment, state quantities obtained as other than learning data (the configuration of sensors installed in the plant), etc. ) can be judged by comprehensively considering the degree of similarity between the two.

The measurement data acquisition unit 16 is configured to acquire operational data from the second plant P2 as measurement data Dm in the second plant P2, which is a prediction target of the state quantity prediction device 1B. The second plant P2 is equipped with measuring equipment for measuring operational data, but due to differences in the type of measuring equipment installed in each plant, the control method, etc., the state quantity x There is a restriction that it is not possible to acquire a sufficient number of data regarding the actual measured value ).

The second prediction model creation unit 17 uses the measurement data Dm of the second plant P2 acquired by the measurement data acquisition unit 16 to adjust the first prediction model M1 read from the first prediction model storage unit 15. , is a configuration for creating a second prediction model M2 corresponding to the second plant P2. Creation of the second prediction model M2 involves learning parameters θ (the above-mentioned machine learning parameters θ _N and physical This is done by finely adjusting the parameter θ _P ). In each plant, the relationship between the input parameter u and the predicted value In plant P2, the prediction model M is relatively close. Therefore, the second prediction model creation unit 17 finely adjusts the first prediction model M1 corresponding to the learned first plant P1 using sufficient learning data, so that the actual measured value X' of the state quantity x is sufficiently adjusted. A second prediction model M2 corresponding to the second plant P2 for which acquisition of learning data is difficult is created.

The creation of the second prediction model M2 by the second prediction model creation unit 17 is performed by regularizing the magnitude of the difference between the machine learning parameter θ _N1 of the first prediction model M1 and the machine learning parameter θ _N2 of the second prediction model M2. This is done by performing learning (transfer learning). Specifically, this learning is performed so as to minimize the evaluation function F2 of the following equation.
F2=Σ(X-X') ² +βΣ(θ _N2 -θ _N1 ) ² (2)

This evaluation function F2 includes a first term similar to the above-described evaluation function F1 (see equation (1) above) and a second term that is a regular term. The second term is expressed as the 2-norm of the difference between the machine learning parameters θ _N of the first prediction model M1 and the second prediction model M2, and is added to the first term using the weighting coefficient β. In such learning to minimize the evaluation function F2, by restricting the degree of freedom for learning the machine learning parameter θ _N of the prediction model M including the machine learning model M _N and the physical model M _P , the actual measurement of the state quantity x is performed. Even when there is little learning data for the value X', updating of the physical parameter θ _P is promoted and overfitting can be avoided. As a result, even when only a small amount of learning data (actual measured value X' of the state quantity x included in the measurement data Dm) is obtained in the second plant P2, the second predictive model creation unit 17 uses the small amount of learning data By fine-tuning the first predictive model M1 using , a reliable second predictive model M2 can be created.

Note that the second predictive model creation unit 17 may set an allowable range for the physical parameter θ _P in learning using the evaluation function F2. This allowable range can be defined in advance as a range that the physical parameter θ _P can take. As a result, in addition to limiting the degree of freedom for learning the machine learning parameter θ _N through the regularization described above, by also limiting the degree of freedom for learning the physical parameter θ _P , a more reliable second learning model M2 can be created. Can be created.

The state quantity prediction unit 18 is configured to calculate a predicted value X of the state quantity x for the second plant P2 using the second prediction model M2 created by the second prediction model creation unit 17. Specifically, as described above with reference to FIG. 1, the state quantity prediction unit 18 inputs the input parameter u, which is the operating data acquired in the second plant P2, to the second prediction model M2. As a result, the predicted value X of the corresponding state quantity x is obtained.

Note that each block configuring the state quantity prediction device 1B shown in FIG. 3 may be configured as a single device or may be configured over a plurality of devices. For example, the state quantity x is predicted using the configuration related to the creation of the second prediction model M2 (first prediction model storage unit 15, measurement data acquisition unit 16, second prediction model creation unit 17, etc.) and the second prediction model M2. The configuration (such as the state quantity prediction unit 18) for performing the calculation may be configured as a different device. In this case, the second prediction model M2 may be created in the former case, and the state quantity x may be predicted using the second prediction model M2 created in the latter case.

Further, the second prediction model creation unit 17 determines that the predicted value X of the state quantity x predicted by the state quantity prediction unit 18 is the actual value X′ of the state quantity The second prediction model M2 may be adjusted so that . Here, FIG. 4 is a diagram showing the transition of the predicted value X calculated based on the second prediction model M2 adjusted by the actually measured value X' of the state quantity x.

In FIG. 4, the predicted value X of the state quantity x predicted by the state quantity prediction unit 18 at times t1, t2, . The second prediction model M2 is adjusted so that X′. The second prediction model M2 is basically created using the above-mentioned evaluation function F2. If the type of fuel (coal type, etc.) used in the boiler attached to the boiler is changed, the prediction accuracy may decrease. In this aspect, the operating conditions of the second plant P2 are changed by adjusting the second prediction model M2 so that the predicted value X of the state quantity x becomes the actual measured value X' every predetermined sampling period Ts. Even in this case, the reliability of the second prediction model M2 can be suitably maintained.

Furthermore, when learning the prediction model M using the aforementioned evaluation function F1 (see equation (1) above), if the number of data included in the training data is small, overfitting may occur and prediction accuracy may decrease. There is. Therefore, overfitting may be avoided by performing regularization during learning of the prediction model M. Specifically, as the evaluation function used when learning the prediction model M, the evaluation function F3 of the following formula may be used instead of the evaluation function F1.
F3=Σ(X-X') ² +αΣθ _i ² (3)

In this evaluation function F3, the first term indicates the sum of prediction squared errors as in the evaluation function F1, and the second term indicates a regular term. The above equation (3) is just an example; for example, the first term is the average value obtained by dividing the first term of equation (3) by the number of evaluation data, and the second term is the second term of equation (3). It may also be an average value obtained by dividing by the number of parameters. By learning using such an evaluation function F3, overfitting can be avoided, and even when the number of data included in the learning data is small, overfitting can be avoided and a highly accurate prediction model can be obtained.

In this way, in learning using the evaluation function F3, the degree of freedom of the machine learning parameter θ _N among the learning parameters θ included in the prediction model M becomes relatively high with respect to the physical parameter θ _P. Therefore, in learning using the evaluation function F3, learning of the machine learning parameter θ _N becomes dominant, and there is a risk that problems such as learning of the physical parameter θ _P may not be performed appropriately or learning may not be reproducible. There is.

In order to solve this problem, learning may be performed using the evaluation function F4 so as to regularize only the machine learning parameter θ _N of the learning parameters θ, as shown in the following equation.
F4=Σ(X-X') ² +αΣ(θ _Ni ) ² (4)
By performing learning so that such evaluation function F4 is minimized, overfitting can be avoided by restricting the degree of freedom for learning the machine learning parameter θ _N and promoting the update of the physical parameter θ _P. , even when the number of learning data (actually measured value X' of the state quantity x included in the measurement data Dm) is small, the machine learning parameter θ _N and the physical parameter θ _P can each be suitably determined by learning.

In addition, in equation (4), the second term of the evaluation function F4 is expressed by adding the two norms of the machine learning parameter θ _N with the weighting coefficient α, but this is just an example as with the evaluation function F3 described above. For example, the first term is the average value obtained by dividing the first term in equation (4) by the number of evaluation data, and the second term is the average value obtained by dividing the second term in equation (4) by the number of parameters. You can also use it as

Note that also in learning using the above-mentioned evaluation functions F3 and F4, an allowable range may be set for the physical parameter _θP of the prediction model M. By restricting the degree of freedom in learning physical parameters, a learning model M that can predict state quantities with higher accuracy can be obtained.

Furthermore, the prediction model M created by learning using the evaluation functions F3 and F4 described above also has a predicted value X of the state quantity x predicted using the prediction model M, as described above with reference to FIG. For example, the state quantity x may be adjusted to the actual value X' measured by equipment installed in the plant. As a result, the predicted value X of the state quantity x predicted using the prediction model M is adjusted at a predetermined timing so that it becomes the actual measured value In the embodiment described below, even when the type of fuel (coal type, etc.) used in the boiler attached to the desulfurization device 20 changes, the prediction accuracy of the prediction model M can be suitably maintained.

Next, a specific application example of the state quantity prediction devices 1A and 1B having the above configuration will be described. Here, as a specific example of predicting the state quantity x of a plant, a case will be described in which the absorbent concentration of the absorbent used in the absorption tower of a flue gas desulfurization plant is predicted as the state quantity x. Other examples include gas turbines, steam turbines, large refrigerators, air conditioners, and the like.

FIG. 5 is a diagram schematically showing the configuration of the desulfurization device 20. As shown in FIG. The desulfurization device 20 is installed along with a boiler (not shown) of plant equipment such as a thermal power plant, and includes a dust collector 22 that collects particulates contained in exhaust gas G0 flowing through an exhaust passage 23a of the boiler, and a dust collector. 22, an absorption tower 24 installed in an exhaust passage 23b through which exhaust gas G1 that has passed through the dust collector 22 flows.

The dust collector 22 performs corona discharge on the exhaust gas G0 supplied into the casing to charge the fine particles contained in the exhaust gas G0, and causes them to adhere to the positively and negatively charged adhering portions by electric attraction. This is an electric precipitator that collects dust by The exhaust gas G1 subjected to dust collection processing by the dust collector 22 is supplied to the absorption tower 24 via the exhaust passage 23b.

The absorption tower 24 absorbs SO ₂ (sulfur dioxide) in the exhaust gas G1 by bringing an absorption liquid 26 containing limestone 30 into contact with the exhaust gas G1 that has been subjected to dust collection processing in the dust collector 22. , perform desulfurization treatment. Absorption liquid 26 is stored at the bottom of absorption tower 24 . The absorption liquid 26 is generated by mixing limestone 30 supplied from a limestone feeder 28 provided outside the absorption tower 24 with water 32 supplied to the bottom of the absorption tower 24 .
In this embodiment, a case is illustrated in which the absorption liquid 26 is generated by supplying limestone 30 to water 32 from the limestone feeder 28, but instead of this, limestone slurry containing limestone is supplied to water 32. The absorption liquid 26 may be generated by supplying the liquid.

The absorption liquid 26 stored at the bottom of the absorption tower 24 is pumped by an absorption liquid circulation pump 34, and is supplied to the upper part of the absorption tower 24 through an absorption liquid circulation pipe 36 provided outside the absorption tower 24. Ru. The absorption liquid circulation pump 34 is composed of a plurality of pump units connected in parallel to each other, and the operating state of each pump unit is controlled. For example, if the pump unit is of a variable capacity type (moving blade type), the flow rate of the absorption liquid 26 pumped from the absorption liquid circulation pump 34 can be controlled by variably adjusting the capacity of each pump unit. Further, when the pump unit is of a fixed capacity type (fixed blade type), the flow rate of the absorption liquid 26 pumped from the absorption liquid circulation pump 34 can be controlled by adjusting the number of operating pump units. The absorption liquid 26 supplied to the upper part of the absorption tower 24 comes into contact with the exhaust gas G1 rising inside the absorption tower 24 in the process of being sprayed from the nozzle 38 provided at the upper part of the absorption tower 24 and falling. do. Thereby, SO ₂ contained in the exhaust gas G1 reacts with the limestone 30 in the absorption liquid 26, and a desulfurization process is performed.
The method of dispersing and dropping the absorption liquid 26 from the nozzle 38 may be a grid method, a liquid column method, or a spray method.

The following formula (5) is a chemical reaction formula for the desulfurization treatment carried out in the absorption tower 24. In the desulfurization reaction, limestone 30 and SO ₂ contained in the exhaust gas G1 react to generate gypsum 34 (CaCO ₄ .2H ₂ O) as a byproduct. The exhaust gas G2 from which SO ₂ has been removed is discharged from the top of the absorption tower 24 to the outside via the desulfurization exhaust gas pipe 25.
SO ₂ +1/2O ₂ +CaCO ₃ +2H ₂ O→CaCO _{4.2H 2} O+CO ₂ (5)

In addition, a part of the absorption liquid 26 stored at the bottom of the absorption tower 24 is pumped by an absorption liquid circulation pump 34 and passes through an extraction pipe 40 branched from an absorption liquid circulation pipe 36 outside the absorption tower 24 to a dehydrator. Sent to 42. The dehydrator 42 is composed of, for example, a belt filter, dehydrates the absorbent liquid 26 while being conveyed by the belt filter, and discharges the generated gypsum 34 to the outside of the system.
Note that the filtrate produced by the dehydration process in the dehydrator 42 is reused by being supplied to the bottom of the absorption tower 24 as water 32.

Further, oxidizing air 46 is supplied to the bottom of the absorption tower 24 . As a result, the oxidizing air 46 is included in the absorption liquid 26, which promotes the oxidation of sulfite groups generated by transferring from the SO2 exhaust gas into the absorption liquid 26 into sulfuric acid groups, and as a result, the SO ₂ in the exhaust gas is reduced. Removal efficiency is also improved.
Note that when the method of spraying and dropping the absorption liquid 26 from the nozzle 38 is a grid type, the supply of the oxidizing air 46 may be omitted because the absorption liquid 26 is oxidized in the process of falling.

At least one sensor that can be selected as the input parameter u described above is arranged in such a desulfurization device 20. In this embodiment, an SO2 concentration sensor 50 for detecting the SO2 concentration u1 at the outlet side of the absorption tower 24 (desulfurization outlet SO2 concentration [ppm]) and an SO2 concentration sensor 50 for detecting the SO2 concentration u2 at the inlet side of the absorption tower 24 (desulfurization inlet SO2 concentration [ppm]) and a limestone slurry flow rate sensor 54 to detect the flow rate u3 of limestone slurry generated in the absorption tower 24 (absorption tower limestone slurry flow rate [m ³ /h]). , a boiler air flow rate sensor 56 for detecting the boiler air flow rate u5 [%], and a concentration u6 of the limestone slurry generated in the absorption tower 24 (a limestone slurry sensor 56 for detecting the concentration [wt%] of the limestone slurry in the absorption tower 24). The concentration sensor 58, the oxidizing air flow rate sensor 60 for detecting the oxidizing air flow rate u7 [m ³ N/h] supplied to the absorption tower 24, and the pHu8 of the absorption liquid 26 in the absorption tower 24 (absorption tower pH ) and a level sensor 64 for detecting the level u9 (absorption tower level [m]) of the absorption liquid 26 in the absorption tower 24.The detected values of these sensors is input to the control device 200, which is a control unit of the desulfurization device 20, so that the control device 200 controls each part of the desulfurization device 20.
Furthermore, a power generation command signal u4 for a generator (not shown) that generates power using steam generated by a boiler (not shown) can also be obtained as an input parameter u.

At least one of the detected values of each of these sensors is inputted as the input parameter u to the state quantity prediction devices 1A and 1B, and the state quantity prediction devices 1A and 1B input the detected value as the state quantity x in the absorption tower 24. Predict the absorbent (calcium carbonate) concentration [mmol/L].

In the state quantity prediction devices 1A and 1B, various parameters related to the desulfurization device 20 can be used as the physical parameter θ _P of the physical model M. For example, the activity of limestone in the absorption liquid 26, the inlet of the absorption tower 24, It may include at least one of the moisture content in the gas and the humidity increase rate in the absorption tower 24.

The state quantity prediction devices 1A and 1B having such a configuration are connected to each sensor disposed in the desulfurization device 20 via a network, so that the results detected by each sensor are used as input parameters u. Can be obtained. Thereby, the state quantity prediction device 1 can predict the absorbent (calcium carbonate) concentration [mmol/L] as the state quantity x corresponding to the input parameter u, and can utilize it for the operation of the desulfurization device 20.

Note that it is possible to adopt a configuration as a state quantity prediction system executed by an information processing device 70 that is communicably connected to the state quantity prediction device 1A or the state quantity prediction device 1B in the above embodiment. Here, FIG. 6 is a block diagram showing the configuration of a state quantity prediction system according to an embodiment.

The information processing device 70 includes a display unit 71 that displays the predicted value X calculated by the state quantity prediction device 1A or the state quantity prediction device 1B. Further, the information processing device 70 may be configured to execute each process in the state quantity prediction device 1A or the state quantity prediction device 1B in response to a request from the information processing device 70 according to instructions inputted through the display unit 71. good.

In addition, the components in the embodiments described above can be replaced with well-known components as appropriate without departing from the spirit of the present disclosure, and the embodiments described above may be combined as appropriate.

The contents described in each of the above embodiments can be understood as follows, for example.

(1) A state quantity prediction device according to one aspect includes:
A state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model,
A first prediction for storing a first prediction model that corresponds to a first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. a model storage section,
a measurement data acquisition unit for acquiring measurement data of the second plant;
a second predictive model creation unit for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
a state quantity prediction unit for predicting the state quantity of the second plant using the second prediction model;
Equipped with
The second predictive model creation unit creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

According to the aspect (1) above, the second prediction model corresponding to the second plant can be created by adjusting the first prediction model corresponding to the first plant. Such adjustment of the prediction model is performed by regularizing and learning the difference between the machine learning parameters of the first prediction model and the second prediction model. For example, even if only a small number of second measurement data can be obtained through simple measurements in a second plant, by restricting the degree of freedom in learning machine learning parameters and promoting the updating of physical parameters, it is possible to Therefore, a reliable second prediction model can be created. By using the second prediction model created in this way, it is possible to predict state quantities with high accuracy even in the second plant where it is difficult to obtain sufficient measurement data.

(2) In another aspect, in the aspect of (1) above,
The second predictive model creation unit creates the second predictive model by performing transfer learning so that a loss function including a difference between the machine learning parameters of the first predictive model and the second predictive model is minimized. do.

According to the aspect (2) above, the machine learning parameters can be learned freely by performing transfer learning so that the loss function including the magnitude of the difference between the machine learning parameters of the first prediction model and the second learning model is minimized. This allows learning to be regularized to prevent the degree from becoming excessively large.

(3) In another aspect, in the aspect (1) or (2) above,
The second predictive model creation unit sets an allowable range for the physical parameters of the first predictive model and the second predictive model.

According to the aspect (3) above, the second learning model is capable of predicting state quantities with higher accuracy by setting tolerance ranges for physical parameters during learning and limiting the degree of freedom in learning physical parameters. is obtained.

(4) In another aspect, in any one of the above (1) to (3),
The second predictive model creation unit adjusts the second predictive model so that the predicted value of the state quantity predicted by the state quantity prediction unit becomes the actual measured value of the state quantity.

According to the aspect (4) above, when the operating conditions of the second plant change by adjusting the second prediction model so that the predicted value of the state quantity predicted by the second prediction model becomes the actual measured value. Even in this case, the prediction accuracy of the second prediction model can be suitably maintained.

(5) In another aspect, in any one of the above (1) to (4),
The plant is a flue gas desulfurization plant for desulfurizing the flue gas by bringing an absorption liquid into contact with the flue gas in an absorption tower,
The state quantity is the absorbent concentration of the absorption liquid in the absorption tower.

According to the aspect (5) above, the absorbent concentration of the absorption liquid in the absorption tower included in the flue gas desulfurization plant can be suitably predicted as a state quantity.

(6) In another aspect, in the aspect (5) above,
The input parameters include the SO2 concentration at the desulfurization outlet of the absorption tower, the SO2 concentration at the desulfurization inlet of the absorption tower, the flow rate or concentration of limestone slurry produced in the absorption tower, and the steam produced in the boiler that discharges the flue gas. A power generation command signal to a generator that generates power, an air flow rate in a boiler that discharges the flue gas, an oxidizing air flow rate supplied to the absorption tower, a pH of the absorption liquid in the absorption tower, or a power generation command signal in the absorption tower. and at least one level of absorbent liquid.

According to the aspect (6) above, by including at least one of these parameters in the input parameters, the absorbent concentration of the absorbent in the absorption tower provided in the flue gas desulfurization plant can be suitably predicted as a state quantity.

(7) A state quantity prediction system according to one aspect includes:
A state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, and an information processing device capable of communicating with the state quantity prediction device, the system comprising:
The state quantity prediction device includes:
A first prediction for storing a first prediction model that corresponds to a first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. a model storage section,
a measurement data acquisition unit for acquiring measurement data of the second plant after being requested by the information processing device;
a second predictive model creation unit for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
a state quantity prediction unit for predicting the state quantity of the second plant using the second prediction model;
Equipped with
The second predictive model creation unit creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

According to the aspect (7) above, a state quantity prediction system including an information processing device capable of communicating with the state quantity prediction device according to each of the above-described aspects is realized. In this system, the state quantity prediction device predicts the state quantity of the second plant as described above in response to a request from the information processing device, and the prediction result can be obtained by the information processing device.

(8) A state quantity prediction method according to one aspect is:
A state quantity prediction method for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, the method comprising:
storing a first predictive model that corresponds to a first plant and is the predictive model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned;
a step of acquiring measurement data of the second plant;
creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
predicting the state quantity of the second plant using the second prediction model;
Equipped with
In the step of creating the second predictive model, the second predictive model is created by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

According to the aspect (8) above, the second prediction model corresponding to the second plant can be created by adjusting the first prediction model corresponding to the first plant. Such adjustment of the prediction model is performed by regularizing and learning the difference between the machine learning parameters of the first prediction model and the second prediction model. For example, even if only a small number of second measurement data can be obtained through simple measurements in a second plant, by restricting the degree of freedom in learning machine learning parameters and promoting the updating of physical parameters, it is possible to Therefore, a reliable second prediction model can be created. By using the second prediction model created in this way, it is possible to predict state quantities with high accuracy even in the second plant where it is difficult to obtain sufficient measurement data.

(9) A control method for a state quantity prediction system according to one aspect includes:
A control method for a state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter and an information processing device capable of communicating with each other using a prediction model including a machine learning model and a physical model. hand,
The state quantity prediction device includes:
a measurement data acquisition step for acquiring measurement data of the second plant after being requested by the information processing device;
Using the measurement data, create a first prediction model that corresponds to the first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. A second predictive model for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model stored in a first predictive model storage unit for storage. creation step,
a state quantity prediction step for predicting the state quantity of the second plant using the second prediction model;
Run
The second predictive model creation step creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model.

According to the aspect (9) above, a state quantity prediction system including an information processing device capable of communicating with the state quantity prediction device according to each of the above-described aspects is controlled. The state quantity prediction device predicts the state quantity of the second plant as described above by implementing the state quantity prediction method in response to a request from the information processing device, and the prediction result can be obtained by the information processing device. .

(10) A state quantity prediction device according to one aspect includes:
A state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model,
Predictive model creation for creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant Department and
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit creates the predictive model by regularizing and learning the machine learning parameters of the predictive model.

According to the aspect (10) above, when learning a predictive model using learning data acquired in a plant, the predictive model is created by regularizing and learning the machine learning parameters of the predictive model. will be held. In general, when a predictive model is trained using a small amount of training data, learning of machine learning parameters becomes dominant, and physical parameter learning may not be performed appropriately. In contrast, in this aspect, even when the number of learning data is small, regularization is performed to prevent machine learning parameters from becoming excessively large, thereby limiting the degree of freedom of machine learning parameters. Updating of physical parameters is promoted, overfitting is avoided and a highly accurate predictive model can be obtained.

(11) In another aspect, in the aspect (10) above,
The predictive model creation unit performs learning so that a loss function including the magnitude of the machine learning parameter of the predictive model is minimized.

According to the aspect (11) above, regularization is performed to prevent the machine learning parameters of the prediction model from becoming excessively large by learning so that the loss function including the size of the machine learning parameters of the prediction model is minimized. learning is possible.

(12) In another aspect, in the aspect (10) or (11) above,
The predictive model creation unit sets an allowable range for the physical parameter during learning.

According to the aspect (12) above, by setting an allowable range for physical parameters during learning and limiting the degree of freedom in learning physical parameters, a learning model that can predict state quantities with higher accuracy is created. can.

(13) In another aspect, in any one of the above (10) to (12),
The prediction model creation unit adjusts the prediction model so that the predicted value of the state quantity predicted by the state quantity prediction unit becomes the actual measured value of the state quantity.

According to the aspect (13) above, by adjusting the prediction model so that the predicted value of the state quantity predicted by the prediction model becomes the actual measured value, the prediction model can be adjusted even when the operating conditions of the plant change. Prediction accuracy can be suitably maintained.

(14) In another aspect, in any one of the above (10) to (13),
The plant is a flue gas desulfurization plant for desulfurizing the flue gas by bringing an absorption liquid into contact with the flue gas in an absorption tower,
The state quantity is the absorbent concentration of the absorption liquid in the absorption tower.

According to the aspect (14) above, the absorbent concentration of the absorption liquid in the absorption tower included in the flue gas desulfurization plant can be suitably predicted as a state quantity.

(15) In another aspect, in the aspect of (14) above,
The input parameters include the SO2 concentration at the desulfurization outlet of the absorption tower, the SO2 concentration at the desulfurization inlet of the absorption tower, the flow rate or concentration of limestone slurry produced in the absorption tower, and the steam produced in the boiler that discharges the flue gas. A power generation command signal to a generator that generates power, an air flow rate in a boiler that discharges the flue gas, an oxidizing air flow rate supplied to the absorption tower, a pH of the absorption liquid in the absorption tower, or a power generation command signal in the absorption tower. and at least one level of absorbent liquid.

According to the aspect (15) above, by including at least one of these parameters in the input parameters, the absorbent concentration of the absorbent in the absorption tower provided in the flue gas desulfurization plant can be suitably predicted as a state quantity.

(16) A state quantity prediction system according to one aspect includes:
A state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, and an information processing device capable of communicating with the state quantity prediction device, the system comprising:
The state quantity prediction device includes:
Predictive model creation for creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant Department and
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit creates the predictive model by regularizing and learning the machine learning parameters of the predictive model.

According to the aspect (16) above, a state quantity prediction system including an information processing device capable of communicating with the state quantity prediction device according to each of the above-described aspects is realized. In this system, the state quantity prediction device predicts the state quantity of the plant as described above in response to a request from the information processing apparatus, and the prediction result can be obtained by the information processing apparatus.

(17) A state quantity prediction method according to one aspect includes:
A state quantity prediction method for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, the method comprising:
creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant;
predicting the state quantity of the plant using the prediction model;
Equipped with
In the step of creating the predictive model, the predictive model is created by regularizing and learning the machine learning parameters of the predictive model.

According to the aspect (17) above, when learning a predictive model using learning data acquired in a plant, the predictive model is trained by regularizing the machine learning parameters of the predictive model. Creation takes place. In general, when a predictive model is trained using a small amount of training data, learning of machine learning parameters becomes dominant, and physical parameter learning may not be performed appropriately. In contrast, in this aspect, even when the number of learning data is small, regularization is performed to prevent machine learning parameters from becoming excessively large, thereby limiting the degree of freedom of machine learning parameters. Updating of physical parameters is promoted, overfitting is avoided and a highly accurate predictive model can be obtained.

(18) A control method for a state quantity prediction system according to one aspect includes:
A control method for a state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter and an information processing device capable of communicating with each other using a prediction model including a machine learning model and a physical model. hand,
The state quantity prediction device includes:
After the request from the information processing device, the machine learning parameters included in the machine learning model and the physical parameters included in the physical model are learned by using learning data acquired in the plant. a predictive model creation unit for creating a predictive model;
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit creates the predictive model by regularizing and learning the machine learning parameters of the predictive model.

According to the aspect (18) above, a state quantity prediction system including an information processing device capable of communicating with the state quantity prediction device according to each of the above-described aspects is controlled. The state quantity prediction device predicts the state quantity of the plant as described above by implementing the state quantity prediction method in response to a request from the information processing device, and the prediction result can be obtained by the information processing device.

1 (1A, 1B) State quantity prediction device 4 First predicted value calculation unit 6 Second predicted value calculation unit 8 State quantity predicted value calculation unit 12 Input layer 13 Intermediate layer 14 Output layer 15 First prediction model storage unit 16 Measured data Acquisition unit 17 Second prediction model creation unit 18 State quantity prediction unit 20 Desulfurization device 22 Dust collectors 23a, 23b Exhaust passage 24 Absorption tower 25 Desulfurization exhaust gas pipe 26 Absorption liquid 28 Limestone feeder 30 Limestone 32 Water 34 Absorption liquid circulation pump 36 Absorption Liquid circulation piping 38 Nozzle 40 Extraction pipe 42 Dehydrator 46 Oxidizing air 50, 52, 58 Concentration sensor 54 Flow sensor 56 Boiler air flow sensor 60 Oxidizing air flow sensor 62 Sensor 64 Level sensor DE Neural О
Dm Measured data M Prediction model M1 First prediction model M2 Second prediction model M _P Physical model M _N Machine learning model P1 First plant P2 Second plant u Input parameter x State quantity

Claims (18)

A state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model,
A first prediction for storing a first prediction model that corresponds to a first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. a model storage section,
a measurement data acquisition unit for acquiring measurement data of the second plant;
a second predictive model creation unit for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
a state quantity prediction unit for predicting the state quantity of the second plant using the second prediction model;
Equipped with
The second predictive model creation unit creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model. Device. The second predictive model creation unit creates the second predictive model by performing transfer learning so that a loss function including a difference between the machine learning parameters of the first predictive model and the second predictive model is minimized. The state quantity prediction device according to claim 1. The state quantity prediction device according to claim 1 or 2, wherein the second prediction model creation section sets an allowable range for the physical parameters of the first prediction model and the second prediction model. 2. The second predictive model creation unit adjusts the second predictive model so that the predicted value of the state quantity predicted by the state quantity prediction unit becomes the actual measured value of the state quantity. The state quantity prediction device described in . The plant is a flue gas desulfurization plant for desulfurizing the flue gas by bringing an absorption liquid into contact with the flue gas in an absorption tower,
The state quantity prediction device according to claim 1 or 2, wherein the state quantity is an absorbent concentration of the absorption liquid in the absorption tower. The input parameters include the SO2 concentration at the desulfurization outlet of the absorption tower, the SO2 concentration at the desulfurization inlet of the absorption tower, the flow rate or concentration of limestone slurry produced in the absorption tower, and the steam produced in the boiler that discharges the flue gas. A power generation command signal to a generator that generates power, an air flow rate in a boiler that discharges the flue gas, an oxidizing air flow rate supplied to the absorption tower, a pH of the absorption liquid in the absorption tower, or a power generation command signal in the absorption tower. The state quantity prediction device according to claim 5, comprising at least one of the levels of the absorption liquid. A state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, and an information processing device capable of communicating with the state quantity prediction device, the system comprising:
The state quantity prediction device includes:
A first prediction for storing a first prediction model that corresponds to a first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. a model storage section,
a measurement data acquisition unit for acquiring measurement data of the second plant after being requested by the information processing device;
a second predictive model creation unit for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
a state quantity prediction unit for predicting the state quantity of the second plant using the second prediction model;
Equipped with
The second predictive model creation unit creates the second predictive model by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model. system. A state quantity prediction method for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, the method comprising:
storing a first predictive model that corresponds to a first plant and is the predictive model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned;
a step of acquiring measurement data of the second plant;
creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model using the measurement data;
predicting the state quantity of the second plant using the second prediction model;
Equipped with
In the step of creating the second predictive model, the second predictive model is created by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model. Volume prediction method. A control method for a state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter and an information processing device capable of communicating with each other using a prediction model including a machine learning model and a physical model. hand,
The state quantity prediction device includes:
a measurement data acquisition step for acquiring measurement data of the second plant after being requested by the information processing device;
Using the measurement data, create a first prediction model that corresponds to the first plant and is the prediction model in which machine learning parameters included in the machine learning model and physical parameters included in the physical model are respectively learned. A second predictive model for creating a second predictive model that is the predictive model corresponding to the second plant by adjusting the first predictive model stored in a first predictive model storage unit for storage. creation step,
a state quantity prediction step for predicting the state quantity of the second plant using the second prediction model;
Run
The second predictive model creation step includes state quantity prediction, in which the second predictive model is created by regularizing and learning the difference between the machine learning parameters of the first predictive model and the second predictive model. How to control the system. A state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model,
Predictive model creation for creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant Department and
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The predictive model creation unit is a state quantity prediction device that creates the predictive model by regularizing and learning the machine learning parameters of the predictive model. The state quantity prediction device according to claim 10, wherein the prediction model creation unit performs learning so that a loss function including the magnitude of the machine learning parameter of the prediction model is minimized. The state quantity prediction device according to claim 10 or 11, wherein the prediction model creation unit sets an allowable range for the physical parameter during learning. The state according to claim 10 or 11, wherein the prediction model creation unit adjusts the prediction model so that the predicted value of the state quantity predicted by the state quantity prediction unit becomes the actual measured value of the state quantity. Volume prediction device. The plant is a flue gas desulfurization plant for desulfurizing the flue gas by bringing an absorption liquid into contact with the flue gas in an absorption tower,
The state quantity prediction device according to claim 10 or 11, wherein the state quantity is an absorbent concentration of the absorption liquid in the absorption tower. The input parameters include the SO2 concentration at the desulfurization outlet of the absorption tower, the SO2 concentration at the desulfurization inlet of the absorption tower, the flow rate or concentration of limestone slurry produced in the absorption tower, and the steam produced in the boiler that discharges the flue gas. A power generation command signal to a generator that generates power, an air flow rate in a boiler that discharges the flue gas, an oxidizing air flow rate supplied to the absorption tower, a pH of the absorption liquid in the absorption tower, or a power generation command signal in the absorption tower. The state quantity prediction device according to claim 14, comprising at least one of the levels of the absorption liquid. A state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, and an information processing device capable of communicating with the state quantity prediction device, the system comprising:
The state quantity prediction device includes:
Predictive model creation for creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant Department and
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The prediction model creation unit is a state quantity prediction system that creates the prediction model by regularizing and learning the machine learning parameters of the prediction model. A state quantity prediction method for predicting a state quantity of a plant corresponding to an input parameter using a prediction model including a machine learning model and a physical model, the method comprising:
creating the predictive model by learning machine learning parameters included in the machine learning model and physical parameters included in the physical model using learning data acquired in the plant;
predicting the state quantity of the plant using the prediction model;
Equipped with
In the step of creating the predictive model, the predictive model is created by regularizing and learning the machine learning parameters of the predictive model. A control method for a state quantity prediction system comprising a state quantity prediction device for predicting a state quantity of a plant corresponding to an input parameter and an information processing device capable of communicating with each other using a prediction model including a machine learning model and a physical model. hand,
The state quantity prediction device includes:
After the request from the information processing device, the machine learning parameters included in the machine learning model and the physical parameters included in the physical model are learned by using learning data acquired in the plant. a predictive model creation unit for creating a predictive model;
a state quantity prediction unit for predicting the state quantity of the plant using the prediction model;
Equipped with
The control method for a state quantity prediction system, wherein the prediction model creation unit creates the prediction model by regularizing and learning the machine learning parameters of the prediction model.

Images