JP2020194453A - Model prediction control system, information processing device, program, and model prediction control method - Google Patents

Abstract

To improve a technique relating to process model prediction control.SOLUTION: A model prediction control system 1 comprises: storage means 131 for storing a process model; process control means 132 for performing model prediction control of a process using the process model; storage means 133 for storing a control output and a control input of the process as time series data; and setting means 134 for setting a target value of a model parameter of the process model by machine learning using the accumulated time series data as training data. The process control means 132 performs model prediction control while asymptotically bringing the model parameters closer to the target value over time.SELECTED DRAWING: Figure 2

Description

The present invention relates to a model predictive control system, an information processing device, a program, and a model predictive control method.

Conventionally, as a process control method, a model predictive control (MPC; Model Predictive Control) using a mathematical model of a process to be controlled (hereinafter, also referred to as a “process model”) is known. Further, there is known a method of periodically updating a process model to solve the problem that the accuracy of model prediction control may decrease when the process characteristics change due to factors such as aged deterioration of equipment. For example, Patent Document 1 describes a model prediction control method using online model identification, and in executing a model prediction control calculation, model reidentification is performed to derive a process dynamic model to be used each time based on actual operation data. The technology is disclosed.

Japanese Unexamined Patent Publication No. 2011-198327

In the prior art, the model parameters of the process model can change each time the model re-identification is performed, so that the model parameters become a discontinuous step function at each execution timing of the model re-identification. However, if a discontinuity occurs in the change of model parameters, inconveniences such as a decrease in control stability may occur. Therefore, there was room for improvement in the technology related to process model predictive control.

An object of the present invention made in view of such circumstances is to improve a technique relating to model predictive control of a process.

The model prediction control system according to the embodiment of the present invention is
A storage means for storing the process model and
A process control means for performing model predictive control of a process using the process model, and
A storage means for accumulating the control output and control input of the process as time series data,
A setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as learning data, and
It is a model prediction control system equipped with
The process control means carries out the model prediction control while asymptotically bringing the model parameter closer to the target value over time.

The information processing device according to the embodiment of the present invention is
A storage means for storing the process model and
A process control means for performing model predictive control of a process using the process model, and
A storage means for accumulating the control output and control input of the process as time series data,
A setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as learning data, and
It is an information processing device equipped with
The process control means carries out the model prediction control while asymptotically bringing the model parameter closer to the target value over time.

The information processing device according to the embodiment of the present invention is
A storage means for storing a process model, a process control means for performing model prediction control of a process using the process model, a storage means for storing control outputs and control inputs of the process as time-series data, and storage means. A setting means for setting a target value of a model parameter of the process model by machine learning using the time-series data as training data is provided, and the process control means gradually brings the model parameter closer to the target value over time. To configure a model predictive control system that implements the model predictive control
It includes at least one of the storage means, the process control means, the storage means, and the setting means.

The program according to the embodiment of the present invention
Information processing device,
A storage means for storing the process model and
A process control means for performing model predictive control of a process using the process model, and
A storage means for accumulating the control output and control input of the process as time series data,
A setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as learning data, and
It is a program that functions as
The process control means carries out the model prediction control while asymptotically bringing the model parameter closer to the target value over time.

The program according to the embodiment of the present invention
A storage means that includes a plurality of information processing devices that are communicably connected and stores a process model, a process control means that performs model prediction control of a process using the process model, and a control output of the process. The process is provided with a storage means for accumulating control inputs as time-series data and a setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as training data. One of the plurality of information processing devices in a model prediction control system in which the control means performs the model prediction control while gradually bringing the model parameters closer to the target value with the passage of time.
It functions as at least one of the storage means, the process control means, the storage means, and the setting means.

The model prediction control method according to the embodiment of the present invention is
It is a model prediction control method of the process executed by the information processing device.
Steps to memorize the process model and
A control step for performing model prediction control of a process using the process model, and
A step of accumulating the control output and control input of the process as time series data,
A step of setting target values of model parameters of the process model by machine learning using the accumulated time series data as learning data, and
Including
In the control step, the model prediction control is performed while asymptotically approaching the model parameter to the target value with the passage of time.

According to one embodiment of the present invention, techniques relating to process model predictive control are improved.

It is a figure which shows the schematic structure of the model prediction control system which concerns on one Embodiment of this invention. It is a block diagram which shows the schematic structure of an information processing apparatus. It is a figure which shows the 1st time series data which is stored in an information processing apparatus. It is a figure which shows the 2nd time series data which is stored in the information processing apparatus. It is a flowchart which shows the operation of an information processing apparatus. It is a figure which shows the 1st example of the time change of a model parameter. It is a figure which shows the 2nd example of the time change of a model parameter.

Hereinafter, embodiments of the present invention will be described.

(Configuration of model prediction control system)
The model prediction control system 1 according to the embodiment of the present invention will be described with reference to FIG. The model predictive control system 1 is a system used to control a process in an arbitrary plant such as a water treatment plant. The model prediction control system 1 includes an information processing device 10 and one or more plant facilities 20. In FIG. 1, only one information processing device 10 and three plant facilities 20 are shown, but the model prediction control system 1 includes an arbitrary number of information processing devices 10 and plant equipment 20. You may.

The information processing device 10 is a device that performs model prediction control of an arbitrary process to be controlled. Specifically, the information processing apparatus 10 predicts the future control output by using the process model of the process to be controlled for each control cycle, and controls the input so that the actual control output approaches a desired value. To determine. For example, when the water purification process in a water treatment plant is to be controlled, the control output is process data such as nitric acid concentration and / or ammonia concentration in the treated water, and the amount of air blown to the treated water or the driving power of the blower. The amount of operation such as, etc. can be used as the control input. However, the process to be controlled and the control output and control input of the process are not limited to the above example and can be arbitrarily determined.

Further, the information processing device 10 can communicate with the plant equipment 20 installed in the plant via, for example, the Internet or a network N such as a LAN (Local Area Network). Here, the plant equipment 20 is any equipment used to carry out a process in the plant. For example, the plant equipment 20 used to carry out a water purification process in a water treatment plant includes on-site equipment such as a sensor that detects process data such as nitrate concentration and / or ammonia concentration in the treated water, and a blower that blows air to the treated water. , And equipment including PLC (Programmable Logic Controller) that controls the field equipment, but is not limited to these. The information processing device 10 is installed in, for example, the central control room of the plant, but the information processing device 10 is not limited to this and can be installed in any place.

Here, the outline of the present embodiment will be described, and the details will be described later. The information processing device 10 executes model prediction control of a process using a process model, and stores the control output and control input of the process as time series data (hereinafter, also referred to as “first time series data”). The information processing device 10 sets the target value of the model parameter of the process model based on the accumulated first time series data. Then, the information processing apparatus 10 carries out model prediction control while asymptotically bringing the model parameters closer to the target value over time.

According to such a configuration, since the model parameters gradually change with the passage of time, the chance of a substantial discontinuity occurring in the changes of the model parameters is reduced. Therefore, the probability that inconveniences such as a decrease in control stability will occur is reduced, and the technique for process model prediction control is improved. Since the model parameters change discretely in each control cycle, strictly speaking, discontinuities occur in each control cycle even when the changes in the model parameters are substantially continuous. "Substantial discontinuity" refers to a discontinuity other than such a discontinuity in the strict sense.

(Hardware configuration of information processing device)
The hardware configuration of the information processing apparatus 10 will be described with reference to FIG. The information processing device 10 includes a communication unit 11, a storage unit 12, and a control unit 13.

The communication unit 11 is one or more communication interfaces that communicate with an external device via wireless or wired. In the present embodiment, the communication unit 11 includes a communication interface that communicates with the plant equipment 20 via the network N. Further, the communication unit 11 includes a communication interface for communicating with each of the display device 30 and the input device 40. Here, the display device 30 is an arbitrary display such as a liquid crystal display or an OEL (Organic Electro-luminescence) display. The input device 40 is an arbitrary input interface that accepts operations by the user, such as a keyboard or a mouse.

The storage unit 12 is one or more memories. In the present embodiment, the "memory" is, for example, a semiconductor memory, a magnetic memory, an optical memory, or the like, but the memory is not limited to these and may be any memory. Each memory included in the storage unit 12 functions as, for example, a main storage device, an auxiliary storage device, or a cache memory. The storage unit 12 is built in the information processing device 10, for example, but can be connected to the information processing device 10 from the outside via an arbitrary interface.

The control unit 13 includes one or more processors. In the present embodiment, the "processor" is a general-purpose processor, a dedicated processor specialized for a specific process, or the like, but is not limited to these, and may be any processor. The control unit 13 controls the operation of the entire information processing device 10. The details of the operation of the information processing device 10 will be described later.

(Software configuration of information processing device)
The software configuration of the information processing apparatus 10 will be described with reference to FIG. One or more programs used for controlling the operation of the information processing device 10 are stored in the storage unit 12. When the one or more programs are read by the control unit 13, the control unit 13 functions as a storage means 131, a process control means 132, a storage means 133, a setting means 134, a determination means 135, and a determination means 136.

The storage means 131 is a means for storing the process model of the process to be controlled in the storage unit 12. A process model is a mathematical model with one or more model parameters. The process model further contains information on the initial values of each model parameter. The process model can be determined in advance, for example by experiment or simulation.

The process control means 132 is a means for performing model prediction control of a process using the process model stored in the storage unit 12. Specifically, the process control means 132 predicts a future control output (for example, process data such as nitric acid concentration and / or ammonia concentration) based on the process model stored in the storage unit 12 for each control cycle. The control input (for example, the amount of air blown or the amount of operation such as the drive power of the blower) is determined so that the actual control output approaches a desired value, and the determined control input is input to the plant equipment 20. The control cycle is sufficiently shorter than the update cycle described later.

Specifically, the process control means 132 performs model prediction control while asymptotically bringing the model parameters closer to the target value over time as described later. According to such a configuration, for example, the model parameter does not change instantaneously from the current value to the target value, but gradually changes with the passage of time, so that there is an opportunity for a substantial discontinuity to occur in the change of the model parameter. Reduce.

Here, when the start value is determined by the determination means 136, the process control means 132 overwrites the current value of the model parameter with the start value, and then asymptoticizes the model parameter to the target value with the passage of time. Let me. According to such a configuration, as will be described later, overfitting in machine learning for determining the target value of the model parameter can be alleviated or prevented.

The storage means 133 is a means for storing the control output and the control input of the process in the storage unit 12 as the first time series data. Specifically, the storage means 133 acquires a control output (for example, process data) from the plant equipment 20 via the communication unit 11 for each control cycle. Further, the storage means 133 acquires a control input (for example, an operation amount) to be input to the plant equipment 20 from the process control means 132 for each control cycle. Then, the storage means 133 stores the control output and the control input in the storage unit 12 as the first time series data. Therefore, the first time series data is data indicating current and past control outputs and control inputs. For example, as shown in FIG. 3, a combination of time data, process data as control output, and operation amount as control input is stored in the storage unit 12 as first time series data.

Further, the storage means 133 stores the target value of the model parameter set for each update cycle by the setting means 134 as time series data (hereinafter, also referred to as “second time series data”) in the storage unit 12. It also functions as a means of accumulating. Therefore, the second time series data is data showing current and past target values. For example, as shown in FIG. 4, a combination of time data and a target value is stored in the storage unit 12 as second time series data.

The setting means 134 is a means for setting the target value of the model parameter of the process model by machine learning using the first time series data accumulated in the storage unit 12 as learning data. Specifically, the setting means 134 searches for the estimated value of the model parameter by machine learning using at least a part of the first time series data stored in the storage unit 12 as learning data for each update cycle. The estimated value is set as the target value of the model parameter. Therefore, the target value of the model parameter is updated every update cycle.

The determination means 135 is a means for determining whether or not there is a possibility that overfitting has occurred in the above-mentioned machine learning based on the change in the target value of the model parameter set by the setting means 134 for each update cycle. Here, any method can be adopted for determining the presence or absence of the possibility of overfitting. For example, the determination means 135 determines that overfitting may have occurred if the variance of the current and past target values of the model parameters is less than a predetermined threshold.

When the determination means 135 determines that overfitting may have occurred, the determination means 136 determines a value not included in the numerical range from the current value of the model parameter to the current target value as a start value. It is a deciding means to do. Here, any method can be adopted for determining the start value. For example, the determination means 136 may determine an arbitrary value not included in the numerical range from the current value of the model parameter to the current target value as the start value. Alternatively, the determination means 136 may determine the start value based on the current and past target values of the model parameters. Specifically, the determination means 136 determines the average value, the weighted average value, or the mode value of the current and past target values within a certain period in the past. Then, when the numerical range from the current value of the model parameter to the current target value does not include the average value, the weighted average value, or the mode value, the determination means 136 determines the average value, the weighted average value, or the mode. Determine the mode as the start value. When the weighted average value is adopted, the weighting coefficient to be multiplied by each target value is determined so as to be larger as the target value is newer, but the present invention is not limited to this example and can be determined by an arbitrary algorithm.

In addition to the above-described means 131 to 136, the control unit 13 generally performs arbitrary processing that can be executed by the processor (for example, means for executing general arithmetic processing and timekeeping processing). It is also possible to have a configuration that further functions as.

(Operation of information processing device)
The operation of the information processing apparatus 10 will be described with reference to FIG. This operation is an operation of performing model prediction control while asymptotically bringing the model parameters closer to the target value with the passage of time.

Step S100: The storage means 131 stores the process model of the process to be controlled in the storage unit 1.

Step S102: The control unit 13 performs initial setting of model prediction control.

Specifically, the control unit 13 sets the variable i to 1. The variable i is a variable indicating which update cycle the current time calculated from the start of the model prediction control belongs to. Specifically, assuming that the current time calculated from the start of model prediction control is t and the length of the update cycle is T, (i-1) When T ≦ t <iT, the current time t is the i-th time. It belongs to the update cycle.

Further, the control unit 13 sets the target value G ₁ of the model parameter in the first update cycle and sets the initial value P ₀ of the model parameter. Here, the value of G ₁ is predetermined, and P ₀ is described as being equal to G ₁ .

Step S104: The process control unit 132 changes the model parameters so as to approach the target value G _i. The target value G _i is the target value of the model parameter in the i-th update cycle. In the i-th update cycle, step S104 is repeatedly executed for each control cycle, so that the model parameters asymptotically approach the target value G _i with the passage of time.

Here, as a function indicating the model parameters, any function that approaches the target value G _i can be employed with the lapse of time. For example, the model parameter in the i-th update cycle can be indicated by a function of time t calculated from the start of the update cycle. In one example, the model parameter P (t) in the i-th update cycle is represented by the following equation (1).
P (t) = G _i-1 + (G _{i −} G _i-1 ) · (1-exp (−α · t)) (1)
Note that α is a coefficient that determines how to approach the target value G _i in the i-th update cycle, and can be arbitrarily determined.

If the current value of the model parameter is equal to the target value G _i , the model parameter does not change. For example, as described above, the initial value P ₀ of the model parameter is equal to the target value G _{1 in the first} update cycle. Therefore, the model parameters do not change at least in the first update cycle (ie, 0 ≦ t <T).

Step S106: The process control means 132 predicts a future control output based on the process model, determines a control input so as to bring the actual control output closer to a desired value, and inputs the determined control input to the plant equipment 20. To do.

Step S108: The storage means 133 stores the control output and the control input of the process in the storage unit 12 as the first time series data.

Step S110: The control unit 13 determines whether or not the i-th update cycle has elapsed. Specifically, when t = iT, the control unit 13 determines that the i-th update cycle has elapsed (step S110-Yes), and the operation proceeds to step S112. On the other hand, when (i-1) T ≦ t <iT, the control unit 13 determines that the i-th update cycle has not elapsed (step S110-No), and the operation returns to step S104. By step S104~S110 are repeated, model predictive control is performed while asymptotic model parameter to the target value G _i with the lapse of time.

Step S112: The setting means 134 searches for an estimated value of the model parameter by machine learning using at least a part of the first time series data stored in the storage unit 12 as learning data, and uses the searched estimated value for the i + 1th time. It is set as the target value G _{i + 1} of the model parameter in the update cycle. The set target value G _{i + 1} is stored in the storage unit 12 by the storage means 133 as the second time series data.

Step S114: The determination means 135 determines whether or not there is a possibility that overfitting has occurred in the above-mentioned machine learning based on the change in the target value of the model parameter set for each update cycle. Specifically, the determination means 135 is overfitted when there is no change in the model parameters in the latest plurality of update cycles (for example, when the variance of the current and past target values of the model parameters is less than a predetermined threshold value). It is determined that there is a possibility that it has occurred (step S114-Yes), and the operation proceeds to step S116. On the other hand, the determination means 135 determines that there is no possibility that overfitting has occurred when there is a change in the model parameters in the latest plurality of update cycles (for example, when the variance is equal to or greater than the predetermined threshold value). (Step S114-No), the operation proceeds to step S120.

Step S116: The determining means 136 is included in the numerical range from the current value of the model parameter to the current target value (that is, the target value G _{i + 1} of the model parameter in the i + 1th update cycle set in step S112). The value that does not exist is determined as the start value S _{i + 1} .

Step S118: The process control means 132 overwrites the current value of the model parameter with the start value S _{i + 1} .

Step S120: The control unit 13 increments the variable i. After that, the operation returns to step S104.

(Time change of model parameters)
With reference to FIGS. 6 and 7, the time change of the model parameter when the information processing apparatus 10 executes the above operation will be specifically described. 6 and 7 are graphs in which the horizontal axis represents time and the vertical axis represents model parameters.

First, with reference to FIG. 6, a state in which the model parameters gradually approach the target value G _i with the passage of time will be described in each update cycle (2 ≦ i) after the second update. FIG. 6 shows an example of time change of model parameters in the first to third update cycles. In the first update cycle (i = 1), the initial value P ₀ of the model parameter is equal to the target value G ₁ , so the model parameter has not changed. Further, in the second update cycle (i = 2), the model parameters gradually approach from G ₁ toward the target value G ₂ with the passage of time. Further, in the third update cycle, the model parameters gradually approach from G ₂ toward the target value G ₃ with the passage of time. The model parameters in each update cycle from the second time onward are represented by, for example, the above equation (1).

As described above, according to the present embodiment, the model parameters do not change instantaneously at the start timing of each update cycle, but gradually change with the passage of time, so that the change of the model parameters is a substantial discontinuity. Is reduced. Therefore, the probability that inconveniences such as a decrease in control stability will occur is reduced, and the technique for process model prediction control is improved.

Next, referring to FIG. 7, when it is determined that overfitting may have occurred, the current value of the model parameter is overwritten with the start value, and then the model parameter is set to the target value Gi over time. I will explain how it approaches. FIG. 7 shows an example of the time change of the model parameter in the m-th to m + second update cycles. Here, it is assumed that m is a relatively large value and the model prediction control is continuously performed for a sufficiently long time. In the mth update cycle (i = m), the model parameters asymptotically approach from G _m-1 to the target value G _m with the passage of time. Further, in the m + 1th update cycle (i = m + 1), the model parameters gradually approach from G _m toward the target value G _{m + 1} with the passage of time. Here, it is assumed that there is a possibility that overfitting has occurred by the determination means 135 at the end of the m + 1th update cycle. Therefore, at time t = (m + 1) T, the model parameter is overwritten with the start value S _{m + 2} . Then, in the m + second update cycle (i = m + 2), the model parameters gradually approach from S _{m + 2} toward the target value G _{m + 2} with the passage of time.

As described above, according to the present embodiment, at the end of the i-th update cycle, the current value of the model parameter is exceptionally started when it is determined that overfitting may have occurred. Overwritten with the value S _{i + 1} . As described above, the start value S _{i + 1} is a numerical value from the value of the model parameter at the end of the i-th update cycle (that is, the value before overwriting) to the target value G _{i + 1} in the i + 1-th update cycle. A value that is not included in the range. Therefore, in the i + 1th update cycle, the model parameter changes in a numerical range different from the case where the model parameter is not overwritten with the start value S _{i + 1} , for example. That is, according to the present embodiment, when it is determined that there is a possibility that overfitting has occurred, fluctuations occur in the change mode of the model parameters.

As described above, in machine learning for determining the target value of the model parameter, the estimated value of the model parameter is searched, and the search range depends on the change mode of the past model parameter. For example, when the steady state of the process continues for a long period of time, the change in the past model parameters becomes small, so that the search range of the estimated value of the model parameters becomes relatively narrow and overfitting may occur. On the other hand, according to the present embodiment, since fluctuations are generated in the change mode of the model parameters, it is possible to alleviate or prevent overfitting in machine learning for determining the target value of the model parameters.

Although the present invention has been described with reference to the drawings and examples, it should be noted that those skilled in the art can easily make various modifications and modifications based on the present disclosure. Therefore, it should be noted that these modifications and modifications are within the scope of the present invention. For example, the functions included in each means, each step, etc. can be rearranged so as not to be logically inconsistent, and a plurality of means, steps, etc. can be combined or divided into one. ..

For example, the model prediction control system 1 may be configured to include a plurality of information processing devices 10. Further, each configuration or each means of one information processing device 10 according to the embodiment may be distributed and arranged in a plurality of devices capable of communicating with each other.

Further, in the above-described embodiment, an example of the operation of the information processing apparatus 10 has been described with reference to FIG. However, it is also possible that some steps included in the above-mentioned operations or some operations included in one step are omitted within a range that is logically consistent. Further, it is possible that the order of the plurality of steps included in the above-described operation is interchanged within a range that is logically consistent.

Further, in the above-described embodiment, each means realized by the control unit 13 has been described as a software configuration, but at least some of these means may be a concept including software resources and / or hardware resources. Good. For example, the storage means 131 and the storage means 133 may include a memory.

Further, in order to function as the information processing device 10 according to the above-described embodiment, a general-purpose information processing device such as a computer or a mobile phone can be used. The information processing apparatus stores a program describing processing contents for realizing each function of the information processing apparatus 10 according to the embodiment in the memory of the information processing apparatus, and reads the program by the processor of the information processing apparatus. It can be realized by executing it.

1 Model prediction control system 10 Information processing device 11 Communication unit 12 Storage unit 13 Control unit 131 Storage means 132 Process control means 133 Storage means 134 Setting means 135 Judgment means 136 Decision means 20 Plant equipment 30 Display device 40 Input device N network

Claims (9)

A storage means for storing the process model and
A process control means for performing model predictive control of a process using the process model, and
A storage means for accumulating the control output and control input of the process as time series data,
A setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as learning data, and
It is a model prediction control system equipped with
The process control means is a model prediction control system that carries out the model prediction control while asymptotically bringing the model parameters closer to the target value over time. The model prediction control system according to claim 1.
A determination means for determining whether or not overfitting may have occurred in the machine learning based on a change in the target value of the model parameter periodically set by the setting means.
When it is determined by the determination means that the overfitting may have occurred, a determination means for determining a value not included in the numerical range from the current value of the model parameter to the current target value as a start value. When,
With more
When the determination means determines that the overfitting may have occurred, the process control means overwrites the current value of the model parameter with the start value, and then the model parameter with the passage of time. A model predictive control system that asymptotically approaches the current target value. The model prediction control system according to claim 2.
The determination means is a model prediction control system that determines that the overfitting may have occurred when the variance of the current and past target values of the model parameters is less than a predetermined threshold value. The model predictive control system according to claim 2 or 3.
The determination means is a model prediction control system that determines the start value based on the current and past target values of the model parameter. A storage means for storing the process model and
A process control means for performing model predictive control of a process using the process model, and
A storage means for accumulating the control output and control input of the process as time series data,
A setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as learning data, and
It is an information processing device equipped with
The process control means is an information processing device that performs the model prediction control while asymptotically bringing the model parameters closer to the target value with the passage of time. A storage means for storing a process model, a process control means for performing model prediction control of a process using the process model, a storage means for storing control outputs and control inputs of the process as time-series data, and storage means. A setting means for setting a target value of a model parameter of the process model by machine learning using the time-series data as training data is provided, and the process control means gradually brings the model parameter closer to the target value over time. To configure a model predictive control system that implements the model predictive control
An information processing device including at least one of the storage means, the process control means, the storage means, and the setting means. Information processing device,
A storage means for storing the process model and
A process control means for performing model predictive control of a process using the process model, and
A storage means for accumulating the control output and control input of the process as time series data,
A setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as learning data, and
It is a program that functions as
The process control means is a program that carries out the model prediction control while asymptotically bringing the model parameters closer to the target value over time. A storage means that includes a plurality of information processing devices that are communicably connected and stores a process model, a process control means that performs model prediction control of a process using the process model, and a control output of the process. The process is provided with a storage means for accumulating control inputs as time-series data and a setting means for setting target values of model parameters of the process model by machine learning using the accumulated time-series data as training data. One of the plurality of information processing devices in a model prediction control system in which the control means performs the model prediction control while gradually bringing the model parameters closer to the target value with the passage of time.
A program that functions as at least one of the storage means, the process control means, the storage means, and the setting means. It is a model prediction control method of the process executed by the information processing device.
Steps to memorize the process model and
A control step for performing model prediction control of a process using the process model, and
A step of accumulating the control output and control input of the process as time series data,
A step of setting target values of model parameters of the process model by machine learning using the accumulated time series data as learning data, and
Including
A model prediction control method in which, in the control step, the model prediction control is performed while the model parameters are asymptotically approached to the target value with the passage of time.

Images