JP7462905B2 - CONTROL DEVICE, METHOD, PROGRAM AND SYSTEM - Google Patents

Description

This invention relates to a control device that performs feedback control.

Feedback control, for example PID control, is widely used to control various devices. Although feedback control belongs to classical control, it is still the main control method in industry today due to its reliability based on past performance and the ease of adjustment based on engineers' empirical rules.

Figure 11 shows the basic configuration of conventional feedback control, that is, a conventional feedback system 200. As is clear from the figure, the output y obtained from a detector (e.g., a sensor, etc.) of the control mechanism 202 is fed back to the input side on the left side of the figure, and the deviation from the target value r is calculated. This calculated deviation is further input to the controller 201, which calculates the manipulated variable u. The operator (e.g., an actuator, etc.) of the control mechanism 202 operates in response to this manipulated variable u, and controls a control target (not shown). At this time, a disturbance w may occur. This series of processes is repeated to perform control that reduces the deviation, that is, control that brings the output y closer to the target value r.

Meanwhile, the field of machine learning has been attracting a lot of attention in recent years. In this context, the inventors of the present application have proposed a new machine learning framework with a tree structure (learning tree) (Patent Document 1).

Figure 12 is an explanatory diagram showing the new machine learning framework described above, that is, an explanatory diagram showing the structure of a learning tree. Figure 12(a) shows the structure of a learning tree in this learning method, and Figure 12(b) shows an image of a state space corresponding to this structure. As is clear from the figure, the learning tree structure is configured by arranging each node corresponding to each hierarchically divided state space in a tree shape or a lattice shape from the top node (start node or root node) to the bottom node (terminal node or leaf node). Note that the figure shows an example of an N-layer d-dimensional n-partition learning tree where d is 2 and n is 2, and the numbers 1 to 4 assigned to the four terminal nodes in the first layer of the learning tree shown in Figure 12(a) correspond to the four state spaces shown in Figure 12(b), respectively.

When performing learning processing using the above learning tree, the input data is sequentially associated with each divided state space, and the data is accumulated in each state space. At this time, when new data is input into a state space where no data existed before, new nodes are generated sequentially. The predicted output is calculated by taking the arithmetic mean of the output values or output vectors corresponding to each piece of data contained in each state space after learning.

This type of machine learning technology makes it possible to achieve high-speed machine learning with low memory consumption.

JP 2016-173686 A

However, in conventional feedback control such as PID control, it was common to adjust and set the gain before control started, and then use the fixed gain after control started. Therefore, for example, when the characteristics of the controlled object or the operator change due to aging or other reasons, it was not possible to respond adaptively, and there was a risk of the accuracy of the control decreasing.

The present invention was made against the above technical background, and its purpose is to perform adaptive control based on data obtained during control while utilizing highly reliable feedback control that has been used for many years.

Further objects and effects of the present invention will be readily understood by those skilled in the art by referring to the following description of the specification.

The above-mentioned technical problems can be solved by a control device, method, program, system, etc. having the following configuration.

That is, the control device according to the present invention is a control device for performing feedback control on a predetermined device, and includes a first controller that generates a first operation amount for the device based on an output fed back from the device and a target value, a predicted output generation unit having a trained model that has been machine-learned to generate a predicted output from the device based on the output fed back from the device and the first operation amount, a second controller that generates a second operation amount for the device based on the predicted output and the target value, an integrated operation amount generation unit that generates an integrated operation amount that is an operation amount for the device based on the first operation amount and the second operation amount, and a storage unit that stores the first operation amount, the output fed back from the device, and the output from the device corresponding to the integrated operation amount as machine learning data when the second operation amount is invalidated.

This configuration makes it possible to use feedback control, a highly reliable control technique that has been used for many years, while also using machine learning technology to perform adaptive control based on data obtained during control.

The control device may further include a learning processing unit that performs learning processing based on the machine learning data to update the trained model.

With this configuration, learning processing can be performed while controlling the device, optimizing control.

The control device may further include a determination unit that determines whether the second operation amount satisfies an invalidation condition, and an invalidation processing unit that invalidates the second operation amount when the determination unit determines that the invalidation condition is satisfied.

With this configuration, when the second operation amount satisfies a predetermined condition, the second operation amount is invalidated and only control based on the first operation amount is performed, so that more reliable control can be performed. In addition, the data from the period is provided as data for machine learning, so that improvement in control accuracy can be expected in the future.

The invalidation condition may be that the second operation amount is greater than a first threshold value or is smaller than a second threshold value that is smaller than the first threshold value.

With this configuration, the second operation amount is invalidated when the operation amount exceeds the expected amount, so that more reliable control can be performed. In addition, the data from the period is provided as data for machine learning, so that improvement in control accuracy can be expected in the future.

The storage unit may further store, when the second operation amount is 0 or a value close to 0, the first operation amount, the output fed back from the device, and the output from the device corresponding to the integrated operation amount as data for machine learning.

With this configuration, learning can be continued even when the second manipulated variable is zero or a value close to zero, so further improvement in control accuracy can be expected.

The storage unit may further store, as machine learning data, the first operation amount relating to a reference time step when the second operation amount is invalidated, the output fed back from the device, and the output from the device corresponding to the integrated operation amount, as well as the first operation amount relating to one or more time steps prior to the reference time, the output fed back from the device, and the output from the device corresponding to the integrated operation amount.

With this configuration, data relating to one or more time steps prior to the reference time step is also learned, making generalization easier and improving learning speed is expected.

The first controller and/or the second controller may each perform any one of P control, PI control, PD control, or PID control.

This configuration allows for the use of highly reliable control technology that has been in use for many years, while also using machine learning technology to further improve control accuracy based on data obtained during operation of the equipment.

The trained model may be obtained by performing machine learning using a training model having a tree structure constructed by hierarchically arranging a plurality of nodes, each of which corresponds to a hierarchically divided state space.

This configuration makes it possible to perform high-speed learning with less memory than learning using artificial neural networks, etc., which is particularly advantageous when performing simultaneous learning (online learning) while the device is operating.

The present invention can also be conceived as a method. That is, the control method according to the present invention is a control method in a control device for performing feedback control on a predetermined device, the control device comprising: a first controller that generates a first operation amount for the device based on an output fed back from the device and a target value; a predicted output generating unit that has a trained model that has been machine-learned to generate a predicted output from the device based on the output fed back from the device and the first operation amount; and a second controller that generates a second operation amount for the device based on the predicted output and the target value, and an integrated operation amount generating step that generates an integrated operation amount that is an operation amount for the device based on the first operation amount and the second operation amount; and a storage step that stores the first operation amount, the output fed back from the device, and the output from the device corresponding to the integrated operation amount as machine learning data when the second operation amount is invalidated.

The present invention can also be conceived as a program. That is, the control program according to the present invention is a control program for a control device for performing feedback control on a predetermined device, the control device comprising: a first controller that generates a first operation amount for the device based on an output fed back from the device and a target value; a predicted output generating unit that has a trained model that has been machine-learned to generate a predicted output from the device based on the output fed back from the device and the first operation amount; and a second controller that generates a second operation amount for the device based on the predicted output and the target value, and an integrated operation amount generating step that generates an integrated operation amount that is an operation amount for the device based on the first operation amount and the second operation amount; and a storage step that stores the first operation amount, the output fed back from the device, and the output from the device corresponding to the integrated operation amount as machine learning data when the second operation amount is invalidated.

The present invention can also be conceived as a system. That is, the control system according to the present invention is a control system for performing feedback control on a predetermined device, and includes a first controller that generates a first operation amount for the device based on an output fed back from the device and a target value, a predicted output generating unit having a trained model that has been machine-learned to generate a predicted output from the device based on the output fed back from the device and the first operation amount, a second controller that generates a second operation amount for the device based on the predicted output and the target value, an integrated operation amount generating unit that generates an integrated operation amount that is an operation amount for the device based on the first operation amount and the second operation amount, and a storage unit that stores the first operation amount, the output fed back from the device, and the output from the device corresponding to the integrated operation amount as machine learning data when the second operation amount is invalidated.

The present invention makes it possible to use highly reliable feedback control while performing adaptive control based on data obtained during control.

FIG. 1 is a hardware configuration diagram of a control system. FIG. 2 is a general flow chart of the operation of the system. FIG. 3 is a block diagram of the basic system. FIG. 4 is a detailed flow chart of the operation of the basic system. FIG. 5 is a detailed flowchart relating to the initial learning. FIG. 6 is a detailed flow chart of the operation of the expansion system. FIG. 7 is a block diagram of an extended system. FIG. 8 is a detailed flowchart (part 1) of the control process in the extended system. FIG. 9 is a detailed flowchart (part 2) of the control process in the extended system. FIG. 10 is an explanatory diagram relating to the second manipulated variable condition. FIG. 11 is a block diagram showing the basic configuration of the feedback system. FIG. 12 is an explanatory diagram of a learning tree.

One embodiment of the present invention will now be described in detail with reference to the accompanying drawings.

1. First embodiment
<1.1 Configuration>
FIG. 1 is a hardware configuration diagram of a control system including a control device 100 and a control mechanism 12. As shown in FIG.

As is clear from the figure, the control device 100 comprises a control unit 1, a memory unit 2, an I/O unit 3, an input unit 4, a display unit 5, and a communication unit 6, which are connected to each other via a bus. The control device 100 is also connected to an operation unit 121 and a detection unit 122 that constitute the control mechanism 12, and is configured to be able to control a control target (not shown).

The control unit 1 is an information processing unit such as a CPU, and reads and executes various programs stored in the memory unit 2. The memory unit 2 is a volatile or non-volatile storage device such as a ROM, RAM, hard disk, or flash memory, and stores various data described below, including data that is the subject of machine learning. The I/O unit 3 is an interface that performs input and output with an external device. The input unit 4 processes signals input via a keyboard, touch panel, button, or the like. The display unit 5 is connected to a display or the like to perform display control and provide a GUI to the user via the display or the like. The communication unit 6 is a communication unit that communicates with external devices via a wired or wireless connection.

The operation unit 121 affects the controlled object based on a predetermined operation amount, and is composed of, for example, an actuator. The detection unit 122 detects the state of the controlled object, and is composed of, for example, a sensor.

The hardware configuration is not limited to the configuration according to this embodiment, and the configuration and functions may be distributed or integrated. For example, processing may be performed in a distributed manner using multiple control devices 100, or a large-capacity storage device may be provided externally and connected to the control device 100. Processing may also be performed by forming a computer network via the Internet, etc.

Furthermore, the processing according to this embodiment may be implemented as so-called hardware using semiconductor circuits (ICs, etc.) such as FPGAs.

1.2 Operation
Next, the operation of the control device 100 will be described with reference to FIGS.

Figure 2 is a general flowchart of the operation of the control device 100.

As is clear from the figure, when processing starts, a process is performed to set each gain (i.e., P (proportional) gain, I (integral) gain, and D (differential) gain) to be set in the first PID controller 11 of the basic system 10 described below (S1).

3 is a block diagram of the basic system 10. As is clear from the figure, the basic system 10 is composed of a first PID controller 11, a control mechanism 12 provided in the rear stage of the first PID controller 11 and equipped with an operation unit 121 and a detection unit 122, and a data logger 13 for recording the operation amount _u0 output from the first PID controller and the output value y output from the detection unit 122 of the control mechanism 12. The operation is substantially the same as that of the feedback system 200 shown in FIG. 11, but is different in that the data logger 13 records the operation amount _u0 output from the first PID controller and the output value y output from the detection unit 122 of the control mechanism 12.

The user adjusts each gain of the first PID controller 11 by a known method, for example by operating the basic system 10 or performing a simulation, and inputs and sets the final gain via the input unit 4, etc. These input gains are stored in the memory unit 2.

Returning to FIG. 2, when the gain setting process (S1) is completed, a process is performed in which the basic system 10 is actually operated using the gain, that is, a process is performed in which data for machine learning is acquired and stored (S3).

Figure 4 is a detailed flowchart of the operation of the basic system 10. As is clear from the figure, when processing begins, a process is performed in which a predetermined integer value t corresponding to a time step is initialized (for example, set to 1) (S31). When initialization is complete, a predetermined target value r(t) and the output value y(t-1) of the previous time step (t-1) are read out, the deviation (r(t)-y(t-1)) is calculated, and this deviation is input to the first controller 11 (S32).

When the deviation is input, the first controller 11 calculates the manipulated variable u(t) based on the set gain (S33). This manipulated variable u(t) is provided to the operation unit 121 of the control mechanism 12, which performs a predetermined control action on the controlled object. After that, the current (t) output value y(t) is detected via the detection unit 122 of the control mechanism 12 (S34).

When the above series of processes is completed, the output value y(t-1) of the previous time step, the manipulated variable u(t), and the output value y(t) at the current time (t) are stored in the memory unit 2 via the data logger 13 (S36). After that, the value of t is incremented by 1 (S38), and the series of processes (S32 to S38) are performed again.

That is, while controlling the controlled object, a process is continuously performed in which the output value y(t-1) of the previous time step, the manipulated variable u(t), and the output value y(t) at the current time are stored in the storage unit 2 via the data logger 13. This allows the desired amount of machine learning data to be accumulated in order to generate a trained model used in the prediction processing unit 35, which will be described later.

Returning to FIG. 2, once data acquisition and storage processing is completed based on the operation of the basic system 10 (S3), processing is performed to perform initial learning based on the acquired data (S5).

Figure 5 is a detailed flowchart of the initial learning. In this embodiment, the machine learning technique used is the one using the tree structure described above and shown in Figure 12.

As is clear from the figure, when processing starts, a parameter file related to learning, including the structure of the learning tree (number of levels, number of dimensions, number of divisions, etc.) and various initial parameters, is read from the storage unit 2. Then, a process is performed to initialize a predetermined integer value t (for example, to 1) (S52).

After this initialization, the tth input data, i.e., the output value y(t-1) of the previous time step and the operation amount u(t) are read and input to the learning tree (S53). After that, the input is classified according to a predetermined branching condition, and multiple nodes from the root node to the leaf nodes are identified and stored in association with each node (S54).

After that, in each node, the arithmetic mean value based on the output value y is updated by calculating the arithmetic mean value including the new output value y(t) and storing it in association with the node (S56).

Then, it is determined whether the value of t matches a predetermined maximum value (t_max), and if the value of t is not yet the maximum value (S57 NO), the value of t is incremented by 1 and the above-mentioned learning process (S53 to S56) is repeated again. On the other hand, if the value of t is the predetermined maximum value (S57 YES), the process ends.

In other words, this generates a trained model that predicts the output value y(t) based on the output value y(t-1) of the previous time step and the operation amount u(t) at the current time.

Returning to FIG. 2, once the initial learning process is complete, the extension system 30, which is an extension of the basic system 10 and will be described later, is then operated (S7).

Figure 6 is a detailed flowchart of the operation of the extended system 30. As is clear from the figure, when processing starts, control processing based on the extended system 30 is performed (S71).

Figure 7 is a block diagram of the extended system 30. As is clear from the figure, the extended system 30 includes a second feedback loop and a learning processing unit 34 in addition to the configuration of the basic system 10 with a first feedback loop. The second feedback loop is made up of a prediction processing unit 35 with a learned model, a second controller 37 provided downstream of the prediction processing unit 35, and an invalidation processing unit 38 and a judgment unit 39 provided downstream of the prediction processing unit 35.

The prediction processing unit 35 includes a learned model that generates a predicted output value y _hat (t) based on the output value y(t-1) of the previous time step and the first operation amount u ₁ (t) of the current time. The second controller 37 generates a second operation amount u 2 (t) based on the deviation (r(t)-y _hat (t)) between the target value r(t) and the predicted output value y _hat (t). The determination unit 39 performs a predetermined condition determination on the second operation amount u ₂ (t) and provides the determination result to the invalidation processing unit 38. Depending on the determination result provided from the determination unit 39, the invalidation processing unit 38 invalidates the second operation amount u ₂ (t) (for example, sets the second operation amount _{u 2 (} t) to 0) or provides it as it is.

The learning processing unit 34 also reads the data stored in the memory unit 2 through the data logger 53, performs learning processing under specified conditions, and provides the updated learned model to the prediction processing unit 35.

Figures 8 and 9 are detailed flowcharts of the control processing in the expansion system 30.

8, when the process starts, a process is performed to initialize flags used in the process described later (S711). Next, a process is performed to input the deviation (r(t)-y(t-1)) between the output value y(t-1) of the previous time step and the target value r(t) to the first controller 31 (S712). The first controller 31 performs a process to calculate a first manipulated variable u ₁ (t) based on the input and a set gain (S713).

Thereafter, a process is performed in which the first manipulated variable u ₁ (t) and the output value y(t-1) of the previous time step are input to the prediction processing unit 35 (S714). The prediction processing unit 35 calculates a predicted output y _hat (t) by inputting the first manipulated variable u ₁ (t) and the output value y(t-1) of the previous time step to the learned model (S715). After this calculation, a process is performed in which the deviation (r(t)-y _hat (t)) between the predicted output y _hat (t) and the target value r(t) is input to the second controller 37 (S716). The second controller 37 calculates a second manipulated variable u ₂ (t) based on the deviation between the predicted output y _hat (t) and the target value r(t) (S717).

Continuing with FIG. 9, when the second operation amount u ₂ (t) is calculated, the determination unit 39 performs a process of determining whether or not the second operation amount u ₂ (t) satisfies a predetermined condition (S719).

10 is an explanatory diagram regarding an outline of the predetermined condition for the second manipulated variable _u2 (t). As is clear from the figure, the predetermined condition is whether or not the second manipulated variable _u2 (t) is in a range between a predetermined threshold value _UL or more and a predetermined threshold value _UH or less (the range indicated by R in the figure).

If it is not within this range (R) (NO in S719), that is, if the second operation amount _u2 (t) is smaller than the predetermined threshold value _UL or larger than the predetermined threshold value _UH , the determination unit 39 provides the invalidation processing unit 38 with a determination signal indicating that the second operation amount _u2 (t) is not within the predetermined range, and the invalidation processing unit 38 performs processing to invalidate the second operation amount _u2 (t) (S720). After performing this invalidation processing, processing is performed to set a flag indicating that invalidation has been performed to ON (S721).

On the other hand, if the second operating variable _u2 (t) is within the above range (R) (S719 YES), the judgment unit 39 provides a judgment signal indicating that the second operating variable _u2 (t) is within the specified range to the invalidation processing unit 38, and the invalidation processing unit 38 provides the second operating variable _u2 (t) as is to the downstream output stage of the first controller 13 in the first feedback loop (S722).

Thereafter, in the output stage of the first controller 13 in the first feedback loop, the first manipulated variable _u1 (t) and the second manipulated variable _u2 (t) are added together to calculate the manipulated variable u(t) (S723). This manipulated variable u(t) is input to the manipulation unit 121 of the control mechanism 32, and the resulting output value y(t) is detected through the detection unit 122 (S724).

After this detection process, the output value y(t-1), the manipulated variable u(t), the output value y(t) and the flag signal of the previous time step are stored (S725), and the process corresponding to one cycle of the control process in the expansion system 30 is completed.

6, when one cycle of the control process in the expansion system 30 is completed, a process is performed to determine the state of the stored flag (S73). If it is determined that the flag is OFF (S73 NO), the process in the next time step of the expansion system 30 is performed again (S71). On the other hand, if it is determined that the flag is ON (S73 YES), that is, if the invalidation process of the second manipulated variable _u2 (t) has been performed, a learning process is performed (S75).

The contents of the learning process (S75) are substantially the same as those shown in FIG. 5, so a detailed explanation is omitted here. After this learning process, processing at the next time step of the extended system 30 is performed again (S71).

This configuration makes it possible to use feedback control, a highly reliable control technique that has been used for many years, while also using machine learning technology to perform adaptive control based on data obtained during control.

Moreover, according to this configuration, when the second manipulated variable _u2 (t) satisfies a predetermined condition, the second manipulated variable _u2 (t) is invalidated and only the control based on the first manipulated variable _u1 (t) is performed, so that highly reliable control can be performed. Moreover, since the data for the period is provided as data for machine learning, it is possible to expect improvement in control accuracy in the future.

2. Modified Examples
The above-described embodiment is an exemplary embodiment, and the present invention can be modified in various ways.

In the above embodiment, a PID controller is used as an example of a controller, but the present invention is not limited to this configuration. Therefore, other controllers having similar functions may be used, or control using only some of the gains, such as P control, PI control, or PD control, may be used.

In the above embodiment, the state of the flag is checked for each time step, and the learning process is performed in real time each time (online learning), but the present invention is not limited to such a configuration. Therefore, for example, learning may be performed in batches (batch learning, mini-batch learning) after waiting for a certain amount of data to be learned to accumulate.

In the above embodiment, when the flag is ON (S721), data relating to the previous step is learned, but the present invention is not limited to such a configuration. Therefore, for example, learning (S75) may be performed using data from one or more steps leading up to the step in question. Such learning can be particularly effective when there is continuity in the learning subject.

In the above embodiment, when the second operation amount u ₂ (t) falls outside a predetermined range (the area indicated by "R" in FIG. 10) (NO in S719), the invalidation process (S720) is performed, and therefore, when the second operation amount u 2 (t) falls outside the area (R), the flag is set to ON and learning is performed (S75). However, the present invention is not limited to such a configuration. Therefore, for example, regardless of whether the second operation amount u ₂ (t) is within the predetermined range (R) or not, learning (S75) may be performed when the second operation amount u ₂ (t) is 0 or close to it (0±ε range) (ε is a small value). In this case, the small value ε may be arbitrarily set by the user.

In the above embodiment, a machine learning model based on a tree structure model was used, but the present invention is not limited to such a configuration. Therefore, for example, other machine learning models such as neural networks and support vector machines may be used.

The present invention can be used in various industries that use control devices.

REFERENCE SIGNS LIST 1 Control unit 2 Memory unit 3 I/O unit 4 Input unit 5 Display unit 6 Communication unit 10 Basic system 11 First PID controller 12 Control mechanism 100 Control device 121 Operation unit 122 Detection unit 13 Data logger 30 Expansion system 31 First controller 32 Control mechanism 33 Data logger 34 Learning processing unit 35 Prediction processing unit 37 Second controller 38 Invalidation processing unit 39 Determination unit 200 Feedback system 201 Controller 202 Control mechanism

Claims (8)

A control device for performing feedback control on a predetermined device,
a first controller that generates a first manipulated variable for the device based on an output and a target value fed back from the device;
A predicted output generating unit including a trained model that has been machine-learned to generate a predicted output from the device based on an output fed back from the device and the first manipulated variable;
a second controller that receives the deviation between the predicted output and the target value as an input and generates a second operation input for the device;
an integrated operation amount generating unit that adds the first operation amount and the second operation amount to generate an integrated operation amount that is an operation amount for the device;
A determination unit that determines whether the second manipulated variable is within a predetermined value range ;
an invalidation processing unit that does not provide the second operation amount to the integrated operation amount generation unit when the determination unit determines that the second operation amount is not within the value range, thereby setting the first operation amount as the integrated operation amount;
a storage unit that stores the first manipulated variable, the output fed back from the device, and the output from the device corresponding to the integrated manipulated variable as machine learning data when it is determined that the second manipulated variable is not within the value range ;
The control device, wherein the first controller and the second controller each perform any one of P control, PI control, PD control, and PID control. The control device further comprises:
The control device according to claim 1 , further comprising a learning processing unit that performs a learning process based on the machine learning data and updates the trained model. The storage unit further includes:
The control device according to claim 1 , wherein when the second operation amount becomes 0 , the first operation amount, the output fed back from the device, and the output from the device corresponding to the integrated operation amount are stored as data for machine learning. The storage unit further includes:
The control device according to claim 1, further storing as machine learning data the first operation amount relating to one or more time steps temporally prior to the reference time step, the output fed back from the device, and the output from the device corresponding to the integrated operation amount, in addition to the first operation amount relating to a reference time step when it is determined that the second operation amount is not within the value range, the output fed back from the device, and the output from the device corresponding to the integrated operation amount. The control device according to claim 1, wherein the trained model is obtained by performing machine learning using a training model having a tree structure constructed by hierarchically arranging a plurality of nodes each corresponding to a hierarchically divided state space. A control method in a control device for performing feedback control on a predetermined device, comprising:
The control device includes:
a first controller that generates a first manipulated variable for the device based on an output and a target value fed back from the device;
A predicted output generating unit including a trained model that has been machine-learned to generate a predicted output from the device based on an output fed back from the device and the first manipulated variable;
a second controller that uses the deviation between the predicted output and the target value as an input to generate a second operation amount for the device;
an integrated operation amount generating step of adding the first operation amount and the second operation amount to generate an integrated operation amount that is an operation amount for the device;
a determination step of determining whether the second manipulated variable is within a predetermined value range ;
an invalidation processing step of not adding the second operation amount to the first operation amount in the integrated operation amount generation step when it is determined in the determination step that the second operation amount is not within the value range, thereby setting the first operation amount as the integrated operation amount;
and a storage step of storing, when it is determined that the second manipulated variable is not within the value range , the first manipulated variable, the output fed back from the device, and the output from the device corresponding to the integrated manipulated variable as machine learning data;
A control method, wherein the first controller and the second controller each perform any one of P control, PI control, PD control, and PID control. A control program for a control device for performing feedback control on a predetermined device,
The control device includes:
a first controller that generates a first manipulated variable for the device based on an output and a target value fed back from the device;
A predicted output generating unit including a trained model that has been machine-learned to generate a predicted output from the device based on an output fed back from the device and the first manipulated variable;
a second controller that uses the deviation between the predicted output and the target value as an input to generate a second operation amount for the device;
an integrated operation amount generating step of adding the first operation amount and the second operation amount to generate an integrated operation amount that is an operation amount for the device;
a determination step of determining whether the second manipulated variable is within a predetermined value range ;
an invalidation processing step of not adding the second operation amount to the first operation amount in the integrated operation amount generation step when it is determined in the determination step that the second operation amount is not within the value range, thereby setting the first operation amount as the integrated operation amount;
and a storage step of storing, when it is determined that the second manipulated variable is not within the value range , the first manipulated variable, the output fed back from the device, and the output from the device corresponding to the integrated manipulated variable as machine learning data;
The control program, wherein the first controller and the second controller each perform any one of P control, PI control, PD control, and PID control. A control system for performing feedback control on a predetermined device, comprising:
a first controller that generates a first manipulated variable for the device based on an output and a target value fed back from the device;
A predicted output generating unit including a trained model that has been machine-learned to generate a predicted output from the device based on an output fed back from the device and the first manipulated variable;
a second controller that receives the deviation between the predicted output and the target value as an input and generates a second operation input for the device;
an integrated operation amount generating unit that adds the first operation amount and the second operation amount to generate an integrated operation amount that is an operation amount for the device;
A determination unit that determines whether the second manipulated variable is within a predetermined value range ;
an invalidation processing unit that does not provide the second operation amount to the integrated operation amount generation unit when the determination unit determines that the second operation amount is not within the value range, thereby setting the first operation amount as the integrated operation amount;
a storage unit that stores the first manipulated variable, the output fed back from the device, and the output from the device corresponding to the integrated manipulated variable as machine learning data when it is determined that the second manipulated variable is not within the value range ;
A control system, wherein the first controller and the second controller each perform any one of P control, PI control, PD control, and PID control.

Images