JP7724621B2 - PID control parameter adjustment method, PID control device, and air conditioner equipped with the same - Google Patents

Description

The present invention relates to a PID control parameter adjustment method, a PID control device, and an air conditioner equipped with the same.

PID control is widely used for machine control and temperature control. When initially installing equipment, autotuning is typically used to determine PID control parameters. However, when using PID control for temperature and humidity in test rooms or inspection rooms that require precise temperature and humidity management, the PID control parameters obtained through autotuning may not provide sufficient accuracy. In such cases, to adjust the PID control parameters, workers would adjust them by trial and error while checking the output waveform. Currently, it takes time to determine the PID control parameters that are appropriate for the equipment being controlled.

There have been attempts to use reinforcement learning, a type of machine learning, to set PID control parameters. Patent Document 1 discloses a technology for determining PID control parameters using reinforcement learning.

Japanese Patent Application Laid-Open No. 2019-197315

When using reinforcement learning to determine PID control parameters, there are many different specific reinforcement learning methods, and no single method can be determined. The reinforcement learning algorithm merely provides a framework, and algorithms for converting environment, behavior, and reward data, observing environmental conditions, calculating rewards, etc. must be set separately depending on the problem being applied.

Reinforcement learning ends when a predetermined condition is met, such as completing a set number of runs or obtaining a reward (good result) greater than or equal to a set value. When large load fluctuations, environmental changes, or equipment deterioration occur, the learning results obtained may not provide sufficient control accuracy. In this case, the PID control parameters must be adjusted again. This adjustment also takes time, just like when the equipment was first installed.
Temperature and humidity control requires particular precision. When load fluctuations, environmental changes, or equipment deterioration progresses, the obtained learning results can lead to situations where the control precision does not meet the required level, making it difficult to apply reinforcement learning to determine PID control parameters.

The object of the present invention is to provide a method for adjusting PID control parameters for a PID control device, in particular a method for adjusting PID control parameters that can automatically respond to load fluctuations and deterioration over time.

The PID control parameter adjustment method disclosed herein includes the steps of PID controlling a control target device to a control target value using predetermined PID control parameters for a first predetermined time period; calculating a control evaluation value based on output data from the control target device for the first predetermined time period; storing the control conditions for PID control for the first predetermined time period, the PID control parameters, and the control evaluation value as learning data; updating the PID control parameters of the PID control device; and re-executing the step of PID controlling the control target device using the updated PID control parameters.

The PID control parameter adjustment method disclosed herein makes it possible to realize a PID control parameter adjustment method that can automatically respond to load fluctuations and deterioration over time.

FIG. 1 is a diagram illustrating a configuration of a PID control device according to an embodiment. FIG. 1 is a block diagram of a general PID control. FIG. 2 is a diagram illustrating a configuration of a reinforcement learning unit according to an embodiment. 10 is a flowchart illustrating the operation of reinforcement learning of the PID control device according to the embodiment. 4 is a flowchart illustrating the operation of the PID control execution process in the embodiment; 10 is a flowchart illustrating the operation of the evaluation value calculation process in the embodiment. 10 is a flowchart illustrating an operation of a PID control parameter update process in the embodiment;

First, we will explain reinforcement learning in this disclosure and define terms related to reinforcement learning.

(Reinforcement learning)
Reinforcement learning is a method of finding a policy that maximizes the reward obtained by an agent placed in a certain environment, taking action in the environment. The agent takes action in the environment, the environment updates the state and evaluates the action, and notifies the agent of the state and reward. This process is repeated chronologically, and the action value function and policy are optimized to maximize the expected value of the total reward obtained.

(Definitions of state, action, and reward in reinforcement learning in this disclosure)
In the reinforcement learning of PID control parameters in the present disclosure, the state is the control conditions of PID control, specifically, the control target value of PID control and surrounding environmental data. The action is the update of the PID control parameters through reinforcement learning. The reward is a value (hereinafter referred to as a control evaluation value) obtained by calculating the convergence, responsiveness, stability, etc. of the output data of the controlled device after executing PID control for a predetermined period.

As mentioned in the issue section, even for equipment that has completed reinforcement learning, if there are large load fluctuations, changes in the environment, or deterioration of the equipment being controlled over time, and control accuracy can no longer be achieved, reinforcement learning will be required again. Facility operators must confirm whether reinforcement learning is necessary again. This confirmation process requires detailed analysis of the output waveform. Furthermore, abnormalities in the control results may not be noticed, making management cumbersome.

The applicants therefore conducted extensive research into PID control parameter adjustment methods that could address the above-mentioned issues, and as a result, arrived at the PID control parameter adjustment method disclosed herein. The PID control parameter adjustment method disclosed herein has the following two features. The first feature is that reinforcement learning is constantly performed in the PID control of the controlled device. The second feature is that it incorporates a method of storing output data obtained by executing PID control for a predetermined period at regular intervals, and calculating convergence, responsiveness, and stability, including the process by which that output data changes.

The first feature makes it possible to automatically correct the PID control parameters even when there are large load fluctuations, changes in the outside air, or deterioration of the system over time.
The second feature makes it possible to set PID control parameters that have high convergence, good responsiveness, and high stability in PID control that requires precision.

Embodiments of the present invention are described in detail below. In the following description, specific shapes, materials, directions, numerical values, etc. are examples intended to facilitate understanding of the present disclosure and can be modified as appropriate to suit the application, purpose, specifications, etc.

Figure 1 shows the configuration of the PID control device 1 of this embodiment. The PID control device 1 has a PID control unit 10, a target setting input unit 30, an environmental data acquisition unit 40, a control data measurement unit 50, and a reinforcement learning unit 60. The components specific to reinforcement learning are the control data measurement unit 50 and the reinforcement learning unit 60.

The PID control unit 10 performs PID control based on preset PID control parameters so that the output of the controlled device 20 matches a target value (control target value). The control target value for the controlled device 20 is obtained from the target value setting input unit 30.

2 is a block diagram of the PID control unit 10 and the controlled device 20. _KP , _KI , and _KD are proportional gain, integral gain, and differential gain, respectively. The PID control unit 10 of this embodiment is configured to change the value of each gain upon receiving PID control parameters from a reinforcement learning unit 60, which will be described later.

The controlled device 20 is a device on which PID control is performed. Specifically, it may be, but is not limited to, a constant temperature and humidity room. The output data of the controlled device 20 is fed back to the PID control unit 10, and PID control is performed based on the deviation from the control target value.

The target setting input unit 30 acquires the control target value of the controlled device 20. The control target value acquired by the target value setting input unit 30 is sent to the PID control unit 10 and becomes the control target value for PID control. The control target value is also sent to the reinforcement learning unit 60. In the reinforcement learning unit 60, the control target value is used as part of the state value in reinforcement learning.

The environmental data acquisition unit 40 acquires environmental data of the controlled device 20, such as the outside air temperature, room temperature, and humidity. The specific configuration for acquiring environmental data is not limited, but it can be configured to acquire data from various sensors. The environmental data acquired by the environmental data acquisition unit 40 is used as part of the state value in reinforcement learning.

The control data measurement unit 50 sequentially measures the output data of the controlled device 20 executing PID control and sends it to the reinforcement learning unit 60. The output data of the control data measurement unit 50 may also be configured to be used as the output data fed back from the controlled device 20 to the PID control unit 10.

The reinforcement learning unit 60 stores the control target value for PID control acquired from the target value setting input unit 30 and the environmental data of the controlled device 20 acquired from the environmental data acquisition unit 40 as part of the state values in reinforcement learning.

The reinforcement learning unit 60 further acquires and stores output data from the control data measurement unit 50 for the controlled device 20 during PID control execution. After PID control has been executed for a predetermined period of time, the acquired output data is used to calculate the control evaluation value.

After executing PID control for a predetermined time, the reinforcement learning unit 60 calculates a control evaluation value as a reward in reinforcement learning. The control evaluation value may be calculated based on the relationship between the control target value, such as the difference between the output data and the control target value. The calculation of the control evaluation value will be explained in detail later.

The reinforcement learning unit 60 updates the PID control parameters based on the control evaluation value and transmits them to the PID control unit 10. The PID control unit 10 performs PID control based on the updated PID control parameters and the control target value of the target setting input 30.

The reinforcement learning unit 60 also has a function of saving the learning results. In this embodiment, the state in reinforcement learning is the control target value and the environmental data value. The action in reinforcement learning is the update of the PID control parameters. Specifically, this is the new update of the PID control gains _KP , _KI , and _KD . The reward in reinforcement learning is the control evaluation value. When updating the PID control parameters, the already saved learning results are referenced.

Figure 3 shows the configuration of the reinforcement learning unit 60. The reinforcement learning unit 60 has at least an input unit 610, a control unit 620, a memory unit 630, a calculation unit 640, and an output unit 650.

The input unit 610 has the function of inputting data necessary to perform PID control and reinforcement learning. The input unit 610 has a target value input unit 611, an environmental data input unit 612, and an output data measurement unit 613. The target value input unit 611 acquires control target values from the target setting input unit 30. The environmental data input unit 612 acquires environmental data from the environmental data acquisition unit 40. In the case of controlling a constant temperature and humidity room, environmental data corresponds to the outside air temperature, room temperature, humidity, etc. The output data measurement unit 613 acquires output data of the controlled device 20 that is executing PID control at predetermined intervals. The output data measurement unit 613 acquires output data of the controlled device 20 from the control data measurement unit 50.

The control unit 620 has a control time measurement unit 621. The control time measurement unit 621 measures a first predetermined time T1 during which the PID control unit 10 executes PID control. The first predetermined time T1 corresponds to the time for one reinforcement learning session. The control time measurement unit 621 may also be configured to measure a second predetermined time T2 during which data output from the controlled device 20 is saved. The second predetermined time T2 may also be configured to be measured by the PID control unit 10.

The memory unit 630 has a measurement data memory unit 631 and a learning data memory unit 632. The measurement data memory unit 631 receives and stores the output data of the control target device 20 acquired by the output data measurement unit 613. For example, it stores the output data of the control target device 20 for each second predetermined time T2. The learning data memory unit 632 stores a control evaluation value calculated by the calculation unit 640 (described below) based on the data from the measurement data memory unit 631.

The calculation unit 640 has a data calculation unit 641 and a parameter setting unit 642. The data calculation unit 641 calculates a control evaluation value based on the output data of the control target device 20 for each second predetermined time T2 stored in the measurement data storage unit 631. The control evaluation value is stored in the learning data storage unit 632 as part of the learning data.

The parameter setting unit 642 of the calculation unit 640 updates the values of the PID control parameters to be used in the next PID control. In this embodiment, the values of the PID control parameters are updated by extracting from the learning data an action (update of the PID control parameters) that results in a control evaluation value of a predetermined value (usually the maximum value) for the control target value obtained from the target value input unit 611 and the environmental data value obtained from the environmental data input unit 612.

The output unit 650 has a parameter output unit 651. The parameter output unit 651 transmits the PID control parameters determined by the parameter setting unit 642 of the calculation unit 640 to the PID control unit 10. The PID control unit 10 changes the values of each gain based on the transmitted PID control parameters.

Next, we will explain the details of reinforcement learning in this embodiment with reference to the flowcharts in Figures 4 to 7. Figure 4 is a flowchart that shows an overview of reinforcement learning.

Step S01: First, determine the initial values of the PID control parameters. The initial values of the PID control parameters may be determined by autotuning, or may be set manually. If the PID control parameters have been set in advance, the initial values may be used, and step S01 may not be executed. Next, proceed to step S02.

Step S02: PID control of the controlled device 20 is performed using the current PID control parameters, and the output data of the controlled device 20 is stored. The control time is the first predetermined time T1. Details of the PID control execution process are explained in Figure 5.

Step S03: Calculate a control evaluation value based on the stored output data. The control evaluation value reflects a measure for determining the quality of the PID control parameters. Specifically, the control evaluation value is calculated from the perspective of the convergence, responsiveness, and stability of the control results. The calculated control evaluation value is stored as part of the learning data. Details of the control evaluation value calculation and learning data storage process are explained in Figure 6.

Step S04: The PID control parameters are updated and the updated PID control parameters are sent to the PID control unit 10. A specific method for updating the PID control parameters is described in Figure 7. Next, return to step S02 and execute PID control for a predetermined time using the updated PID control parameters.

The reinforcement learning flowchart shown in Figure 4 differs from general reinforcement learning in that there is no termination condition for reinforcement learning. According to the reinforcement learning policy described later, if the control evaluation value of PID control deteriorates in relation to the control target value and environmental data, reinforcement learning is configured to proceed again so that new optimal PID control parameters are found.

Note that learning results are constantly being accumulated in the memory unit 630 of the reinforcement learning unit 60. Therefore, the memory unit 630 must be configured to take into account the fact that learning results are constantly being accumulated. For example, the memory unit 630 may be configured as a hard disk drive. Alternatively, it may be configured on a server connected to a network. As reinforcement learning progresses, the increase in learning result data will slow down unless there is a large change in the target value or sudden deterioration of the equipment.

Next, we will explain the details of each process in steps S02 to S04 in the flowchart in Figure 4.

(PID control execution process)
5 is a flowchart of the PID control execution process of step S02 in FIG. 4. The PID control execution process executes PID control of the controlled device 20 using set PID control parameters for a first predetermined time. Output data of the controlled device 20 is acquired during the first predetermined time T1, and the output data is saved at intervals of a second predetermined time T2. The saved output data is used in the control evaluation value calculation process in the next step.

Step S11: Set the elapsed time of PID control to 0.

Step S12: Execute PID control using the current PID control parameters.

Step S13: Obtain output data from the controlled device 20. In step S13, the output data is not necessarily saved.

Step S14: If the second predetermined time T2 has elapsed, proceed to the next step S15. If the second predetermined time T2 has not elapsed, return to step S12.

Step S15: The output data acquired in step S13 is stored. The output data is stored in the measurement data storage unit 631. This allows the output data for each second predetermined time T2 to be stored.

Step S16: If the first predetermined time T1 has elapsed, end the PID control execution process. If the first predetermined time T1 has not elapsed, return to step S12.

The above is the processing content of the PID control execution process. Here, the first predetermined time T1 is the time for executing PID control and is set to, for example, one hour. The second predetermined time T2 is shorter than the first predetermined time T1 and is the interval for acquiring output data. For example, the second predetermined time T2 is set to one second. The first predetermined time T1 and second predetermined time T2 shown here are examples. Depending on the physical size of the controlled device 20, etc., and taking into account the time it takes for the output to stabilize, the first predetermined time T1 and second predetermined time T2 may be changed as appropriate.

(Control evaluation value calculation and learning data storage processing)
6 is a flowchart of the control evaluation value calculation and learning data storage process. The control evaluation value is calculated based on the PID control output data for the first predetermined time T1 (the output data saved every second predetermined time T2). The calculation for the control evaluation value reflects the quality of the current PID control parameters, and a high score is assigned when the PID control result precisely matches the output target value.

Step S21: Calculate a control evaluation value based on the output data of PID control for the first predetermined time T1. Calculations are performed so that the more accurately the output matches the target value, the higher the control evaluation value. Specifically, in the case of control of a constant temperature and humidity chamber, the smaller the deviation between the target temperature and the measured output temperature, the better the control. Therefore, in this case, calculations are performed so that the smaller the deviation, the higher the control evaluation value calculated.

Step S22: Once calculation of the control evaluation value has been completed, the learning data is saved. The control conditions for PID control execution (control target value, environmental data), PID control parameters, and control evaluation value are paired and stored as learning data. This stores as learning data how to update the PID control parameters for the control conditions to increase the control evaluation value (= obtain the desired good control results).

Step S23: The learning policy for updating the PID control parameters in the next step is determined based on the control evaluation value. Details will be described later.

The control evaluation value is calculated based on measurement data to reflect, but is not limited to, (1) convergence, (2) responsiveness, and (3) stability of the PID control results.

(1) Convergence may be measured by the difference between the output data value and the control target value for each second predetermined time T2, or by the cumulative value thereof. The smaller the difference between the output data value and the control target value, the better the control accuracy and the better the convergence. Furthermore, the cumulative value of the difference between the output data value and the control target value over time reflects how quickly the convergence occurred.

(2) Responsiveness may be measured by the range of change in the output data for each second predetermined time period T2. The range of change in the output data for each fixed time period corresponds to the rate of change in the output data, and therefore is a value that reflects responsiveness.

(3) Stability may be measured by the difference between the maximum and minimum values of the output data after a predetermined time has elapsed, or by the number of times the output data increases or decreases around the control target value, or the magnitude of the increase or decrease. These values reflect whether hunting is occurring, and if so, the extent of hunting, and therefore reflect stability.

The performance of convergence, responsiveness, and stability are intricately related and cannot be evaluated completely separately. Therefore, the control evaluation value may be calculated, for example, by calculating multiple control evaluation values, multiplying them by a coefficient, and adding them together. Alternatively, multiple control evaluation values may be used.

The method for calculating the control evaluation value explained above is an example. It is important to calculate the control evaluation value based on output data within a specified period of time. The specific calculations performed can be adjusted to suit the control target being applied.

Next, we will explain how to determine the learning policy. The learning policy determines how to update the PID control parameters (= behavior in reinforcement learning), which will be described later. For example, it is preferable to change the reinforcement learning policy depending on the progress of learning.

The learning policy is as follows:
・Policy A is after sufficient learning has progressed, and optimal PID control parameters are selected. ・Policy B is likely to be affected by aging, etc., and new PID control parameters are revised using reinforcement learning. ・Policy C is in the early stages of learning, and data is repeatedly acquired and learning data is accumulated.

For example, it is possible to determine which of these strategies to select based on a control evaluation value obtained by executing PID control for a predetermined period of time.
If the same control evaluation value is obtained for the same control conditions and the same PID control parameters, strategy A is selected.
When a lower control evaluation value is obtained than the previous time for the same control conditions and the same PID control parameters, strategy B is selected. When the control evaluation value decreases significantly, strategy C may be selected. For example, when the control evaluation value decreases by 10% or more, strategy C is selected.
If the amount of data obtained by learning is small, policy C is selected unconditionally.

The learning policy is determined by the control evaluation value corresponding to the state (control condition) and action (update of PID control parameters) in reinforcement learning.

(PID control parameter update process)
7 is a flowchart of the PID control parameter update process, in which the PID control parameters are updated for the next PID control execution.

Step S31: Check the current reinforcement learning learning policy. Change the update method depending on the learning policy.
Steps S32 to S34: Update of PID control parameters corresponding to the current learning policy is executed (described later).
Step S35: The PID control parameters updated in the previous step are sent to the PID control unit 10.

In this embodiment, steps S32 to S34 show the case where there are three learning policies. For policy A, update A is executed (step S32). For policy B, update B is executed (step S33). For policy C, update C is executed (step S34). Three examples of update methods are shown, but the present invention is not limited to these.

Specific examples of update methods corresponding to policies A to C are given below.
Update A corresponding to policy A: Select the action (=PID control parameter) that maximizes the control evaluation value.
Update B corresponding to policy B: The value is set to a value that is randomly increased or decreased from the value of the action that maximizes the control evaluation value.
Update C corresponding to policy C: The value of the action is determined randomly.
The above are examples of update methods corresponding to the learning policy, but other update methods may also be used.

The PID control parameter adjustment method described in this embodiment is characterized by the fact that reinforcement learning is not terminated when a predetermined number of times or when a predetermined condition is reached, but is instead constantly continued. Because reinforcement learning is constantly ongoing, a control evaluation value is constantly calculated. As aging progresses in the controlled device, cases will arise in which the control evaluation value becomes lower than its previous value. This corresponds to the above-mentioned strategy B. In this case, the PID control parameters are updated to values that are randomly increased or decreased from the PID control parameter value that maximized the previous control evaluation value. Therefore, reinforcement learning of new PID control parameters progresses, and by repeating this process, optimal PID control parameters for the new state in reinforcement learning are learned. As a result, it is possible to automatically respond to cases in which PID control parameters need to be modified due to aging, etc.

Many of the components of the PID control device of this embodiment described above can be configured using computer-based hardware and software. They may be implemented using a single computer system, or by linking separate computer systems. Alternatively, a configuration that does not use a computer is also possible. How the overall system is configured is irrelevant to the essence of this disclosure.

In addition, because the PID control device of this embodiment constantly performs reinforcement learning, there is a possibility that the amount of learning data will become enormous. For this reason, the storage capacity of the storage unit 630 must be determined taking this into consideration. The storage unit 630 can be configured as a storage device of a computer system, for example, a hard disk drive. Alternatively, it may be configured to use a storage device on a cloud connected to a network. From this perspective, it is preferable that the reinforcement learning unit 60 be configured as a computer system.

(Application example)
The PID control parameter adjustment method of the present disclosure can be applied to air conditioners used in constant temperature and humidity rooms. Constant temperature and humidity rooms are widely used in various laboratories, measurement rooms, clean rooms, etc., and require highly accurate temperature and humidity control. Once installed, the equipment is used for a long period of time, so it is essential to optimize the PID control parameters to address deterioration over time. By incorporating the PID control parameter adjustment method of the present disclosure into a PID control device and controlling the air conditioner, an air conditioner with a temperature and humidity control function that can automatically address deterioration over time can be realized. Note that the PID control parameter adjustment method of the present disclosure can be applied not only to temperature and humidity control, but also to general control devices using PID control.

1 PID control device, 10 PID control unit, 20 Control target device, 30 Target setting input unit, 40 Environmental data acquisition unit, 50 Control data measurement unit, 60 Reinforcement learning unit, 610 Input unit, 611 Target value input unit, 612 Environmental data input unit, 613 Output data measurement unit, 620 Control unit, 621 Control time measurement unit, 630 Memory unit, 631 Measurement data memory unit, 632 Learning data memory unit, 640 Calculation unit, 641 Data calculation unit, 642 Parameter setting unit, 650 Output unit, 651 Parameter output unit, T1 First predetermined time, T2 Second predetermined time

Claims (8)

1. A method for adjusting PID control parameters for a PID controller, comprising:
a first step of PID-controlling a value of output data indicating at least one of temperature and humidity of a controlled device to a control target value using predetermined PID control parameters for a first predetermined time;
a second step of calculating a control evaluation value using the value of the output data of the control target device during the first predetermined time period , the control target value, and an equation indicating the relationship between the value of the output data and the control target value, after the first step is executed;
a third step of storing, after execution of the second step, the control conditions of the PID control for the first predetermined time, the PID control parameters, and the control evaluation value as learning data;
and a fourth step of updating a PID control parameter of the PID control device based on the control evaluation value after the third step is executed,
a PID control parameter adjusting method, wherein after the fourth step is performed, the first step is always performed using the updated PID control parameters; 2. The PID control parameter adjusting method according to claim 1, wherein the first step includes a step of measuring a value of the output data of the controlled device for the first predetermined time period and storing the value of the output data. The second step includes:
a difference between the value of the stored output data for each second predetermined time period that is shorter than the first predetermined time period and the control target value;
a change range of the value of the stored output data for each second predetermined time;
the difference between the maximum and minimum values of the stored output data;
3. The PID control parameter adjustment method according to claim 2, wherein the control evaluation value is calculated based on at least one of the number of times the value of the stored output data increases or decreases around the control target value, or the width of the change. The fourth step includes:
selecting an update method based on a learning policy determined by the control conditions and the control evaluation values corresponding to the PID control parameters;
updating the PID control parameters based on the selected update method;
4. The PID control parameter adjusting method according to claim 1, further comprising the step of transmitting the updated PID control parameters to the PID control device. A PID controller,
a PID control unit that executes a first process of PID-controlling a value of output data indicating at least one of temperature and humidity of the control target device to a control target value using predetermined PID control parameters for a first predetermined time;
a data calculation unit that executes, after the execution of the first process, a second process of calculating a control evaluation value using the value of the output data of the control target device during the first predetermined time, the control target value, and an equation indicating the relationship between the value of the output data and the control target value ;
a storage unit that executes a third process of storing, after the execution of the second process, the control conditions of the PID control for the first predetermined time, the PID control parameters, and the control evaluation value as learning data;
a parameter setting unit that executes a fourth process to update a PID control parameter of the PID control device based on the control evaluation value after the third process is executed,
A PID control device in which, after the fourth process is executed, the first process is always executed using the updated PID control parameters. An air conditioner having the PID control device described in claim 5. A PID control device,
The PID control device
a PID control unit that PID controls a value of output data indicating at least one of the temperature and humidity of the controlled device to a control target value;
a target setting input unit for setting the control target value of the PID control unit;
an environmental data acquisition unit that acquires environmental conditions;
a control data measurement unit that measures the value of the output data of the control target device;
a reinforcement learning unit that learns a PID control parameter after the PID control is executed,
The reinforcement learning unit
an input unit that acquires the control target value from the target setting input unit, acquires the environmental condition from the environmental data acquisition unit, and acquires the value of the output data of the control target device from the control data measurement unit;
a control unit that controls a control time of the PID control unit;
a storage unit that stores the value of the output data of the control target device acquired from the control data measurement unit;
a calculation unit having a function of calculating a control evaluation value using the output data value of the control target device stored in the storage unit, the control target value, and an equation indicating the relationship between the output data value and the control target value, and a function of updating a PID control parameter based on the control evaluation value;
an output unit that transmits the updated PID control parameters of the calculation unit to the PID control unit;
A PID control device in which the PID control is always performed after the PID control parameters are learned. An air conditioner having the PID control device described in claim 7.