Learning processing device, control device, learning processing method, control method, and recording medium

Abstract

Provided is a learning processing device, a control device, a learning processing method, a control method, and a recording medium, the learning processing device including: a learning processing section that generates, by machine learning, a control model that outputs an operation amount corresponding to a prescribed system's command value and a measured value; a generation section that generates, using the control model, control data that indicates a correspondence between a combination of the command value and the measured value and the operation amount corresponding to the combination; and a supply section that supplies the control data to a prescribed control device.

Description

Technical Field

[0001] This invention relates to a learning processing device, a control device, a learning processing method, a control method, and a recording medium. Background Technology

[0002] Patent document 1 describes a "temperature control device that uses temperature control information to perform PID control on cylinder temperature".

[0003] Patent Document 1: Japanese Patent Application Publication No. 2019-130771 Summary of the Invention

[0004] In a first aspect of the present invention, a learning processing apparatus is provided. The learning processing apparatus may include a learning processing unit that generates a control model by machine learning, outputting operational quantities corresponding to predetermined system indication values ​​and measurement values. The learning processing apparatus may include a generation unit that uses the control model to generate control data representing the correspondence between combinations of indication values ​​and measurement values ​​and operational quantities corresponding to those combinations. The learning processing apparatus may include a supply unit that supplies the control data to a predetermined control device.

[0005] The operating quantity can be the maximum or minimum operating quantity within a predefined operable range.

[0006] The control model can output the operating quantity in a way that makes the measured value reach the set target value. The generation unit can generate different control data for each target value.

[0007] It can generate control models corresponding to multiple predefined systems. The generation unit can generate different control data for each of the multiple systems.

[0008] Control data may include an operation correspondence diagram that matches the combination of indicated and measured values ​​with the corresponding operation quantities for that combination.

[0009] In a second aspect of the present invention, a control device is provided. The control device may include an input data acquisition unit that acquires predetermined system indication values ​​and measured values. The control device may also include a control data acquisition unit that acquires control data generated using a control model learned to output operational quantities corresponding to the indication and measured values, representing the correspondence between combinations of indication and measured values ​​and operational quantities output corresponding to those combinations. The control device may include a calculation unit that uses the control data to calculate the operational quantity corresponding to the combination of indication and measured values. Finally, the control device may include an output unit that outputs the operational quantity to a predetermined controlled object.

[0010] As control data, the control data acquisition unit can acquire multiple operational quantity correspondence diagrams that match the correspondence between combinations of indicated values ​​and measured values ​​and the corresponding operational quantities. The input data acquisition unit can acquire feature data related to the system. The control device may have a correspondence diagram selection unit that selects an arbitrary operational quantity correspondence diagram from multiple operational quantity correspondence diagrams based on the feature data.

[0011] The control device may include a scheduling unit that schedules indicated and measured values ​​to values ​​corresponding to the operational quantity correspondence diagram. The control device may also include a reverse scheduling unit that performs reverse scheduling on the operational quantities calculated based on the operational quantity correspondence diagram according to the system.

[0012] As an operational quantity, the output unit can output the maximum or minimum operational quantity within a predefined operable range to the controlled object.

[0013] In a third aspect of the present invention, a learning processing method is provided. The learning processing method may include the following stages: generating a control model that outputs operational quantities corresponding to predefined system indication values ​​and measurement values ​​through machine learning; generating control data representing the correspondence between combinations of indication values ​​and measurement values ​​and operational quantities corresponding to those combinations using the control model; and supplying the control data to a predefined control device.

[0014] In a fourth aspect of the present invention, a control method is provided. The control method may include the following stages: acquiring predetermined system indication values ​​and measured values; acquiring control data generated using a control model learned to output operational quantities corresponding to the indication values ​​and measured values, representing the correspondence between combinations of indication values ​​and measured values ​​and operational quantities output corresponding to those combinations; calculating operational quantities corresponding to the combinations of indication values ​​and measured values ​​using the control data; and outputting the operational quantities to a predetermined controlled object.

[0015] In a fifth aspect of the present invention, a recording medium is provided that records a learning program. The learning program can be executed by a computer, which functions as a learning processing unit that generates a control model by means of machine learning, outputting operational quantities corresponding to predefined system indication values ​​and measurement values. The learning program can also be executed by a computer, which functions as a generation unit that uses the control model to generate control data representing the correspondence between combinations of indication values ​​and measurement values ​​and the operational quantities corresponding to those combinations. Finally, the learning program can be executed by a computer, which functions as a supply unit that supplies the control data to a predefined control device.

[0016] In a sixth aspect of the present invention, a recording medium is provided, which records a control program. The control program can be executed by a computer, which functions as an input data acquisition unit, acquiring predetermined system indication values ​​and measurement values. The control program can also be executed by a computer, which functions as a control data acquisition unit, acquiring control data generated using a control model learned to output operation quantities corresponding to the indication and measurement values, representing the correspondence between combinations of indication and measurement values ​​and the operation quantities output corresponding to those combinations. The control program can also be executed by a computer, which functions as a calculation unit, using the control data to calculate the operation quantity corresponding to the combination of indication and measurement values. Finally, the control program can be executed by a computer, which functions as an output unit, outputting the operation quantity to a predetermined controlled object.

[0017] Furthermore, the above summary of the invention does not list all the essential features of the invention. Additionally, sub-components of the above-described feature set can also constitute the invention. Attached Figure Description

[0018] Figure 1AAn outline of the structure of the control device 100 is shown together with the device 300.

[0019] Figure 1B This is an example of the process by which the control device 100 controls the action of the controlled object 310.

[0020] Figure 2A This section outlines the structure of the learning processing device 200.

[0021] Figure 2B This illustrates an example of a process for performing machine learning using the learning processing device 200.

[0022] Figure 3 An example of a more specific structure of the computing unit 30.

[0023] Figure 4A An example of a diagram showing the correspondence between operational quantities.

[0024] Figure 4B This illustrates an example of a control method for the control device 100 involved in the embodiment.

[0025] Figure 4C This represents an example of a method for generating a graph that utilizes machine learning operations.

[0026] Figure 5 This represents an example of the control method involved in proportional control.

[0027] Figure 6A This illustrates one embodiment of the learning processing device 200.

[0028] Figure 6B This illustrates one embodiment of the learning processing device 200.

[0029] Figure 7 Examples of computers 2200 that can embody all or part of the various methods of the present invention are shown. Detailed Implementation

[0030] The present invention will now be described through embodiments thereof, which do not limit the invention as defined in the claims. Furthermore, not all combinations of the features described in the embodiments are essential to the solution of the invention.

[0031] Figure 1A Together with the device 300, this section outlines the structure of the control device 100. The control device 100 controls the operation of the controlled object 310 disposed on the device 300.

[0032] Equipment 300 refers to facilities, devices, etc., equipped with the controlled object 310. For example, equipment 300 can be a factory or a composite device composed of multiple instruments. As a factory, in addition to chemical and biological industrial factories, examples include factories that manage and control the wellheads and surrounding areas of gas fields and oil fields, factories that manage and control power generation from hydropower, thermal power, and nuclear power, factories that manage and control environmental resource power generation from solar energy and wind power, and factories that manage and control water supply and drainage, dams, etc.

[0033] Device 300 is provided with a control object 310. In this figure, only one control object 310 is shown as an example, but it is not limited to this. Multiple control objects 310 may also be provided for device 300.

[0034] Additionally, one or more sensors (not shown) can be installed in device 300 to measure various states (physical quantities) inside and outside device 300. Such sensors, for example, acquire operational data representing the operating state as a result of controlling the controlled object 310. For example, the operational data can represent the measured value PV (Process Variable) obtained for the controlled object 310, and as an example, it can represent the output (control quantity) of the controlled object 310, or various values ​​that change according to the output of the controlled object 310.

[0035] The controlled object 310 refers to field instruments and devices that are controlled. For example, the controlled object 310 may be a sensor instrument such as a pressure gauge, flow meter, or temperature sensor; a valve instrument such as a flow control valve or an on / off valve; or an actuator instrument such as a fan or a motor.

[0036] In this example, the control device 100 performs process control through a single input and single output, consisting of a measured value PV and an operational quantity MV. For example, the control device 100 performs process control such as temperature regulation, liquid level adjustment, or flow rate adjustment.

[0037] The control device 100 can be a PC (personal computer), tablet computer, smartphone, workstation, server computer, or general-purpose computer, or a computer system composed of multiple connected computers. Furthermore, such a computer system is also considered a computer in a broad sense. Additionally, the control device 100 can be installed within a computer using one or more virtual computer environments. Alternatively, the control device 100 can be a dedicated computer designed for AI control, or dedicated hardware implemented using dedicated circuitry. Furthermore, if the control device 100 can connect to the internet, it can be implemented through cloud computing.

[0038] The control device 100 includes an input data acquisition unit 10, a control data acquisition unit 20, a calculation unit 30, and an output unit 40. Furthermore, the above modules can be functionally separate modules and may not conform to the actual device structure. That is, in this figure, they are shown as a single module, but are not limited to being composed of a single device. Additionally, in this figure, they are shown as different modules, but are not limited to being composed of different devices.

[0039] The input data acquisition unit 10 acquires pre-defined input data. For example, as input data, the input data acquisition unit 10 acquires a pre-defined system indication value IV and a measured value PV from the device 300. The indication value IV and the measured value PV can be measured by a sensor installed in the device 300, or they can be transmitted to the input data acquisition unit 10 via a network. In addition, as input data, the input data acquisition unit 10 acquires a target value SV set by the operator or others. The input data acquisition unit 10 can supply the acquired input data to the control data acquisition unit 20 and the calculation unit 30.

[0040] The control data acquisition unit 20 acquires control data representing the correspondence between a combination of indicated value IV and measured value PV and an operational quantity MV (manipulated variable) output based on that combination. The control data acquisition unit 20 may include a storage unit for storing the acquired control data. The control data is generated using a control model learned in a manner that outputs the operational quantity MV corresponding to the indicated value IV and measured value PV.

[0041] Control data only needs to represent the correspondence between the combination of indicated values ​​(IV) and measured values ​​(PV) and the operational quantity (MV); the data format is not particularly limited. In one example, the control data includes an operational quantity mapping chart that matches the combination of indicated values ​​(IV) and measured values ​​(PV) with the operational quantity (MV). The operational quantity mapping chart will be described later. Control data can also represent the correspondence between the combination of indicated values ​​(IV) and measured values ​​(PV) and the operational quantity (MV) in tabular form.

[0042] The calculation unit 30 uses control data to calculate the operating quantity MV corresponding to the combination of the indicated value IV and the measured value PV. In one example, the calculation unit 30 calculates the maximum or minimum operating quantity MV within a predefined operating range as the operating quantity MV. The maximum operating quantity MV can be the maximum value of the positive operating quantity MV within the predefined operating range. The minimum operating quantity MV can be the minimum value of the negative operating quantity MV within the predefined operating range.

[0043] Furthermore, the calculation unit 30 can also calculate the operation quantity MV based on the target value SV. For example, the calculation unit 30 selects an operation quantity correspondence chart based on the target value SV and calculates the operation quantity MV using the selected operation quantity correspondence chart. In addition, if the difference between the target value SV and the measured value PV is less than a predetermined value, the calculation unit 30 can switch to other control methods such as PID control.

[0044] The output unit 40 outputs the operation quantity MV calculated by the calculation unit 30 to the controlled object 310. As the operation quantity MV, the output unit 40 can output the maximum or minimum operation quantity MV within a predefined operable range to the controlled object.

[0045] For example, when the control device 100 adjusts the rotational speed used to open and close the valve to control the water level in the container, the indicated value IV can be the valve opening degree, the measured value PV can be the water level in the container, and the operating amount MV can be the valve rotational speed. Similarly, when the control device 100 adjusts the current flowing in the heating wire to control the furnace temperature, the indicated value IV can be the temperature of the heating wire itself, the measured value PV can be the temperature of the entire furnace, and the operating amount MV can be the current in the heating wire.

[0046] The control device 100 in this example calculates the operating quantity MV using control data generated through machine learning. Therefore, compared to other controls such as PID control, it can reduce overshoot and achieve higher-speed control. The control device 100 in this example can be used as a single-cycle controller, outputting either the maximum or minimum operating quantity MV to the controlled object, thereby achieving theoretically the fastest control. Thus, compared to PID control, the control device 100 can shorten the adjustment time of the container's water level or the time for the furnace temperature to rise. Furthermore, it can address the increased production volume associated with the shortened rise time, energy conservation during initial startup, or faster multi-variety, multi-batch manufacturing systems. In addition, by reducing overshoot, it can achieve reduced waste due to early quality stabilization, increased equipment operating rate, or extended equipment lifespan due to reduced equipment load.

[0047] Figure 1B This is an example of the process by which the control device 100 controls the action of the controlled object 310. In step S100, the control device 100 acquires an indication value IV and a measurement value PV. Alternatively, in step S100, the control device 100 may acquire a target value SV, or it may acquire the target value SV in advance. For example, the control device 100 acquires the target value SV pre-set by the operator, and acquires the indication value IV and the measurement value PV in real time from the sensors of the device 300.

[0048] In step S102, the control device 100 acquires control data. The control device 100 may acquire control data generated through machine learning in advance, or it may acquire control data obtained by machine learning using a simulator or an actual system.

[0049] In step S104, the control device 100 calculates the operation quantity MV based on the control data. In step S106, the operation quantity MV is output to the controlled object 310. Then, the control device 100 determines whether to terminate the control. If the control has not terminated, the control device 100 returns to step S100 and obtains the indication value IV and the measurement value PV.

[0050] For example, the control device 100 can combine control using control data generated through machine learning with feedback control. The feedback control can be at least one of proportional control (P control), integral control (I control), or derivative control (D control), and in one example, it can also be PID control. As an example, in feedback control, the operating quantity MV is calculated based on the measured value PV and the target value SV. In feedback control, the operating quantity MV that reduces the difference between the target value SV and the current measured value PV can be calculated in accordance with the setting of the target value SV.

[0051] Figure 2A This section outlines the structure of the learning processing device 200. The learning processing device 200 includes a status data acquisition unit 210, a learning processing unit 220, a generation unit 230, and a supply unit 240.

[0052] The status data acquisition unit 210 acquires status data indicating the status of the device 300 on which the controlled object 310 is installed. For example, the status data acquisition unit 210 acquires the indication value IV and the measurement value PV measured by the sensors installed on the device 300. The status data acquisition unit 210 can supply the acquired status data to the learning processing unit 220 and the generation unit 230.

[0053] The learning processing unit 220 generates a control model 235 that outputs an operational quantity MV corresponding to the indicated value IV and the measured value PV through machine learning. In this example, the input values ​​of the control model 235 are the indicated value IV and the measured value PV, and the output value is the operational quantity MV. Having obtained the measured value PV corresponding to the indicated value IV, the learning processing unit 220 generates a control model 235 that outputs an operational quantity MV with minimal overshoot and fastest convergence to the target value SV through machine learning. The control model 235 can be generated corresponding to multiple predefined systems. For example, machine learning can be performed to generate the control model 235 for each system, such as a system for controlling the temperature of a furnace or a system for controlling the water level in a container.

[0054] Control model 235 is a learning model generated by learning processing unit 220 through reinforcement learning, which outputs operation quantity MV corresponding to the state of device 300. In this example, control model 235 outputs operation quantity MV in a manner that causes the measured value PV to reach a predetermined target value SV. Furthermore, in this figure, as an example, the case where control model 235 is built into learning processing device 200 is shown, but it is not limited to this. Control model 235 may be stored in control device 100. In addition, control model 235 may be a model that outputs operation quantity MV for each target value SV. Control model 235 may contain multiple models after reinforcement learning for each target value SV. For example, control model 235 may output operation quantity MV when the target value SV is the first value (e.g., 5), or when the target value SV is the second value (e.g., 10).

[0055] The generation unit 230 uses the control model 235 to generate control data representing the correspondence between combinations of indicated values ​​IV and measured values ​​PV and the corresponding operating quantities MV. The generation unit 230 can generate different control data for each target value SV. In this example, the generation unit 230 uses the control model 235, which is different for each target value SV, to generate different control data for each target value SV. For example, the generation unit 230 generates an operating quantity correspondence diagram, which will be described later, for each target value SV. Furthermore, when the control model 235 is generated corresponding to multiple systems, the generation unit 230 can generate different control data for each of the multiple systems.

[0056] The supply unit 240 supplies control data to the control device 100. Additionally, the supply unit 240 can supply and store the control data in a pre-defined storage unit. In this example, the supply unit 240 can supply optimal control data for each target value SV or each system. Furthermore, the supply unit 240 can supply optimal control data based on a combination of target value SV and system. That is, the control data is different for each system, and different for each target value within each system.

[0057] Figure 2B This illustrates an example of a machine learning process performed by the learning processing unit 200. In step S200, the learning processing unit 200 acquires state data. In step S202, the learning processing unit 200 generates a control model 235 through machine learning. In step S204, the learning processing unit 200 generates control data using the control model 235. In step S206, the learning processing unit 200 supplies the control data.

[0058] Figure 3Here is an example of a more specific structure of the computation unit 30. In this example, the computation unit 30 has a scheduling unit 32, a correspondence selection unit 34, a decision unit 36, and an inverse scheduling unit 38.

[0059] As control data, the control data acquisition unit 20 acquires multiple operation quantity correspondence diagrams that match the correspondence between combinations of indicated values ​​IV and measured values ​​PV and the corresponding operation quantities MV. These multiple operation quantity correspondence diagrams may include different operation quantity correspondence diagrams for each target value SV. Furthermore, the multiple operation quantity correspondence diagrams may include different operation quantity correspondence diagrams for the system designated as the controlled object 310.

[0060] The scheduling unit 32 schedules the indicated value IV and the measured value PV to values ​​corresponding to the operational quantity correspondence diagram. The scheduling unit 32 schedules the ranges of the indicated value IV and the measured value PV to match the ranges of the indicated value IV' and the measured value PV' to the ranges of the operational quantity correspondence diagram, performing linear processing to ensure range consistency. For example, the range of the measured value PV' in the operational quantity correspondence diagram is [0, 100], while the actual range of the measured value PV in the system is [0, 1]. When the input measured value PV is 0.3, the scheduling unit 32 performs linear processing to set the measured value PV' to 0.3 × 100 = 30 in a manner consistent with the range of the operational quantity correspondence diagram.

[0061] The correspondence diagram selection unit 34 selects a pre-defined operation quantity correspondence diagram from a plurality of operation quantity correspondence diagrams stored in the control data acquisition unit 20. The correspondence diagram selection unit 34 can select an appropriate operation quantity correspondence diagram based on the indication value IV and the measured value PV input to the calculation unit 30. For example, the correspondence diagram selection unit 34 selects the system operation quantity correspondence diagram that is closest to the system of the controlled object 310 from a plurality of operation quantity correspondence diagrams.

[0062] For example, the correspondence diagram selection unit 34 selects an operation quantity correspondence diagram suitable for the actual system from multiple operation quantity correspondence diagrams based on system-related feature data. Here, the multiple operation quantity correspondence diagrams may include operation quantity correspondence diagrams specific to each application, such as those for a heating furnace or for three-stage water tank level control. Furthermore, the multiple operation quantity correspondence diagrams may include operation quantity correspondence diagrams for single-hysteresis systems or double-hysteresis systems, depending on the mathematical characteristics of the system. That is, as feature data, the correspondence diagram selection unit 34 can select any operation quantity correspondence diagram from multiple operation quantity correspondence diagrams based on information related to the system's application or information related to the system's mathematical characteristics. The correspondence diagram selection unit 34 can select any operation quantity correspondence diagram from multiple operation quantity correspondence diagrams by referring to application-related information and information related to the system's mathematical characteristics in a combined manner. More specifically, the correspondence diagram selection unit 34 can compare the feature data and additional information from multiple operation quantity correspondence diagrams to select the operation quantity correspondence diagram that best approximates the information of the actual system. The additional information from multiple operation quantity correspondence diagrams refers to any information related to the operation quantity correspondence diagrams. Furthermore, feature data can be acquired using the input data acquisition unit 10. The feature data can be input by the user or calculated based on the indication value IV and the measured value PV input to the input data acquisition unit 10.

[0063] The decision unit 36 ​​determines the operation quantity MV' corresponding to the input indication value IV' and measurement value PV' using the operation quantity correspondence chart selected by the correspondence chart selection unit 34. Alternatively, the indication value IV and measurement value PV can be directly input to the decision unit 36 ​​without prior scheduling.

[0064] The inverse scheduling unit 38 performs inverse scheduling on the operand MV' calculated from the operand correspondence diagram according to the system. The inverse scheduling unit 38 performs the inverse operation of the scheduling unit 32 in accordance with the actual output range of the system, and outputs the result. For example, the range of the operand MV' in the operand correspondence diagram is [0, 100], and the range of the operand MV in the actual system is [0, 1]. When the value of the operand MV' obtained using the operand correspondence diagram is 5, the inverse scheduling unit 38 performs linear processing to set the operand MV to 5 ÷ 100 = 0.05 in accordance with the actual output range of the system.

[0065] Figure 4A This is an example of an operation quantity correspondence graph. The vertical axis represents the measured value PV, and the horizontal axis represents the indicated value IV. Furthermore, in this example, the operation quantity correspondence graph is divided into region A and region B based on the combination of the indicated value IV and the measured value PV. The control device 100 can output different operation quantities MV in region A and region B.

[0066] For example, when the combination of indicated value IV and measured value PV is located in region A, the control device 100 controls the controlled object 310 with the maximum operating amount MV within a predefined operating range. Conversely, when the combination of indicated value IV and measured value PV is located in region B, the control device 100 can control the controlled object 310 with the minimum operating amount MV within a predefined operating range.

[0067] In this example, the system's stable trajectory is represented by full acceleration and full braking control. For instance, full braking control based on the operational variable correspondence diagram is applied from a predefined initial state a to eventually stabilize the system at a steady state c. Similarly, full acceleration control is applied from a predefined initial state b to stabilize the system at the same steady state c.

[0068] Figure 4B This example illustrates a control method for the control device 100 involved in the embodiment. In this example, it shows the control result when the water level of a single-stage lag system, referred to as a "three-stage tank," is controlled on a simulator. In this example, the control method is compared to... (The sentence is incomplete and requires further context to translate accurately.) Figure 5 The PID control shown converges to the predefined target value SV more quickly. Thus, the control device 100 in this example uses control data to set an appropriate operating quantity MV, thereby avoiding overshoot and achieving high-speed control.

[0069] Figure 4C This example illustrates a method for generating operand-to-operational graphs using machine learning. Operand-to-operational graphs can be generated using predefined reinforcement learning algorithms. This example uses Kernel Dynamic Policy Programming (KDPP) to generate the operand-to-operational graph, but is not limited to this method.

[0070] Evaluation functions, such as f(t) = |Measured value PV(t) - Target value SV|, are used. Several points between -MAX and +MAX are selected, and the operation value MV is set as the output value of the reinforcement learning. In KDPP, if learning is fully performed in a system with one input and one output, the final reinforcement learning model converges to a model using only two values: +MAX and -MAX. Furthermore, the generated reinforcement learning model is assigned a combination of indicator value IV and measured value PV, and the operation value MV (+MAX or -MAX) is calculated to form a curve, thus generating an operation-to-output plot.

[0071] Figure 5 This illustrates an example of a control method involved in proportional control. In this example, PID control is used as a single-input, single-output control algorithm for one operand MV and one measured value PV.

[0072] Here, in PID control, stable control can be achieved, but the time until the target value SV is reached is not optimal. PID control calculates the solution by performing a Laplace transform on the system's differential equations, transforming them into algebraic equations. In general, smooth functions such as exponential functions or combinations of trigonometric functions form the solution space. However, even when performing a Laplace transform on a single triangular wave, the exponential function terms remain, and it does not become an algebraic equation, thus excluding solutions such as single triangular waves.

[0073] In contrast, the current Laplace transform does not consider the summation of solutions from a single triangular wave for control utilizing full acceleration and full braking. That is, the Laplace transform method cannot calculate the fastest control solution using full acceleration and full braking. Therefore, in proportional PID control, overshoot makes high-speed control difficult to achieve.

[0074] Furthermore, while advanced control methods can be considered to achieve control similar to full acceleration and full braking, this requires complex parameter adjustments and is therefore difficult to integrate into small controllers capable of executing actions using microcomputers with limited processing power. The control device 100 in this example can be replaced with a structure that performs PID calculations for a single-cycle controller; the structure can be the same as existing structures, such as analog signal processing or digital signal processing. The control device 100 can also be integrated into a small controller.

[0075] Figure 6A This illustrates one embodiment of the learning processing device 200. The learning processing device 200 includes a simulator 250. The simulator 250 may be located externally to the learning processing device 200.

[0076] The simulator 250 supplies the pre-defined indication value IV and the measured value PV to the status data acquisition unit 210. For example, the simulator 250 can be created using any system identification technique and actual measurement data of the system. In this example, the learning processing device 200 generates a control model 235 by utilizing machine learning from the simulator 250. Thus, even when the controlled object 310 is a more complex system, the learning processing device 200 in this example can perform learning processing using the simulator 250.

[0077] Figure 6BThis illustrates an example of an implementation of the learning processing apparatus 200. In this example, the control data acquisition unit 20 sets the indication value IV and the measurement value PV acquired from the device 300 as state data and generates a control model 235 through machine learning. Even when it is difficult to generate a simulator 250 corresponding to the controlled object 310, the learning processing apparatus 200 in this example can still generate the control model 235. The control data generated using the actual system can be combined with control data generated using other methods such as the simulator 250. That is, multiple operation quantity correspondence diagrams can include operation quantity correspondence diagrams obtained through machine learning using different methods.

[0078] Figure 7 Examples of computer 2200 that can embody all or part of the various embodiments of the present invention are shown. Programs installed in computer 2200 enable computer 2200 to function as an operation associated with or one or more parts of a device according to embodiments of the present invention, or to perform such operation or such parts, and / or to perform such processes or stages of processes according to embodiments of the present invention. Such programs may be executed by CPU 2212 to enable computer 2200 to perform specific operations associated with several or all of the modules in the flowcharts and block diagrams described in this specification.

[0079] The computer 2200 according to this embodiment includes a CPU 2212, RAM 2214, a graphics controller 2216, and a display device 2218, which are interconnected via a main controller 2210. Additionally, the computer 2200 includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card driver, which are connected to the main controller 2210 via an input / output controller 2220. Furthermore, the computer includes conventional input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0080] CPU 2212 executes actions according to the programs stored in ROM 2230 and RAM 2214, thereby controlling each unit. Graphics controller 2216 acquires image data provided to RAM 2214, generated by CPU 2212 in frame buffer or other locations, and displays the image data on display device 2218.

[0081] Communication interface 2222 communicates with other electronic devices via a network. Hard disk drive 2224 stores programs and data used by CPU 2212 within computer 2200. DVD-ROM drive 2226 reads programs or data from DVD-ROM 2201 and provides the programs or data to hard disk drive 2224 via RAM 2214. IC card drive reads programs and data from IC card and / or writes programs and data to IC card.

[0082] ROM 2230 stores startup programs and / or programs that depend on the hardware of computer 2200, which are executed by computer 2200 upon activation. Additionally, input / output chip 2240 connects various input / output units to input / output controller 2220 via parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0083] The program is provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from the computer-readable medium and installed on a hard disk drive 2224, RAM 2214, or ROM 2230, which are also examples of computer-readable media, and executed by the CPU 2212. The information processing described in the program is read by the computer 2200, enabling cooperation between the program and the aforementioned hardware resources of various types. An apparatus or method can be constituted by using the computer 2200 to perform information manipulation or processing.

[0084] For example, when communication is performed between computer 2200 and external devices, CPU 2212 can execute a communication program loaded in RAM 2214 and issue communication processing instructions to communication interface 2222 based on the processing described in the communication program. Under the control of CPU 2212, communication interface 2222 reads transmission data stored in the transmission buffer processing area provided in recording media such as RAM 2214, hard disk drive 2224, DVD-ROM 2201, or IC card, and sends the read transmission data to the network, or writes received data received from the network to the receive buffer processing area provided on the recording medium, etc.

[0085] In addition, CPU 2212 can read all or a portion of files or databases stored on external recording media such as hard disk drive 2224, DVD-ROM drive 2226 (DVD-ROM 2201), IC card, etc., into RAM 2214, and perform various types of processing on the data in RAM 2214. CPU 2212 then writes the processed data back to the external recording media.

[0086] Various types of information, such as programs, data, tables, and databases, can be stored in recording media and processed. The CPU 2212 can perform various types of processing, including operations, information processing, conditional judgments, conditional branches, unconditional branches, and information retrieval / replacement, as specified by a sequence of instructions described at any location in this disclosure, on data read from RAM 2214, and write back the results to RAM 2214. Furthermore, the CPU 2212 can retrieve information from files, databases, etc., within the recording medium. For example, when multiple entries, each having an attribute value of a first attribute associated with a second attribute value, are stored in the recording medium, the CPU 2212 can retrieve from these multiple records an entry whose first attribute value matches a condition, read the attribute value of the second attribute stored in that entry, and obtain the attribute value of the second attribute associated with the first attribute that satisfies the pre-defined condition.

[0087] The programs or software modules described above can be stored on or near computer-readable media on computer 2200. Alternatively, recording media such as hard disks or RAM provided to a server system connected to a dedicated communication network or the Internet can be used as computer-readable media, thereby providing the program to computer 2200 via the network.

[0088] The present invention has been described above using embodiments, but the technical scope of the present invention is not limited to the scope described in the above embodiments. Those skilled in the art will understand that various modifications or improvements can be made to the above embodiments. As clearly understood from the claims, such modifications or improvements can also be included within the technical scope of the present invention.

[0089] Regarding the execution order of actions, sequence, steps, and stages in the apparatus, system, program, and method shown in the claims, specification, and drawings, it should be noted that, unless explicitly stated as "earlier" or "earlier," and the output of a previous process is not used in a subsequent process, they can be implemented in any order. Even if terms such as "firstly," "nextly," etc., are used for convenience in describing the flow of actions in the claims, specification, and drawings, it does not mean that they must be performed in that order.

[0090] Explanation of the label

[0091] 10…Input data acquisition unit, 20…Control data acquisition unit, 30…Calculation unit, 32…Scheduling unit, 34…Correspondence diagram selection unit, 36…Decision unit, 38…Inverse scheduling unit, 40…Output unit, 100…Control device, 200…Learning processing device, 210…Status data acquisition unit, 220…Learning processing unit, 230…Generation unit, 235…Control model, 240…Supply unit, 250…Simulator, 300…Equipment, 310…Control object, 2200…Computer, 2201…DVD-ROM, 2210…Main controller, 2212…CPU, 2214…RAM, 2216…Graphics controller, 2218…Display device, 2220…Input / output controller, 2222…Communication interface, 2224…Hard disk drive, 2226…DVD-ROM drive, 2230…ROM, 2240…Input / output chip, 2242…Keyboard.

Claims

1. A learning processing device, wherein, The learning processing device has: The learning processing unit generates, through machine learning, a control model that outputs operational quantities to the control object set in the device in accordance with the indication values ​​of the state of the controlled object set in the device, which are represented by sensors set in the device according to a predefined system, and the measured values ​​of the operating state, which are measured by sensors set in the device and represent the result of controlling the controlled object. The generation unit, using the control model, generates control data representing the correspondence between combinations of indicated values ​​and measured values, and the corresponding operational quantities; and The supply department supplies the control data to the pre-defined control device.

2. The learning processing apparatus according to claim 1, wherein, The operating quantity is either the maximum operating quantity or the minimum operating quantity within a predefined operable range.

3. The learning processing apparatus according to claim 1 or 2, wherein, The control model outputs the operational quantity in a manner that causes the measured value to reach the set target value. The generation unit generates control data that is different for each of the target values.

4. The learning processing apparatus according to claim 1 or 2, wherein, The control model is generated corresponding to multiple predefined systems. The generation unit generates different control data for each of the plurality of systems.

5. The learning processing apparatus according to claim 1 or 2, wherein, The control data includes an operation quantity correspondence diagram that matches the combination of the indicated value and the measured value with the corresponding operation quantity.

6. A control device, wherein, The control device has: The input data acquisition unit acquires a pre-defined system of indicator values ​​representing the state of a controlled object installed on the device, as indicated by sensors installed on the device, and a measurement value representing the operating state of the controlled object, as measured by sensors installed on the device. The control data acquisition unit acquires control data generated using a control model that has been learned to output an operational quantity of a control object set on the device in accordance with the indicated value and the measured value. The control data represents the correspondence between the combination of the indicated value and the measured value and the operational quantity output in accordance with the combination. The calculation unit, using the control data, calculates the operating quantity corresponding to the combination of the indicated value and the measured value; and The output unit outputs the operation quantity to the controlled object.

7. The control device according to claim 6, wherein, As control data, the control data acquisition unit acquires multiple operation quantity correspondence maps that match the correspondence between combinations of indicated values ​​and measured values, and operation quantities corresponding to those combinations. The input data acquisition unit acquires information related to the application of the system or information related to the mathematical characteristics of the system. The control device includes a mapping selection unit that selects an arbitrary operand mapping map from the plurality of operand mapping maps based on information related to the application of the system or information related to the mathematical characteristics of the system. The system is used for applications including heating furnaces or three-stage water tank level control. The mathematical characteristics of the system include a single-lag system or a double-lag system.

8. The control device according to claim 7, wherein, The control device has: The scheduling unit schedules the indicated value and the measured value to values ​​corresponding to the operation quantity correspondence diagram; and The inverse scheduling unit performs inverse scheduling on the operands calculated based on the operand correspondence diagram according to the system.

9. The control device according to any one of claims 6 to 8, wherein, As the operation quantity, the output unit outputs the maximum operation quantity or the minimum operation quantity within the predefined operable range to the controlled object.

10. A learning processing method, wherein, The learning processing method has the following stages: Through machine learning, a control model is generated that outputs the operational quantities of the controlled object to the device in accordance with the indication values ​​of the sensors set on the device, which represent the state of the controlled object set on the device, and the measured values ​​of the operating state, which represent the result of controlling the controlled object, as measured by the sensors set on the device. Using the control model, control data is generated representing the correspondence between combinations of indicated values ​​and measured values, and the corresponding operational quantities; and The control data is supplied to a pre-defined control device.

11. A control method, wherein, The control method has the following stages: Acquire a pre-defined system of indicator values ​​representing the state of a controlled object installed on the device, as indicated by sensors installed on the device, and a measurement value representing the operating state of the controlled object, as measured by sensors installed on the device. Acquire control data, which is generated using a control model that has been learned to output the operational quantity of the control object set on the device in accordance with the indicated value and the measured value, and represents the correspondence between the combination of the indicated value and the measured value and the operational quantity output in accordance with the combination; Using the control data, the operating quantity corresponding to the combination of the indicated value and the measured value is calculated; and The operation quantity is output to the controlled object.

12. A recording medium that records a learning program, wherein, The learning program is executed by a computer, which then functions as a functional unit: The learning processing unit generates, through machine learning, a control model that outputs operational quantities to the control object set in the device in accordance with the indication values ​​of the state of the controlled object set in the device, which are represented by sensors set in the device according to a predefined system, and the measured values ​​of the operating state, which are measured by sensors set in the device and represent the result of controlling the controlled object. The generation unit, using the control model, generates control data representing the correspondence between combinations of indicated values ​​and measured values, and the corresponding operational quantities; and The supply department supplies the control data to the pre-defined control device.

13. A recording medium having a control program recorded thereon, wherein, The control program is executed by a computer, causing the computer to function as a functional unit: The input data acquisition unit acquires a pre-defined system of indicator values ​​representing the state of a controlled object installed on the device, as indicated by sensors installed on the device, and a measurement value representing the operating state of the controlled object, as measured by sensors installed on the device. The control data acquisition unit acquires control data generated using a control model that has been learned to output an operational quantity of a control object set on the device in accordance with the indicated value and the measured value. The control data represents the correspondence between the combination of the indicated value and the measured value and the operational quantity output in accordance with the combination. The calculation unit, using the control data, calculates the operating quantity corresponding to the combination of the indicated value and the measured value; and The output unit outputs the operation quantity to the controlled object.

Images